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 #ifndef arch_mmap_check
37 #define arch_mmap_check(addr, len, flags) (0)
40 static void unmap_region(struct mm_struct *mm,
41 struct vm_area_struct *vma, struct vm_area_struct *prev,
42 unsigned long start, unsigned long end);
45 * WARNING: the debugging will use recursive algorithms so never enable this
46 * unless you know what you are doing.
50 /* description of effects of mapping type and prot in current implementation.
51 * this is due to the limited x86 page protection hardware. The expected
52 * behavior is in parens:
55 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
56 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
57 * w: (no) no w: (no) no w: (yes) yes w: (no) no
58 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
60 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
61 * w: (no) no w: (no) no w: (copy) copy w: (no) no
62 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
65 pgprot_t protection_map[16] = {
66 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
67 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
70 pgprot_t vm_get_page_prot(unsigned long vm_flags)
72 return protection_map[vm_flags &
73 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
75 EXPORT_SYMBOL(vm_get_page_prot);
77 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
78 int sysctl_overcommit_ratio = 50; /* default is 50% */
79 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
80 atomic_t vm_committed_space = ATOMIC_INIT(0);
83 * Check that a process has enough memory to allocate a new virtual
84 * mapping. 0 means there is enough memory for the allocation to
85 * succeed and -ENOMEM implies there is not.
87 * We currently support three overcommit policies, which are set via the
88 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
90 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
91 * Additional code 2002 Jul 20 by Robert Love.
93 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
95 * Note this is a helper function intended to be used by LSMs which
96 * wish to use this logic.
98 int __vm_enough_memory(long pages, int cap_sys_admin)
100 unsigned long free, allowed;
102 vm_acct_memory(pages);
105 * Sometimes we want to use more memory than we have
107 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
110 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
113 free = global_page_state(NR_FILE_PAGES);
114 free += nr_swap_pages;
117 * Any slabs which are created with the
118 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
119 * which are reclaimable, under pressure. The dentry
120 * cache and most inode caches should fall into this
122 free += atomic_read(&slab_reclaim_pages);
125 * Leave the last 3% for root
134 * nr_free_pages() is very expensive on large systems,
135 * only call if we're about to fail.
140 * Leave reserved pages. The pages are not for anonymous pages.
142 if (n <= totalreserve_pages)
145 n -= totalreserve_pages;
148 * Leave the last 3% for root
160 allowed = (totalram_pages - hugetlb_total_pages())
161 * sysctl_overcommit_ratio / 100;
163 * Leave the last 3% for root
166 allowed -= allowed / 32;
167 allowed += total_swap_pages;
169 /* Don't let a single process grow too big:
170 leave 3% of the size of this process for other processes */
171 allowed -= current->mm->total_vm / 32;
174 * cast `allowed' as a signed long because vm_committed_space
175 * sometimes has a negative value
177 if (atomic_read(&vm_committed_space) < (long)allowed)
180 vm_unacct_memory(pages);
185 EXPORT_SYMBOL(__vm_enough_memory);
188 * Requires inode->i_mapping->i_mmap_lock
190 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
191 struct file *file, struct address_space *mapping)
193 if (vma->vm_flags & VM_DENYWRITE)
194 atomic_inc(&file->f_dentry->d_inode->i_writecount);
195 if (vma->vm_flags & VM_SHARED)
196 mapping->i_mmap_writable--;
198 flush_dcache_mmap_lock(mapping);
199 if (unlikely(vma->vm_flags & VM_NONLINEAR))
200 list_del_init(&vma->shared.vm_set.list);
202 vma_prio_tree_remove(vma, &mapping->i_mmap);
203 flush_dcache_mmap_unlock(mapping);
207 * Unlink a file-based vm structure from its prio_tree, to hide
208 * vma from rmap and vmtruncate before freeing its page tables.
210 void unlink_file_vma(struct vm_area_struct *vma)
212 struct file *file = vma->vm_file;
215 struct address_space *mapping = file->f_mapping;
216 spin_lock(&mapping->i_mmap_lock);
217 __remove_shared_vm_struct(vma, file, mapping);
218 spin_unlock(&mapping->i_mmap_lock);
223 * Close a vm structure and free it, returning the next.
225 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
227 struct vm_area_struct *next = vma->vm_next;
230 if (vma->vm_ops && vma->vm_ops->close)
231 vma->vm_ops->close(vma);
234 mpol_free(vma_policy(vma));
235 kmem_cache_free(vm_area_cachep, vma);
239 asmlinkage unsigned long sys_brk(unsigned long brk)
241 unsigned long rlim, retval;
242 unsigned long newbrk, oldbrk;
243 struct mm_struct *mm = current->mm;
245 down_write(&mm->mmap_sem);
247 if (brk < mm->end_code)
251 * Check against rlimit here. If this check is done later after the test
252 * of oldbrk with newbrk then it can escape the test and let the data
253 * segment grow beyond its set limit the in case where the limit is
254 * not page aligned -Ram Gupta
256 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
257 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
260 newbrk = PAGE_ALIGN(brk);
261 oldbrk = PAGE_ALIGN(mm->brk);
262 if (oldbrk == newbrk)
265 /* Always allow shrinking brk. */
266 if (brk <= mm->brk) {
267 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
272 /* Check against existing mmap mappings. */
273 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
276 /* Ok, looks good - let it rip. */
277 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
283 up_write(&mm->mmap_sem);
288 static int browse_rb(struct rb_root *root)
291 struct rb_node *nd, *pn = NULL;
292 unsigned long prev = 0, pend = 0;
294 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
295 struct vm_area_struct *vma;
296 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
297 if (vma->vm_start < prev)
298 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
299 if (vma->vm_start < pend)
300 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
301 if (vma->vm_start > vma->vm_end)
302 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
307 for (nd = pn; nd; nd = rb_prev(nd)) {
311 printk("backwards %d, forwards %d\n", j, i), i = 0;
315 void validate_mm(struct mm_struct *mm)
319 struct vm_area_struct *tmp = mm->mmap;
324 if (i != mm->map_count)
325 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
326 i = browse_rb(&mm->mm_rb);
327 if (i != mm->map_count)
328 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
332 #define validate_mm(mm) do { } while (0)
335 static struct vm_area_struct *
336 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
337 struct vm_area_struct **pprev, struct rb_node ***rb_link,
338 struct rb_node ** rb_parent)
340 struct vm_area_struct * vma;
341 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
343 __rb_link = &mm->mm_rb.rb_node;
344 rb_prev = __rb_parent = NULL;
348 struct vm_area_struct *vma_tmp;
350 __rb_parent = *__rb_link;
351 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
353 if (vma_tmp->vm_end > addr) {
355 if (vma_tmp->vm_start <= addr)
357 __rb_link = &__rb_parent->rb_left;
359 rb_prev = __rb_parent;
360 __rb_link = &__rb_parent->rb_right;
366 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
367 *rb_link = __rb_link;
368 *rb_parent = __rb_parent;
373 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
374 struct vm_area_struct *prev, struct rb_node *rb_parent)
376 if (vma->vm_flags & VM_EXEC)
377 arch_add_exec_range(mm, vma->vm_end);
379 vma->vm_next = prev->vm_next;
384 vma->vm_next = rb_entry(rb_parent,
385 struct vm_area_struct, vm_rb);
391 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
392 struct rb_node **rb_link, struct rb_node *rb_parent)
394 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
395 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
398 static inline void __vma_link_file(struct vm_area_struct *vma)
404 struct address_space *mapping = file->f_mapping;
406 if (vma->vm_flags & VM_DENYWRITE)
407 atomic_dec(&file->f_dentry->d_inode->i_writecount);
408 if (vma->vm_flags & VM_SHARED)
409 mapping->i_mmap_writable++;
411 flush_dcache_mmap_lock(mapping);
412 if (unlikely(vma->vm_flags & VM_NONLINEAR))
413 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
415 vma_prio_tree_insert(vma, &mapping->i_mmap);
416 flush_dcache_mmap_unlock(mapping);
421 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
422 struct vm_area_struct *prev, struct rb_node **rb_link,
423 struct rb_node *rb_parent)
425 __vma_link_list(mm, vma, prev, rb_parent);
426 __vma_link_rb(mm, vma, rb_link, rb_parent);
427 __anon_vma_link(vma);
430 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
431 struct vm_area_struct *prev, struct rb_node **rb_link,
432 struct rb_node *rb_parent)
434 struct address_space *mapping = NULL;
437 mapping = vma->vm_file->f_mapping;
440 spin_lock(&mapping->i_mmap_lock);
441 vma->vm_truncate_count = mapping->truncate_count;
445 __vma_link(mm, vma, prev, rb_link, rb_parent);
446 __vma_link_file(vma);
448 anon_vma_unlock(vma);
450 spin_unlock(&mapping->i_mmap_lock);
457 * Helper for vma_adjust in the split_vma insert case:
458 * insert vm structure into list and rbtree and anon_vma,
459 * but it has already been inserted into prio_tree earlier.
462 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
464 struct vm_area_struct * __vma, * prev;
465 struct rb_node ** rb_link, * rb_parent;
467 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
468 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
469 __vma_link(mm, vma, prev, rb_link, rb_parent);
474 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
475 struct vm_area_struct *prev)
477 prev->vm_next = vma->vm_next;
478 rb_erase(&vma->vm_rb, &mm->mm_rb);
479 if (mm->mmap_cache == vma)
480 mm->mmap_cache = prev;
481 if (vma->vm_flags & VM_EXEC)
482 arch_remove_exec_range(mm, vma->vm_end);
486 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
487 * is already present in an i_mmap tree without adjusting the tree.
488 * The following helper function should be used when such adjustments
489 * are necessary. The "insert" vma (if any) is to be inserted
490 * before we drop the necessary locks.
492 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
493 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
495 struct mm_struct *mm = vma->vm_mm;
496 struct vm_area_struct *next = vma->vm_next;
497 struct vm_area_struct *importer = NULL;
498 struct address_space *mapping = NULL;
499 struct prio_tree_root *root = NULL;
500 struct file *file = vma->vm_file;
501 struct anon_vma *anon_vma = NULL;
502 long adjust_next = 0;
505 if (next && !insert) {
506 if (end >= next->vm_end) {
508 * vma expands, overlapping all the next, and
509 * perhaps the one after too (mprotect case 6).
511 again: remove_next = 1 + (end > next->vm_end);
513 anon_vma = next->anon_vma;
515 } else if (end > next->vm_start) {
517 * vma expands, overlapping part of the next:
518 * mprotect case 5 shifting the boundary up.
520 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
521 anon_vma = next->anon_vma;
523 } else if (end < vma->vm_end) {
525 * vma shrinks, and !insert tells it's not
526 * split_vma inserting another: so it must be
527 * mprotect case 4 shifting the boundary down.
529 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
530 anon_vma = next->anon_vma;
536 mapping = file->f_mapping;
537 if (!(vma->vm_flags & VM_NONLINEAR))
538 root = &mapping->i_mmap;
539 spin_lock(&mapping->i_mmap_lock);
541 vma->vm_truncate_count != next->vm_truncate_count) {
543 * unmap_mapping_range might be in progress:
544 * ensure that the expanding vma is rescanned.
546 importer->vm_truncate_count = 0;
549 insert->vm_truncate_count = vma->vm_truncate_count;
551 * Put into prio_tree now, so instantiated pages
552 * are visible to arm/parisc __flush_dcache_page
553 * throughout; but we cannot insert into address
554 * space until vma start or end is updated.
556 __vma_link_file(insert);
561 * When changing only vma->vm_end, we don't really need
562 * anon_vma lock: but is that case worth optimizing out?
565 anon_vma = vma->anon_vma;
567 spin_lock(&anon_vma->lock);
569 * Easily overlooked: when mprotect shifts the boundary,
570 * make sure the expanding vma has anon_vma set if the
571 * shrinking vma had, to cover any anon pages imported.
573 if (importer && !importer->anon_vma) {
574 importer->anon_vma = anon_vma;
575 __anon_vma_link(importer);
580 flush_dcache_mmap_lock(mapping);
581 vma_prio_tree_remove(vma, root);
583 vma_prio_tree_remove(next, root);
586 vma->vm_start = start;
588 vma->vm_pgoff = pgoff;
590 next->vm_start += adjust_next << PAGE_SHIFT;
591 next->vm_pgoff += adjust_next;
596 vma_prio_tree_insert(next, root);
597 vma_prio_tree_insert(vma, root);
598 flush_dcache_mmap_unlock(mapping);
603 * vma_merge has merged next into vma, and needs
604 * us to remove next before dropping the locks.
606 __vma_unlink(mm, next, vma);
608 __remove_shared_vm_struct(next, file, mapping);
610 __anon_vma_merge(vma, next);
613 * split_vma has split insert from vma, and needs
614 * us to insert it before dropping the locks
615 * (it may either follow vma or precede it).
617 __insert_vm_struct(mm, insert);
621 spin_unlock(&anon_vma->lock);
623 spin_unlock(&mapping->i_mmap_lock);
629 mpol_free(vma_policy(next));
630 kmem_cache_free(vm_area_cachep, next);
632 * In mprotect's case 6 (see comments on vma_merge),
633 * we must remove another next too. It would clutter
634 * up the code too much to do both in one go.
636 if (remove_next == 2) {
646 * If the vma has a ->close operation then the driver probably needs to release
647 * per-vma resources, so we don't attempt to merge those.
649 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
651 static inline int is_mergeable_vma(struct vm_area_struct *vma,
652 struct file *file, unsigned long vm_flags)
654 if (vma->vm_flags != vm_flags)
656 if (vma->vm_file != file)
658 if (vma->vm_ops && vma->vm_ops->close)
663 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
664 struct anon_vma *anon_vma2)
666 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
670 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
671 * in front of (at a lower virtual address and file offset than) the vma.
673 * We cannot merge two vmas if they have differently assigned (non-NULL)
674 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
676 * We don't check here for the merged mmap wrapping around the end of pagecache
677 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
678 * wrap, nor mmaps which cover the final page at index -1UL.
681 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
682 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
684 if (is_mergeable_vma(vma, file, vm_flags) &&
685 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
686 if (vma->vm_pgoff == vm_pgoff)
693 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
694 * beyond (at a higher virtual address and file offset than) the vma.
696 * We cannot merge two vmas if they have differently assigned (non-NULL)
697 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
700 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
701 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
703 if (is_mergeable_vma(vma, file, vm_flags) &&
704 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
706 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
707 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
714 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
715 * whether that can be merged with its predecessor or its successor.
716 * Or both (it neatly fills a hole).
718 * In most cases - when called for mmap, brk or mremap - [addr,end) is
719 * certain not to be mapped by the time vma_merge is called; but when
720 * called for mprotect, it is certain to be already mapped (either at
721 * an offset within prev, or at the start of next), and the flags of
722 * this area are about to be changed to vm_flags - and the no-change
723 * case has already been eliminated.
725 * The following mprotect cases have to be considered, where AAAA is
726 * the area passed down from mprotect_fixup, never extending beyond one
727 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
729 * AAAA AAAA AAAA AAAA
730 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
731 * cannot merge might become might become might become
732 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
733 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
734 * mremap move: PPPPNNNNNNNN 8
736 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
737 * might become case 1 below case 2 below case 3 below
739 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
740 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
742 struct vm_area_struct *vma_merge(struct mm_struct *mm,
743 struct vm_area_struct *prev, unsigned long addr,
744 unsigned long end, unsigned long vm_flags,
745 struct anon_vma *anon_vma, struct file *file,
746 pgoff_t pgoff, struct mempolicy *policy)
748 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
749 struct vm_area_struct *area, *next;
752 * We later require that vma->vm_flags == vm_flags,
753 * so this tests vma->vm_flags & VM_SPECIAL, too.
755 if (vm_flags & VM_SPECIAL)
759 next = prev->vm_next;
763 if (next && next->vm_end == end) /* cases 6, 7, 8 */
764 next = next->vm_next;
767 * Can it merge with the predecessor?
769 if (prev && prev->vm_end == addr &&
770 mpol_equal(vma_policy(prev), policy) &&
771 can_vma_merge_after(prev, vm_flags,
772 anon_vma, file, pgoff)) {
774 * OK, it can. Can we now merge in the successor as well?
776 if (next && end == next->vm_start &&
777 mpol_equal(policy, vma_policy(next)) &&
778 can_vma_merge_before(next, vm_flags,
779 anon_vma, file, pgoff+pglen) &&
780 is_mergeable_anon_vma(prev->anon_vma,
783 vma_adjust(prev, prev->vm_start,
784 next->vm_end, prev->vm_pgoff, NULL);
785 } else /* cases 2, 5, 7 */
786 vma_adjust(prev, prev->vm_start,
787 end, prev->vm_pgoff, NULL);
788 if (prev->vm_flags & VM_EXEC)
789 arch_add_exec_range(mm, prev->vm_end);
794 * Can this new request be merged in front of next?
796 if (next && end == next->vm_start &&
797 mpol_equal(policy, vma_policy(next)) &&
798 can_vma_merge_before(next, vm_flags,
799 anon_vma, file, pgoff+pglen)) {
800 if (prev && addr < prev->vm_end) /* case 4 */
801 vma_adjust(prev, prev->vm_start,
802 addr, prev->vm_pgoff, NULL);
803 else /* cases 3, 8 */
804 vma_adjust(area, addr, next->vm_end,
805 next->vm_pgoff - pglen, NULL);
813 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
814 * neighbouring vmas for a suitable anon_vma, before it goes off
815 * to allocate a new anon_vma. It checks because a repetitive
816 * sequence of mprotects and faults may otherwise lead to distinct
817 * anon_vmas being allocated, preventing vma merge in subsequent
820 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
822 struct vm_area_struct *near;
823 unsigned long vm_flags;
830 * Since only mprotect tries to remerge vmas, match flags
831 * which might be mprotected into each other later on.
832 * Neither mlock nor madvise tries to remerge at present,
833 * so leave their flags as obstructing a merge.
835 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
836 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
838 if (near->anon_vma && vma->vm_end == near->vm_start &&
839 mpol_equal(vma_policy(vma), vma_policy(near)) &&
840 can_vma_merge_before(near, vm_flags,
841 NULL, vma->vm_file, vma->vm_pgoff +
842 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
843 return near->anon_vma;
846 * It is potentially slow to have to call find_vma_prev here.
847 * But it's only on the first write fault on the vma, not
848 * every time, and we could devise a way to avoid it later
849 * (e.g. stash info in next's anon_vma_node when assigning
850 * an anon_vma, or when trying vma_merge). Another time.
852 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
856 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
857 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
859 if (near->anon_vma && near->vm_end == vma->vm_start &&
860 mpol_equal(vma_policy(near), vma_policy(vma)) &&
861 can_vma_merge_after(near, vm_flags,
862 NULL, vma->vm_file, vma->vm_pgoff))
863 return near->anon_vma;
866 * There's no absolute need to look only at touching neighbours:
867 * we could search further afield for "compatible" anon_vmas.
868 * But it would probably just be a waste of time searching,
869 * or lead to too many vmas hanging off the same anon_vma.
870 * We're trying to allow mprotect remerging later on,
871 * not trying to minimize memory used for anon_vmas.
876 #ifdef CONFIG_PROC_FS
877 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
878 struct file *file, long pages)
880 const unsigned long stack_flags
881 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
884 mm->shared_vm += pages;
885 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
886 mm->exec_vm += pages;
887 } else if (flags & stack_flags)
888 mm->stack_vm += pages;
889 if (flags & (VM_RESERVED|VM_IO))
890 mm->reserved_vm += pages;
892 #endif /* CONFIG_PROC_FS */
895 * The caller must hold down_write(current->mm->mmap_sem).
898 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
899 unsigned long len, unsigned long prot,
900 unsigned long flags, unsigned long pgoff)
902 struct mm_struct * mm = current->mm;
903 struct vm_area_struct * vma, * prev;
905 unsigned int vm_flags;
906 int correct_wcount = 0;
908 struct rb_node ** rb_link, * rb_parent;
910 unsigned long charged = 0, reqprot = prot;
913 if (is_file_hugepages(file))
916 if (!file->f_op || !file->f_op->mmap)
919 if ((prot & PROT_EXEC) &&
920 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
924 * Does the application expect PROT_READ to imply PROT_EXEC?
926 * (the exception is when the underlying filesystem is noexec
927 * mounted, in which case we dont add PROT_EXEC.)
929 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
930 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
936 error = arch_mmap_check(addr, len, flags);
940 /* Careful about overflows.. */
941 len = PAGE_ALIGN(len);
942 if (!len || len > TASK_SIZE)
945 /* offset overflow? */
946 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
949 /* Too many mappings? */
950 if (mm->map_count > sysctl_max_map_count)
953 /* Obtain the address to map to. we verify (or select) it and ensure
954 * that it represents a valid section of the address space.
956 addr = get_unmapped_area_prot(file, addr, len, pgoff, flags, prot & PROT_EXEC);
957 if (addr & ~PAGE_MASK)
960 /* Do simple checking here so the lower-level routines won't have
961 * to. we assume access permissions have been handled by the open
962 * of the memory object, so we don't do any here.
964 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
965 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
967 if (flags & MAP_LOCKED) {
970 vm_flags |= VM_LOCKED;
972 /* mlock MCL_FUTURE? */
973 if (vm_flags & VM_LOCKED) {
974 unsigned long locked, lock_limit;
975 locked = len >> PAGE_SHIFT;
976 locked += mm->locked_vm;
977 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
978 lock_limit >>= PAGE_SHIFT;
979 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
983 inode = file ? file->f_dentry->d_inode : NULL;
986 switch (flags & MAP_TYPE) {
988 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
992 * Make sure we don't allow writing to an append-only
995 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
999 * Make sure there are no mandatory locks on the file.
1001 if (locks_verify_locked(inode))
1004 vm_flags |= VM_SHARED | VM_MAYSHARE;
1005 if (!(file->f_mode & FMODE_WRITE))
1006 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1010 if (!(file->f_mode & FMODE_READ))
1018 switch (flags & MAP_TYPE) {
1020 vm_flags |= VM_SHARED | VM_MAYSHARE;
1024 * Set pgoff according to addr for anon_vma.
1026 pgoff = addr >> PAGE_SHIFT;
1033 error = security_file_mmap(file, reqprot, prot, flags);
1037 /* Clear old maps */
1040 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1041 if (vma && vma->vm_start < addr + len) {
1042 if (do_munmap(mm, addr, len))
1047 /* Check against address space limit. */
1048 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1051 if (accountable && (!(flags & MAP_NORESERVE) ||
1052 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1053 if (vm_flags & VM_SHARED) {
1054 /* Check memory availability in shmem_file_setup? */
1055 vm_flags |= VM_ACCOUNT;
1056 } else if (vm_flags & VM_WRITE) {
1058 * Private writable mapping: check memory availability
1060 charged = len >> PAGE_SHIFT;
1061 if (security_vm_enough_memory(charged))
1063 vm_flags |= VM_ACCOUNT;
1068 * Can we just expand an old private anonymous mapping?
1069 * The VM_SHARED test is necessary because shmem_zero_setup
1070 * will create the file object for a shared anonymous map below.
1072 if (!file && !(vm_flags & VM_SHARED) &&
1073 vma_merge(mm, prev, addr, addr + len, vm_flags,
1074 NULL, NULL, pgoff, NULL))
1078 * Determine the object being mapped and call the appropriate
1079 * specific mapper. the address has already been validated, but
1080 * not unmapped, but the maps are removed from the list.
1082 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1089 vma->vm_start = addr;
1090 vma->vm_end = addr + len;
1091 vma->vm_flags = vm_flags;
1092 vma->vm_page_prot = protection_map[vm_flags &
1093 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
1094 vma->vm_pgoff = pgoff;
1098 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1100 if (vm_flags & VM_DENYWRITE) {
1101 error = deny_write_access(file);
1106 vma->vm_file = file;
1108 error = file->f_op->mmap(file, vma);
1110 goto unmap_and_free_vma;
1111 } else if (vm_flags & VM_SHARED) {
1112 error = shmem_zero_setup(vma);
1117 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1118 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1119 * that memory reservation must be checked; but that reservation
1120 * belongs to shared memory object, not to vma: so now clear it.
1122 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1123 vma->vm_flags &= ~VM_ACCOUNT;
1125 /* Can addr have changed??
1127 * Answer: Yes, several device drivers can do it in their
1128 * f_op->mmap method. -DaveM
1130 addr = vma->vm_start;
1131 pgoff = vma->vm_pgoff;
1132 vm_flags = vma->vm_flags;
1134 if (vma_wants_writenotify(vma))
1136 protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC)];
1138 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1139 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1140 file = vma->vm_file;
1141 vma_link(mm, vma, prev, rb_link, rb_parent);
1143 atomic_inc(&inode->i_writecount);
1147 atomic_inc(&inode->i_writecount);
1150 mpol_free(vma_policy(vma));
1151 kmem_cache_free(vm_area_cachep, vma);
1154 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1155 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1156 if (vm_flags & VM_LOCKED) {
1157 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1158 make_pages_present(addr, addr + len);
1160 if (flags & MAP_POPULATE) {
1161 up_write(&mm->mmap_sem);
1162 sys_remap_file_pages(addr, len, 0,
1163 pgoff, flags & MAP_NONBLOCK);
1164 down_write(&mm->mmap_sem);
1170 atomic_inc(&inode->i_writecount);
1171 vma->vm_file = NULL;
1174 /* Undo any partial mapping done by a device driver. */
1175 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1178 kmem_cache_free(vm_area_cachep, vma);
1181 vm_unacct_memory(charged);
1185 EXPORT_SYMBOL(do_mmap_pgoff);
1187 /* Get an address range which is currently unmapped.
1188 * For shmat() with addr=0.
1190 * Ugly calling convention alert:
1191 * Return value with the low bits set means error value,
1193 * if (ret & ~PAGE_MASK)
1196 * This function "knows" that -ENOMEM has the bits set.
1198 #ifndef HAVE_ARCH_UNMAPPED_AREA
1200 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1201 unsigned long len, unsigned long pgoff, unsigned long flags)
1203 struct mm_struct *mm = current->mm;
1204 struct vm_area_struct *vma;
1205 unsigned long start_addr;
1207 if (len > TASK_SIZE)
1211 addr = PAGE_ALIGN(addr);
1212 vma = find_vma(mm, addr);
1213 if (TASK_SIZE - len >= addr &&
1214 (!vma || addr + len <= vma->vm_start))
1217 if (len > mm->cached_hole_size) {
1218 start_addr = addr = mm->free_area_cache;
1220 start_addr = addr = TASK_UNMAPPED_BASE;
1221 mm->cached_hole_size = 0;
1225 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1226 /* At this point: (!vma || addr < vma->vm_end). */
1227 if (TASK_SIZE - len < addr) {
1229 * Start a new search - just in case we missed
1232 if (start_addr != TASK_UNMAPPED_BASE) {
1233 addr = TASK_UNMAPPED_BASE;
1235 mm->cached_hole_size = 0;
1240 if (!vma || addr + len <= vma->vm_start) {
1242 * Remember the place where we stopped the search:
1244 mm->free_area_cache = addr + len;
1247 if (addr + mm->cached_hole_size < vma->vm_start)
1248 mm->cached_hole_size = vma->vm_start - addr;
1254 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1257 * Is this a new hole at the lowest possible address?
1259 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1260 mm->free_area_cache = addr;
1261 mm->cached_hole_size = ~0UL;
1266 * This mmap-allocator allocates new areas top-down from below the
1267 * stack's low limit (the base):
1269 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1271 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1272 const unsigned long len, const unsigned long pgoff,
1273 const unsigned long flags)
1275 struct vm_area_struct *vma;
1276 struct mm_struct *mm = current->mm;
1277 unsigned long addr = addr0;
1279 /* requested length too big for entire address space */
1280 if (len > TASK_SIZE)
1283 /* requesting a specific address */
1285 addr = PAGE_ALIGN(addr);
1286 vma = find_vma(mm, addr);
1287 if (TASK_SIZE - len >= addr &&
1288 (!vma || addr + len <= vma->vm_start))
1292 /* check if free_area_cache is useful for us */
1293 if (len <= mm->cached_hole_size) {
1294 mm->cached_hole_size = 0;
1295 mm->free_area_cache = mm->mmap_base;
1298 /* either no address requested or can't fit in requested address hole */
1299 addr = mm->free_area_cache;
1301 /* make sure it can fit in the remaining address space */
1303 vma = find_vma(mm, addr-len);
1304 if (!vma || addr <= vma->vm_start)
1305 /* remember the address as a hint for next time */
1306 return (mm->free_area_cache = addr-len);
1309 if (mm->mmap_base < len)
1312 addr = mm->mmap_base-len;
1316 * Lookup failure means no vma is above this address,
1317 * else if new region fits below vma->vm_start,
1318 * return with success:
1320 vma = find_vma(mm, addr);
1321 if (!vma || addr+len <= vma->vm_start)
1322 /* remember the address as a hint for next time */
1323 return (mm->free_area_cache = addr);
1325 /* remember the largest hole we saw so far */
1326 if (addr + mm->cached_hole_size < vma->vm_start)
1327 mm->cached_hole_size = vma->vm_start - addr;
1329 /* try just below the current vma->vm_start */
1330 addr = vma->vm_start-len;
1331 } while (len < vma->vm_start);
1335 * A failed mmap() very likely causes application failure,
1336 * so fall back to the bottom-up function here. This scenario
1337 * can happen with large stack limits and large mmap()
1340 mm->cached_hole_size = ~0UL;
1341 mm->free_area_cache = TASK_UNMAPPED_BASE;
1342 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1344 * Restore the topdown base:
1346 mm->free_area_cache = mm->mmap_base;
1347 mm->cached_hole_size = ~0UL;
1353 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1356 * Is this a new hole at the highest possible address?
1358 if (addr > mm->free_area_cache)
1359 mm->free_area_cache = addr;
1361 /* dont allow allocations above current base */
1362 if (mm->free_area_cache > mm->mmap_base)
1363 mm->free_area_cache = mm->mmap_base;
1368 get_unmapped_area_prot(struct file *file, unsigned long addr, unsigned long len,
1369 unsigned long pgoff, unsigned long flags, int exec)
1373 if (!(flags & MAP_FIXED)) {
1374 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1376 if (exec && current->mm->get_unmapped_exec_area)
1377 get_area = current->mm->get_unmapped_exec_area;
1379 get_area = current->mm->get_unmapped_area;
1381 if (file && file->f_op && file->f_op->get_unmapped_area)
1382 get_area = file->f_op->get_unmapped_area;
1383 addr = get_area(file, addr, len, pgoff, flags);
1384 if (IS_ERR_VALUE(addr))
1388 if (addr > TASK_SIZE - len)
1390 if (addr & ~PAGE_MASK)
1392 if (file && is_file_hugepages(file)) {
1394 * Check if the given range is hugepage aligned, and
1395 * can be made suitable for hugepages.
1397 ret = prepare_hugepage_range(addr, len);
1400 * Ensure that a normal request is not falling in a
1401 * reserved hugepage range. For some archs like IA-64,
1402 * there is a separate region for hugepages.
1404 ret = is_hugepage_only_range(current->mm, addr, len);
1411 EXPORT_SYMBOL(get_unmapped_area_prot);
1413 #define SHLIB_BASE 0x00110000
1415 unsigned long arch_get_unmapped_exec_area(struct file *filp, unsigned long addr0,
1416 unsigned long len0, unsigned long pgoff, unsigned long flags)
1418 unsigned long addr = addr0, len = len0;
1419 struct mm_struct *mm = current->mm;
1420 struct vm_area_struct *vma;
1423 if (len > TASK_SIZE)
1426 if (!addr && !(flags & MAP_FIXED))
1427 addr = randomize_range(SHLIB_BASE, 0x01000000, len);
1430 addr = PAGE_ALIGN(addr);
1431 vma = find_vma(mm, addr);
1432 if (TASK_SIZE - len >= addr &&
1433 (!vma || addr + len <= vma->vm_start)) {
1439 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1440 /* At this point: (!vma || addr < vma->vm_end). */
1441 if (TASK_SIZE - len < addr)
1444 if (!vma || addr + len <= vma->vm_start) {
1446 * Must not let a PROT_EXEC mapping get into the
1449 if (addr + len > mm->brk)
1453 * Up until the brk area we randomize addresses
1454 * as much as possible:
1456 if (addr >= 0x01000000) {
1457 tmp = randomize_range(0x01000000, PAGE_ALIGN(max(mm->start_brk, (unsigned long)0x08000000)), len);
1458 vma = find_vma(mm, tmp);
1459 if (TASK_SIZE - len >= tmp &&
1460 (!vma || tmp + len <= vma->vm_start))
1464 * Ok, randomization didnt work out - return
1465 * the result of the linear search:
1473 return current->mm->get_unmapped_area(filp, addr0, len0, pgoff, flags);
1477 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1478 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1480 struct vm_area_struct *vma = NULL;
1483 /* Check the cache first. */
1484 /* (Cache hit rate is typically around 35%.) */
1485 vma = mm->mmap_cache;
1486 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1487 struct rb_node * rb_node;
1489 rb_node = mm->mm_rb.rb_node;
1493 struct vm_area_struct * vma_tmp;
1495 vma_tmp = rb_entry(rb_node,
1496 struct vm_area_struct, vm_rb);
1498 if (vma_tmp->vm_end > addr) {
1500 if (vma_tmp->vm_start <= addr)
1502 rb_node = rb_node->rb_left;
1504 rb_node = rb_node->rb_right;
1507 mm->mmap_cache = vma;
1513 EXPORT_SYMBOL(find_vma);
1515 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1516 struct vm_area_struct *
1517 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1518 struct vm_area_struct **pprev)
1520 struct vm_area_struct *vma = NULL, *prev = NULL;
1521 struct rb_node * rb_node;
1525 /* Guard against addr being lower than the first VMA */
1528 /* Go through the RB tree quickly. */
1529 rb_node = mm->mm_rb.rb_node;
1532 struct vm_area_struct *vma_tmp;
1533 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1535 if (addr < vma_tmp->vm_end) {
1536 rb_node = rb_node->rb_left;
1539 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1541 rb_node = rb_node->rb_right;
1547 return prev ? prev->vm_next : vma;
1550 static int over_stack_limit(unsigned long sz)
1552 if (sz < EXEC_STACK_BIAS)
1554 return (sz - EXEC_STACK_BIAS) >
1555 current->signal->rlim[RLIMIT_STACK].rlim_cur;
1559 * Verify that the stack growth is acceptable and
1560 * update accounting. This is shared with both the
1561 * grow-up and grow-down cases.
1563 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1565 struct mm_struct *mm = vma->vm_mm;
1566 struct rlimit *rlim = current->signal->rlim;
1568 /* address space limit tests */
1569 if (!may_expand_vm(mm, grow))
1572 /* Stack limit test */
1573 if (over_stack_limit(size))
1576 /* mlock limit tests */
1577 if (vma->vm_flags & VM_LOCKED) {
1578 unsigned long locked;
1579 unsigned long limit;
1580 locked = mm->locked_vm + grow;
1581 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1582 if (locked > limit && !capable(CAP_IPC_LOCK))
1587 * Overcommit.. This must be the final test, as it will
1588 * update security statistics.
1590 if (security_vm_enough_memory(grow))
1593 /* Ok, everything looks good - let it rip */
1594 vx_vmpages_add(mm, grow);
1595 if (vma->vm_flags & VM_LOCKED)
1596 vx_vmlocked_add(mm, grow);
1597 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1601 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1603 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1604 * vma is the last one with address > vma->vm_end. Have to extend vma.
1609 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1613 if (!(vma->vm_flags & VM_GROWSUP))
1617 * We must make sure the anon_vma is allocated
1618 * so that the anon_vma locking is not a noop.
1620 if (unlikely(anon_vma_prepare(vma)))
1625 * vma->vm_start/vm_end cannot change under us because the caller
1626 * is required to hold the mmap_sem in read mode. We need the
1627 * anon_vma lock to serialize against concurrent expand_stacks.
1629 address += 4 + PAGE_SIZE - 1;
1630 address &= PAGE_MASK;
1633 /* Somebody else might have raced and expanded it already */
1634 if (address > vma->vm_end) {
1635 unsigned long size, grow;
1637 size = address - vma->vm_start;
1638 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1640 error = acct_stack_growth(vma, size, grow);
1642 vma->vm_end = address;
1644 anon_vma_unlock(vma);
1647 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1649 #ifdef CONFIG_STACK_GROWSUP
1650 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1652 return expand_upwards(vma, address);
1655 struct vm_area_struct *
1656 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1658 struct vm_area_struct *vma, *prev;
1661 vma = find_vma_prev(mm, addr, &prev);
1662 if (vma && (vma->vm_start <= addr))
1664 if (!prev || expand_stack(prev, addr))
1666 if (prev->vm_flags & VM_LOCKED) {
1667 make_pages_present(addr, prev->vm_end);
1673 * vma is the first one with address < vma->vm_start. Have to extend vma.
1675 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1680 * We must make sure the anon_vma is allocated
1681 * so that the anon_vma locking is not a noop.
1683 if (unlikely(anon_vma_prepare(vma)))
1688 * vma->vm_start/vm_end cannot change under us because the caller
1689 * is required to hold the mmap_sem in read mode. We need the
1690 * anon_vma lock to serialize against concurrent expand_stacks.
1692 address &= PAGE_MASK;
1695 /* Somebody else might have raced and expanded it already */
1696 if (address < vma->vm_start) {
1697 unsigned long size, grow;
1699 size = vma->vm_end - address;
1700 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1702 error = acct_stack_growth(vma, size, grow);
1704 vma->vm_start = address;
1705 vma->vm_pgoff -= grow;
1708 anon_vma_unlock(vma);
1712 struct vm_area_struct *
1713 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1715 struct vm_area_struct * vma;
1716 unsigned long start;
1719 vma = find_vma(mm,addr);
1722 if (vma->vm_start <= addr)
1724 if (!(vma->vm_flags & VM_GROWSDOWN))
1726 start = vma->vm_start;
1727 if (expand_stack(vma, addr))
1729 if (vma->vm_flags & VM_LOCKED) {
1730 make_pages_present(addr, start);
1737 * Ok - we have the memory areas we should free on the vma list,
1738 * so release them, and do the vma updates.
1740 * Called with the mm semaphore held.
1742 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1744 /* Update high watermark before we lower total_vm */
1745 update_hiwater_vm(mm);
1747 long nrpages = vma_pages(vma);
1749 vx_vmpages_sub(mm, nrpages);
1750 if (vma->vm_flags & VM_LOCKED)
1751 vx_vmlocked_sub(mm, nrpages);
1752 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1753 vma = remove_vma(vma);
1759 * Get rid of page table information in the indicated region.
1761 * Called with the mm semaphore held.
1763 static void unmap_region(struct mm_struct *mm,
1764 struct vm_area_struct *vma, struct vm_area_struct *prev,
1765 unsigned long start, unsigned long end)
1767 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1768 struct mmu_gather *tlb;
1769 unsigned long nr_accounted = 0;
1772 tlb = tlb_gather_mmu(mm, 0);
1773 update_hiwater_rss(mm);
1774 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1775 vm_unacct_memory(nr_accounted);
1776 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1777 next? next->vm_start: 0);
1778 tlb_finish_mmu(tlb, start, end);
1782 * Create a list of vma's touched by the unmap, removing them from the mm's
1783 * vma list as we go..
1786 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1787 struct vm_area_struct *prev, unsigned long end)
1789 struct vm_area_struct **insertion_point;
1790 struct vm_area_struct *tail_vma = NULL;
1793 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1795 rb_erase(&vma->vm_rb, &mm->mm_rb);
1799 } while (vma && vma->vm_start < end);
1800 *insertion_point = vma;
1801 tail_vma->vm_next = NULL;
1802 if (mm->unmap_area == arch_unmap_area)
1803 addr = prev ? prev->vm_end : mm->mmap_base;
1805 addr = vma ? vma->vm_start : mm->mmap_base;
1806 mm->unmap_area(mm, addr);
1807 mm->mmap_cache = NULL; /* Kill the cache. */
1811 * Split a vma into two pieces at address 'addr', a new vma is allocated
1812 * either for the first part or the the tail.
1814 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1815 unsigned long addr, int new_below)
1817 struct mempolicy *pol;
1818 struct vm_area_struct *new;
1820 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1823 if (mm->map_count >= sysctl_max_map_count)
1826 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1830 /* most fields are the same, copy all, and then fixup */
1836 new->vm_start = addr;
1837 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1840 pol = mpol_copy(vma_policy(vma));
1842 kmem_cache_free(vm_area_cachep, new);
1843 return PTR_ERR(pol);
1845 vma_set_policy(new, pol);
1848 get_file(new->vm_file);
1850 if (new->vm_ops && new->vm_ops->open)
1851 new->vm_ops->open(new);
1854 unsigned long old_end = vma->vm_end;
1856 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1857 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1858 if (vma->vm_flags & VM_EXEC)
1859 arch_remove_exec_range(mm, old_end);
1861 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1866 /* Munmap is split into 2 main parts -- this part which finds
1867 * what needs doing, and the areas themselves, which do the
1868 * work. This now handles partial unmappings.
1869 * Jeremy Fitzhardinge <jeremy@goop.org>
1871 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1874 struct vm_area_struct *vma, *prev, *last;
1876 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1879 if ((len = PAGE_ALIGN(len)) == 0)
1882 /* Find the first overlapping VMA */
1883 vma = find_vma_prev(mm, start, &prev);
1886 /* we have start < vma->vm_end */
1888 /* if it doesn't overlap, we have nothing.. */
1890 if (vma->vm_start >= end)
1894 * If we need to split any vma, do it now to save pain later.
1896 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1897 * unmapped vm_area_struct will remain in use: so lower split_vma
1898 * places tmp vma above, and higher split_vma places tmp vma below.
1900 if (start > vma->vm_start) {
1901 int error = split_vma(mm, vma, start, 0);
1907 /* Does it split the last one? */
1908 last = find_vma(mm, end);
1909 if (last && end > last->vm_start) {
1910 int error = split_vma(mm, last, end, 1);
1914 vma = prev? prev->vm_next: mm->mmap;
1917 * Remove the vma's, and unmap the actual pages
1919 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1920 unmap_region(mm, vma, prev, start, end);
1922 /* Fix up all other VM information */
1923 remove_vma_list(mm, vma);
1928 EXPORT_SYMBOL(do_munmap);
1930 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1933 struct mm_struct *mm = current->mm;
1935 profile_munmap(addr);
1937 down_write(&mm->mmap_sem);
1938 ret = do_munmap(mm, addr, len);
1939 up_write(&mm->mmap_sem);
1943 static inline void verify_mm_writelocked(struct mm_struct *mm)
1945 #ifdef CONFIG_DEBUG_VM
1946 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1948 up_read(&mm->mmap_sem);
1954 * this is really a simplified "do_mmap". it only handles
1955 * anonymous maps. eventually we may be able to do some
1956 * brk-specific accounting here.
1958 unsigned long do_brk(unsigned long addr, unsigned long len)
1960 struct mm_struct * mm = current->mm;
1961 struct vm_area_struct * vma, * prev;
1962 unsigned long flags;
1963 struct rb_node ** rb_link, * rb_parent;
1964 pgoff_t pgoff = addr >> PAGE_SHIFT;
1967 len = PAGE_ALIGN(len);
1971 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1974 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1976 error = arch_mmap_check(addr, len, flags);
1983 if (mm->def_flags & VM_LOCKED) {
1984 unsigned long locked, lock_limit;
1985 locked = len >> PAGE_SHIFT;
1986 locked += mm->locked_vm;
1987 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1988 lock_limit >>= PAGE_SHIFT;
1989 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1991 if (!vx_vmlocked_avail(mm, len >> PAGE_SHIFT))
1996 * mm->mmap_sem is required to protect against another thread
1997 * changing the mappings in case we sleep.
1999 verify_mm_writelocked(mm);
2002 * Clear old maps. this also does some error checking for us
2005 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2006 if (vma && vma->vm_start < addr + len) {
2007 if (do_munmap(mm, addr, len))
2012 /* Check against address space limits *after* clearing old maps... */
2013 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2016 if (mm->map_count > sysctl_max_map_count)
2019 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
2020 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
2023 /* Can we just expand an old private anonymous mapping? */
2024 if (vma_merge(mm, prev, addr, addr + len, flags,
2025 NULL, NULL, pgoff, NULL))
2029 * create a vma struct for an anonymous mapping
2031 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2033 vm_unacct_memory(len >> PAGE_SHIFT);
2038 vma->vm_start = addr;
2039 vma->vm_end = addr + len;
2040 vma->vm_pgoff = pgoff;
2041 vma->vm_flags = flags;
2042 vma->vm_page_prot = protection_map[flags &
2043 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
2044 vma_link(mm, vma, prev, rb_link, rb_parent);
2046 vx_vmpages_add(mm, len >> PAGE_SHIFT);
2047 if (flags & VM_LOCKED) {
2048 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
2049 make_pages_present(addr, addr + len);
2054 EXPORT_SYMBOL(do_brk);
2056 /* Release all mmaps. */
2057 void exit_mmap(struct mm_struct *mm)
2059 struct mmu_gather *tlb;
2060 struct vm_area_struct *vma = mm->mmap;
2061 unsigned long nr_accounted = 0;
2064 #ifdef arch_exit_mmap
2070 tlb = tlb_gather_mmu(mm, 1);
2071 /* Don't update_hiwater_rss(mm) here, do_exit already did */
2072 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2073 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2074 vm_unacct_memory(nr_accounted);
2075 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2076 tlb_finish_mmu(tlb, 0, end);
2077 arch_flush_exec_range(mm);
2079 set_mm_counter(mm, file_rss, 0);
2080 set_mm_counter(mm, anon_rss, 0);
2081 vx_vmpages_sub(mm, mm->total_vm);
2082 vx_vmlocked_sub(mm, mm->locked_vm);
2085 * Walk the list again, actually closing and freeing it,
2086 * with preemption enabled, without holding any MM locks.
2089 vma = remove_vma(vma);
2091 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2094 /* Insert vm structure into process list sorted by address
2095 * and into the inode's i_mmap tree. If vm_file is non-NULL
2096 * then i_mmap_lock is taken here.
2098 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2100 struct vm_area_struct * __vma, * prev;
2101 struct rb_node ** rb_link, * rb_parent;
2104 * The vm_pgoff of a purely anonymous vma should be irrelevant
2105 * until its first write fault, when page's anon_vma and index
2106 * are set. But now set the vm_pgoff it will almost certainly
2107 * end up with (unless mremap moves it elsewhere before that
2108 * first wfault), so /proc/pid/maps tells a consistent story.
2110 * By setting it to reflect the virtual start address of the
2111 * vma, merges and splits can happen in a seamless way, just
2112 * using the existing file pgoff checks and manipulations.
2113 * Similarly in do_mmap_pgoff and in do_brk.
2115 if (!vma->vm_file) {
2116 BUG_ON(vma->anon_vma);
2117 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2119 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2120 if (__vma && __vma->vm_start < vma->vm_end)
2122 if ((vma->vm_flags & VM_ACCOUNT) &&
2123 (security_vm_enough_memory(vma_pages(vma)) ||
2124 !vx_vmpages_avail(mm, vma_pages(vma))))
2126 vma_link(mm, vma, prev, rb_link, rb_parent);
2131 * Copy the vma structure to a new location in the same mm,
2132 * prior to moving page table entries, to effect an mremap move.
2134 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2135 unsigned long addr, unsigned long len, pgoff_t pgoff)
2137 struct vm_area_struct *vma = *vmap;
2138 unsigned long vma_start = vma->vm_start;
2139 struct mm_struct *mm = vma->vm_mm;
2140 struct vm_area_struct *new_vma, *prev;
2141 struct rb_node **rb_link, *rb_parent;
2142 struct mempolicy *pol;
2145 * If anonymous vma has not yet been faulted, update new pgoff
2146 * to match new location, to increase its chance of merging.
2148 if (!vma->vm_file && !vma->anon_vma)
2149 pgoff = addr >> PAGE_SHIFT;
2151 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2152 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2153 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2156 * Source vma may have been merged into new_vma
2158 if (vma_start >= new_vma->vm_start &&
2159 vma_start < new_vma->vm_end)
2162 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2165 pol = mpol_copy(vma_policy(vma));
2167 kmem_cache_free(vm_area_cachep, new_vma);
2170 vma_set_policy(new_vma, pol);
2171 new_vma->vm_start = addr;
2172 new_vma->vm_end = addr + len;
2173 new_vma->vm_pgoff = pgoff;
2174 if (new_vma->vm_file)
2175 get_file(new_vma->vm_file);
2176 if (new_vma->vm_ops && new_vma->vm_ops->open)
2177 new_vma->vm_ops->open(new_vma);
2178 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2185 * Return true if the calling process may expand its vm space by the passed
2188 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2190 unsigned long cur = mm->total_vm; /* pages */
2193 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2195 if (cur + npages > lim)
2197 if (!vx_vmpages_avail(mm, npages))
2203 static struct page *
2204 special_mapping_nopage(struct vm_area_struct *vma,
2205 unsigned long address, int *type)
2207 struct page **pages;
2209 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
2211 address -= vma->vm_start;
2212 for (pages = vma->vm_private_data; address > 0 && *pages; ++pages)
2213 address -= PAGE_SIZE;
2220 return NOPAGE_SIGBUS;
2223 static struct vm_operations_struct special_mapping_vmops = {
2224 .nopage = special_mapping_nopage,
2227 unsigned int vdso_populate = 1;
2230 * Insert a new vma covering the given region, with the given flags and
2231 * protections. Its pages are supplied by the given null-terminated array.
2232 * The region past the last page supplied will always produce SIGBUS.
2233 * The array pointer and the pages it points to are assumed to stay alive
2234 * for as long as this mapping might exist.
2236 int install_special_mapping(struct mm_struct *mm,
2237 unsigned long addr, unsigned long len,
2238 unsigned long vm_flags, pgprot_t pgprot,
2239 struct page **pages)
2241 struct vm_area_struct *vma;
2244 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2245 if (unlikely(vma == NULL))
2247 memset(vma, 0, sizeof(*vma));
2250 vma->vm_start = addr;
2251 vma->vm_end = addr + len;
2253 vma->vm_flags = vm_flags;
2254 vma->vm_page_prot = pgprot;
2256 vma->vm_ops = &special_mapping_vmops;
2257 vma->vm_private_data = pages;
2259 insert_vm_struct(mm, vma);
2260 vx_vmpages_add(mm, len >> PAGE_SHIFT);
2267 struct page *page = *pages++;
2269 err = install_page(mm, vma, addr, page, vma->vm_page_prot);