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/init.h>
17 #include <linux/file.h>
19 #include <linux/personality.h>
20 #include <linux/security.h>
21 #include <linux/hugetlb.h>
22 #include <linux/profile.h>
23 #include <linux/module.h>
24 #include <linux/acct.h>
25 #include <linux/mount.h>
26 #include <linux/mempolicy.h>
27 #include <linux/rmap.h>
29 #include <asm/uaccess.h>
30 #include <asm/cacheflush.h>
34 * WARNING: the debugging will use recursive algorithms so never enable this
35 * unless you know what you are doing.
39 /* description of effects of mapping type and prot in current implementation.
40 * this is due to the limited x86 page protection hardware. The expected
41 * behavior is in parens:
44 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
45 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
46 * w: (no) no w: (no) no w: (yes) yes w: (no) no
47 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
49 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
50 * w: (no) no w: (no) no w: (copy) copy w: (no) no
51 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
54 pgprot_t protection_map[16] = {
55 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
56 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
59 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
60 int sysctl_overcommit_ratio = 50; /* default is 50% */
61 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
62 atomic_t vm_committed_space = ATOMIC_INIT(0);
65 * Check that a process has enough memory to allocate a new virtual
66 * mapping. 0 means there is enough memory for the allocation to
67 * succeed and -ENOMEM implies there is not.
69 * We currently support three overcommit policies, which are set via the
70 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
72 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
73 * Additional code 2002 Jul 20 by Robert Love.
75 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
77 * Note this is a helper function intended to be used by LSMs which
78 * wish to use this logic.
80 int __vm_enough_memory(long pages, int cap_sys_admin)
82 unsigned long free, allowed;
84 vm_acct_memory(pages);
87 * Sometimes we want to use more memory than we have
89 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
92 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
95 free = get_page_cache_size();
96 free += nr_swap_pages;
99 * Any slabs which are created with the
100 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
101 * which are reclaimable, under pressure. The dentry
102 * cache and most inode caches should fall into this
104 free += atomic_read(&slab_reclaim_pages);
107 * Leave the last 3% for root
116 * nr_free_pages() is very expensive on large systems,
117 * only call if we're about to fail.
126 vm_unacct_memory(pages);
130 allowed = (totalram_pages - hugetlb_total_pages())
131 * sysctl_overcommit_ratio / 100;
133 * Leave the last 3% for root
136 allowed -= allowed / 32;
137 allowed += total_swap_pages;
139 /* Don't let a single process grow too big:
140 leave 3% of the size of this process for other processes */
141 allowed -= current->mm->total_vm / 32;
143 if (atomic_read(&vm_committed_space) < allowed)
146 vm_unacct_memory(pages);
151 EXPORT_SYMBOL(sysctl_overcommit_memory);
152 EXPORT_SYMBOL(sysctl_overcommit_ratio);
153 EXPORT_SYMBOL(sysctl_max_map_count);
154 EXPORT_SYMBOL(vm_committed_space);
155 EXPORT_SYMBOL(__vm_enough_memory);
158 * Requires inode->i_mapping->i_mmap_lock
160 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
161 struct file *file, struct address_space *mapping)
163 if (vma->vm_flags & VM_DENYWRITE)
164 atomic_inc(&file->f_dentry->d_inode->i_writecount);
165 if (vma->vm_flags & VM_SHARED)
166 mapping->i_mmap_writable--;
168 flush_dcache_mmap_lock(mapping);
169 if (unlikely(vma->vm_flags & VM_NONLINEAR))
170 list_del_init(&vma->shared.vm_set.list);
172 vma_prio_tree_remove(vma, &mapping->i_mmap);
173 flush_dcache_mmap_unlock(mapping);
177 * Remove one vm structure and free it.
179 static void remove_vm_struct(struct vm_area_struct *vma)
181 struct file *file = vma->vm_file;
185 struct address_space *mapping = file->f_mapping;
186 spin_lock(&mapping->i_mmap_lock);
187 __remove_shared_vm_struct(vma, file, mapping);
188 spin_unlock(&mapping->i_mmap_lock);
190 if (vma->vm_ops && vma->vm_ops->close)
191 vma->vm_ops->close(vma);
194 anon_vma_unlink(vma);
195 mpol_free(vma_policy(vma));
196 kmem_cache_free(vm_area_cachep, vma);
200 * sys_brk() for the most part doesn't need the global kernel
201 * lock, except when an application is doing something nasty
202 * like trying to un-brk an area that has already been mapped
203 * to a regular file. in this case, the unmapping will need
204 * to invoke file system routines that need the global lock.
206 asmlinkage unsigned long sys_brk(unsigned long brk)
208 unsigned long rlim, retval;
209 unsigned long newbrk, oldbrk;
210 struct mm_struct *mm = current->mm;
212 down_write(&mm->mmap_sem);
214 if (brk < mm->end_code)
216 newbrk = PAGE_ALIGN(brk);
217 oldbrk = PAGE_ALIGN(mm->brk);
218 if (oldbrk == newbrk)
221 /* Always allow shrinking brk. */
222 if (brk <= mm->brk) {
223 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
228 /* Check against rlimit.. */
229 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
230 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
233 /* Check against existing mmap mappings. */
234 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
237 /* Ok, looks good - let it rip. */
238 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
244 up_write(&mm->mmap_sem);
249 static int browse_rb(struct rb_root *root)
252 struct rb_node *nd, *pn = NULL;
253 unsigned long prev = 0, pend = 0;
255 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
256 struct vm_area_struct *vma;
257 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
258 if (vma->vm_start < prev)
259 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
260 if (vma->vm_start < pend)
261 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
262 if (vma->vm_start > vma->vm_end)
263 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
268 for (nd = pn; nd; nd = rb_prev(nd)) {
272 printk("backwards %d, forwards %d\n", j, i), i = 0;
276 void validate_mm(struct mm_struct *mm)
280 struct vm_area_struct *tmp = mm->mmap;
285 if (i != mm->map_count)
286 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
287 i = browse_rb(&mm->mm_rb);
288 if (i != mm->map_count)
289 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
294 #define validate_mm(mm) do { } while (0)
297 static struct vm_area_struct *
298 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
299 struct vm_area_struct **pprev, struct rb_node ***rb_link,
300 struct rb_node ** rb_parent)
302 struct vm_area_struct * vma;
303 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
305 __rb_link = &mm->mm_rb.rb_node;
306 rb_prev = __rb_parent = NULL;
310 struct vm_area_struct *vma_tmp;
312 __rb_parent = *__rb_link;
313 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
315 if (vma_tmp->vm_end > addr) {
317 if (vma_tmp->vm_start <= addr)
319 __rb_link = &__rb_parent->rb_left;
321 rb_prev = __rb_parent;
322 __rb_link = &__rb_parent->rb_right;
328 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
329 *rb_link = __rb_link;
330 *rb_parent = __rb_parent;
335 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
336 struct vm_area_struct *prev, struct rb_node *rb_parent)
339 vma->vm_next = prev->vm_next;
344 vma->vm_next = rb_entry(rb_parent,
345 struct vm_area_struct, vm_rb);
351 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
352 struct rb_node **rb_link, struct rb_node *rb_parent)
354 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
355 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
358 static inline void __vma_link_file(struct vm_area_struct *vma)
364 struct address_space *mapping = file->f_mapping;
366 if (vma->vm_flags & VM_DENYWRITE)
367 atomic_dec(&file->f_dentry->d_inode->i_writecount);
368 if (vma->vm_flags & VM_SHARED)
369 mapping->i_mmap_writable++;
371 flush_dcache_mmap_lock(mapping);
372 if (unlikely(vma->vm_flags & VM_NONLINEAR))
373 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
375 vma_prio_tree_insert(vma, &mapping->i_mmap);
376 flush_dcache_mmap_unlock(mapping);
381 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
382 struct vm_area_struct *prev, struct rb_node **rb_link,
383 struct rb_node *rb_parent)
385 __vma_link_list(mm, vma, prev, rb_parent);
386 __vma_link_rb(mm, vma, rb_link, rb_parent);
387 __anon_vma_link(vma);
390 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
391 struct vm_area_struct *prev, struct rb_node **rb_link,
392 struct rb_node *rb_parent)
394 struct address_space *mapping = NULL;
397 mapping = vma->vm_file->f_mapping;
400 spin_lock(&mapping->i_mmap_lock);
401 vma->vm_truncate_count = mapping->truncate_count;
405 __vma_link(mm, vma, prev, rb_link, rb_parent);
406 __vma_link_file(vma);
408 anon_vma_unlock(vma);
410 spin_unlock(&mapping->i_mmap_lock);
417 * Helper for vma_adjust in the split_vma insert case:
418 * insert vm structure into list and rbtree and anon_vma,
419 * but it has already been inserted into prio_tree earlier.
422 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
424 struct vm_area_struct * __vma, * prev;
425 struct rb_node ** rb_link, * rb_parent;
427 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
428 if (__vma && __vma->vm_start < vma->vm_end)
430 __vma_link(mm, vma, prev, rb_link, rb_parent);
435 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
436 struct vm_area_struct *prev)
438 prev->vm_next = vma->vm_next;
439 rb_erase(&vma->vm_rb, &mm->mm_rb);
440 if (mm->mmap_cache == vma)
441 mm->mmap_cache = prev;
445 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
446 * is already present in an i_mmap tree without adjusting the tree.
447 * The following helper function should be used when such adjustments
448 * are necessary. The "insert" vma (if any) is to be inserted
449 * before we drop the necessary locks.
451 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
452 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
454 struct mm_struct *mm = vma->vm_mm;
455 struct vm_area_struct *next = vma->vm_next;
456 struct vm_area_struct *importer = NULL;
457 struct address_space *mapping = NULL;
458 struct prio_tree_root *root = NULL;
459 struct file *file = vma->vm_file;
460 struct anon_vma *anon_vma = NULL;
461 long adjust_next = 0;
464 if (next && !insert) {
465 if (end >= next->vm_end) {
467 * vma expands, overlapping all the next, and
468 * perhaps the one after too (mprotect case 6).
470 again: remove_next = 1 + (end > next->vm_end);
472 anon_vma = next->anon_vma;
474 } else if (end > next->vm_start) {
476 * vma expands, overlapping part of the next:
477 * mprotect case 5 shifting the boundary up.
479 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
480 anon_vma = next->anon_vma;
482 } else if (end < vma->vm_end) {
484 * vma shrinks, and !insert tells it's not
485 * split_vma inserting another: so it must be
486 * mprotect case 4 shifting the boundary down.
488 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
489 anon_vma = next->anon_vma;
495 mapping = file->f_mapping;
496 if (!(vma->vm_flags & VM_NONLINEAR))
497 root = &mapping->i_mmap;
498 spin_lock(&mapping->i_mmap_lock);
500 vma->vm_truncate_count != next->vm_truncate_count) {
502 * unmap_mapping_range might be in progress:
503 * ensure that the expanding vma is rescanned.
505 importer->vm_truncate_count = 0;
508 insert->vm_truncate_count = vma->vm_truncate_count;
510 * Put into prio_tree now, so instantiated pages
511 * are visible to arm/parisc __flush_dcache_page
512 * throughout; but we cannot insert into address
513 * space until vma start or end is updated.
515 __vma_link_file(insert);
520 * When changing only vma->vm_end, we don't really need
521 * anon_vma lock: but is that case worth optimizing out?
524 anon_vma = vma->anon_vma;
526 spin_lock(&anon_vma->lock);
528 * Easily overlooked: when mprotect shifts the boundary,
529 * make sure the expanding vma has anon_vma set if the
530 * shrinking vma had, to cover any anon pages imported.
532 if (importer && !importer->anon_vma) {
533 importer->anon_vma = anon_vma;
534 __anon_vma_link(importer);
539 flush_dcache_mmap_lock(mapping);
540 vma_prio_tree_remove(vma, root);
542 vma_prio_tree_remove(next, root);
545 vma->vm_start = start;
547 vma->vm_pgoff = pgoff;
549 next->vm_start += adjust_next << PAGE_SHIFT;
550 next->vm_pgoff += adjust_next;
555 vma_prio_tree_insert(next, root);
556 vma_prio_tree_insert(vma, root);
557 flush_dcache_mmap_unlock(mapping);
562 * vma_merge has merged next into vma, and needs
563 * us to remove next before dropping the locks.
565 __vma_unlink(mm, next, vma);
567 __remove_shared_vm_struct(next, file, mapping);
569 __anon_vma_merge(vma, next);
572 * split_vma has split insert from vma, and needs
573 * us to insert it before dropping the locks
574 * (it may either follow vma or precede it).
576 __insert_vm_struct(mm, insert);
580 spin_unlock(&anon_vma->lock);
582 spin_unlock(&mapping->i_mmap_lock);
588 mpol_free(vma_policy(next));
589 kmem_cache_free(vm_area_cachep, next);
591 * In mprotect's case 6 (see comments on vma_merge),
592 * we must remove another next too. It would clutter
593 * up the code too much to do both in one go.
595 if (remove_next == 2) {
605 * If the vma has a ->close operation then the driver probably needs to release
606 * per-vma resources, so we don't attempt to merge those.
608 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
610 static inline int is_mergeable_vma(struct vm_area_struct *vma,
611 struct file *file, unsigned long vm_flags)
613 if (vma->vm_flags != vm_flags)
615 if (vma->vm_file != file)
617 if (vma->vm_ops && vma->vm_ops->close)
622 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
623 struct anon_vma *anon_vma2)
625 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
629 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
630 * in front of (at a lower virtual address and file offset than) the vma.
632 * We cannot merge two vmas if they have differently assigned (non-NULL)
633 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
635 * We don't check here for the merged mmap wrapping around the end of pagecache
636 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
637 * wrap, nor mmaps which cover the final page at index -1UL.
640 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
641 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
643 if (is_mergeable_vma(vma, file, vm_flags) &&
644 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
645 if (vma->vm_pgoff == vm_pgoff)
652 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
653 * beyond (at a higher virtual address and file offset than) the vma.
655 * We cannot merge two vmas if they have differently assigned (non-NULL)
656 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
659 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
660 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
662 if (is_mergeable_vma(vma, file, vm_flags) &&
663 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
665 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
666 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
673 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
674 * whether that can be merged with its predecessor or its successor.
675 * Or both (it neatly fills a hole).
677 * In most cases - when called for mmap, brk or mremap - [addr,end) is
678 * certain not to be mapped by the time vma_merge is called; but when
679 * called for mprotect, it is certain to be already mapped (either at
680 * an offset within prev, or at the start of next), and the flags of
681 * this area are about to be changed to vm_flags - and the no-change
682 * case has already been eliminated.
684 * The following mprotect cases have to be considered, where AAAA is
685 * the area passed down from mprotect_fixup, never extending beyond one
686 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
688 * AAAA AAAA AAAA AAAA
689 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
690 * cannot merge might become might become might become
691 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
692 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
693 * mremap move: PPPPNNNNNNNN 8
695 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
696 * might become case 1 below case 2 below case 3 below
698 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
699 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
701 struct vm_area_struct *vma_merge(struct mm_struct *mm,
702 struct vm_area_struct *prev, unsigned long addr,
703 unsigned long end, unsigned long vm_flags,
704 struct anon_vma *anon_vma, struct file *file,
705 pgoff_t pgoff, struct mempolicy *policy)
707 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
708 struct vm_area_struct *area, *next;
711 * We later require that vma->vm_flags == vm_flags,
712 * so this tests vma->vm_flags & VM_SPECIAL, too.
714 if (vm_flags & VM_SPECIAL)
718 next = prev->vm_next;
722 if (next && next->vm_end == end) /* cases 6, 7, 8 */
723 next = next->vm_next;
726 * Can it merge with the predecessor?
728 if (prev && prev->vm_end == addr &&
729 mpol_equal(vma_policy(prev), policy) &&
730 can_vma_merge_after(prev, vm_flags,
731 anon_vma, file, pgoff)) {
733 * OK, it can. Can we now merge in the successor as well?
735 if (next && end == next->vm_start &&
736 mpol_equal(policy, vma_policy(next)) &&
737 can_vma_merge_before(next, vm_flags,
738 anon_vma, file, pgoff+pglen) &&
739 is_mergeable_anon_vma(prev->anon_vma,
742 vma_adjust(prev, prev->vm_start,
743 next->vm_end, prev->vm_pgoff, NULL);
744 } else /* cases 2, 5, 7 */
745 vma_adjust(prev, prev->vm_start,
746 end, prev->vm_pgoff, NULL);
751 * Can this new request be merged in front of next?
753 if (next && end == next->vm_start &&
754 mpol_equal(policy, vma_policy(next)) &&
755 can_vma_merge_before(next, vm_flags,
756 anon_vma, file, pgoff+pglen)) {
757 if (prev && addr < prev->vm_end) /* case 4 */
758 vma_adjust(prev, prev->vm_start,
759 addr, prev->vm_pgoff, NULL);
760 else /* cases 3, 8 */
761 vma_adjust(area, addr, next->vm_end,
762 next->vm_pgoff - pglen, NULL);
770 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
771 * neighbouring vmas for a suitable anon_vma, before it goes off
772 * to allocate a new anon_vma. It checks because a repetitive
773 * sequence of mprotects and faults may otherwise lead to distinct
774 * anon_vmas being allocated, preventing vma merge in subsequent
777 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
779 struct vm_area_struct *near;
780 unsigned long vm_flags;
787 * Since only mprotect tries to remerge vmas, match flags
788 * which might be mprotected into each other later on.
789 * Neither mlock nor madvise tries to remerge at present,
790 * so leave their flags as obstructing a merge.
792 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
793 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
795 if (near->anon_vma && vma->vm_end == near->vm_start &&
796 mpol_equal(vma_policy(vma), vma_policy(near)) &&
797 can_vma_merge_before(near, vm_flags,
798 NULL, vma->vm_file, vma->vm_pgoff +
799 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
800 return near->anon_vma;
803 * It is potentially slow to have to call find_vma_prev here.
804 * But it's only on the first write fault on the vma, not
805 * every time, and we could devise a way to avoid it later
806 * (e.g. stash info in next's anon_vma_node when assigning
807 * an anon_vma, or when trying vma_merge). Another time.
809 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
814 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
815 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
817 if (near->anon_vma && near->vm_end == vma->vm_start &&
818 mpol_equal(vma_policy(near), vma_policy(vma)) &&
819 can_vma_merge_after(near, vm_flags,
820 NULL, vma->vm_file, vma->vm_pgoff))
821 return near->anon_vma;
824 * There's no absolute need to look only at touching neighbours:
825 * we could search further afield for "compatible" anon_vmas.
826 * But it would probably just be a waste of time searching,
827 * or lead to too many vmas hanging off the same anon_vma.
828 * We're trying to allow mprotect remerging later on,
829 * not trying to minimize memory used for anon_vmas.
834 #ifdef CONFIG_PROC_FS
835 void __vm_stat_account(struct mm_struct *mm, unsigned long flags,
836 struct file *file, long pages)
838 const unsigned long stack_flags
839 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
841 #ifdef CONFIG_HUGETLB
842 if (flags & VM_HUGETLB) {
843 if (!(flags & VM_DONTCOPY))
844 mm->shared_vm += pages;
847 #endif /* CONFIG_HUGETLB */
850 mm->shared_vm += pages;
851 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
852 mm->exec_vm += pages;
853 } else if (flags & stack_flags)
854 mm->stack_vm += pages;
855 if (flags & (VM_RESERVED|VM_IO))
856 mm->reserved_vm += pages;
858 #endif /* CONFIG_PROC_FS */
861 * The caller must hold down_write(current->mm->mmap_sem).
864 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
865 unsigned long len, unsigned long prot,
866 unsigned long flags, unsigned long pgoff)
868 struct mm_struct * mm = current->mm;
869 struct vm_area_struct * vma, * prev;
871 unsigned int vm_flags;
872 int correct_wcount = 0;
874 struct rb_node ** rb_link, * rb_parent;
876 unsigned long charged = 0;
879 if (is_file_hugepages(file))
882 if (!file->f_op || !file->f_op->mmap)
885 if ((prot & PROT_EXEC) &&
886 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
890 * Does the application expect PROT_READ to imply PROT_EXEC?
892 * (the exception is when the underlying filesystem is noexec
893 * mounted, in which case we dont add PROT_EXEC.)
895 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
896 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
902 /* Careful about overflows.. */
903 len = PAGE_ALIGN(len);
904 if (!len || len > TASK_SIZE)
907 /* offset overflow? */
908 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
911 /* Too many mappings? */
912 if (mm->map_count > sysctl_max_map_count)
915 /* Obtain the address to map to. we verify (or select) it and ensure
916 * that it represents a valid section of the address space.
918 addr = get_unmapped_area(file, addr, len, pgoff, flags);
919 if (addr & ~PAGE_MASK)
922 /* Do simple checking here so the lower-level routines won't have
923 * to. we assume access permissions have been handled by the open
924 * of the memory object, so we don't do any here.
926 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
927 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
929 if (flags & MAP_LOCKED) {
932 vm_flags |= VM_LOCKED;
934 /* mlock MCL_FUTURE? */
935 if (vm_flags & VM_LOCKED) {
936 unsigned long locked, lock_limit;
937 locked = mm->locked_vm << PAGE_SHIFT;
938 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
940 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
944 inode = file ? file->f_dentry->d_inode : NULL;
947 switch (flags & MAP_TYPE) {
949 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
953 * Make sure we don't allow writing to an append-only
956 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
960 * Make sure there are no mandatory locks on the file.
962 if (locks_verify_locked(inode))
965 vm_flags |= VM_SHARED | VM_MAYSHARE;
966 if (!(file->f_mode & FMODE_WRITE))
967 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
971 if (!(file->f_mode & FMODE_READ))
979 switch (flags & MAP_TYPE) {
981 vm_flags |= VM_SHARED | VM_MAYSHARE;
985 * Set pgoff according to addr for anon_vma.
987 pgoff = addr >> PAGE_SHIFT;
994 error = security_file_mmap(file, prot, flags);
1001 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1002 if (vma && vma->vm_start < addr + len) {
1003 if (do_munmap(mm, addr, len))
1008 /* Check against address space limit. */
1009 if ((mm->total_vm << PAGE_SHIFT) + len
1010 > current->signal->rlim[RLIMIT_AS].rlim_cur)
1013 /* check context space, maybe only Private writable mapping? */
1014 if (!vx_vmpages_avail(mm, len >> PAGE_SHIFT))
1017 if (accountable && (!(flags & MAP_NORESERVE) ||
1018 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1019 if (vm_flags & VM_SHARED) {
1020 /* Check memory availability in shmem_file_setup? */
1021 vm_flags |= VM_ACCOUNT;
1022 } else if (vm_flags & VM_WRITE) {
1024 * Private writable mapping: check memory availability
1026 charged = len >> PAGE_SHIFT;
1027 if (security_vm_enough_memory(charged))
1029 vm_flags |= VM_ACCOUNT;
1034 * Can we just expand an old private anonymous mapping?
1035 * The VM_SHARED test is necessary because shmem_zero_setup
1036 * will create the file object for a shared anonymous map below.
1038 if (!file && !(vm_flags & VM_SHARED) &&
1039 vma_merge(mm, prev, addr, addr + len, vm_flags,
1040 NULL, NULL, pgoff, NULL))
1044 * Determine the object being mapped and call the appropriate
1045 * specific mapper. the address has already been validated, but
1046 * not unmapped, but the maps are removed from the list.
1048 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1053 memset(vma, 0, sizeof(*vma));
1056 vma->vm_start = addr;
1057 vma->vm_end = addr + len;
1058 vma->vm_flags = vm_flags;
1059 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1060 vma->vm_pgoff = pgoff;
1064 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1066 if (vm_flags & VM_DENYWRITE) {
1067 error = deny_write_access(file);
1072 vma->vm_file = file;
1074 error = file->f_op->mmap(file, vma);
1076 goto unmap_and_free_vma;
1077 } else if (vm_flags & VM_SHARED) {
1078 error = shmem_zero_setup(vma);
1083 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1084 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1085 * that memory reservation must be checked; but that reservation
1086 * belongs to shared memory object, not to vma: so now clear it.
1088 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1089 vma->vm_flags &= ~VM_ACCOUNT;
1091 /* Can addr have changed??
1093 * Answer: Yes, several device drivers can do it in their
1094 * f_op->mmap method. -DaveM
1096 addr = vma->vm_start;
1097 pgoff = vma->vm_pgoff;
1098 vm_flags = vma->vm_flags;
1100 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1101 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1102 file = vma->vm_file;
1103 vma_link(mm, vma, prev, rb_link, rb_parent);
1105 atomic_inc(&inode->i_writecount);
1109 atomic_inc(&inode->i_writecount);
1112 mpol_free(vma_policy(vma));
1113 kmem_cache_free(vm_area_cachep, vma);
1116 // mm->total_vm += len >> PAGE_SHIFT;
1117 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1118 __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1119 if (vm_flags & VM_LOCKED) {
1120 // mm->locked_vm += len >> PAGE_SHIFT;
1121 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1122 make_pages_present(addr, addr + len);
1124 if (flags & MAP_POPULATE) {
1125 up_write(&mm->mmap_sem);
1126 sys_remap_file_pages(addr, len, 0,
1127 pgoff, flags & MAP_NONBLOCK);
1128 down_write(&mm->mmap_sem);
1130 acct_update_integrals();
1131 update_mem_hiwater();
1136 atomic_inc(&inode->i_writecount);
1137 vma->vm_file = NULL;
1140 /* Undo any partial mapping done by a device driver. */
1141 zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1143 kmem_cache_free(vm_area_cachep, vma);
1146 vm_unacct_memory(charged);
1150 EXPORT_SYMBOL(do_mmap_pgoff);
1152 /* Get an address range which is currently unmapped.
1153 * For shmat() with addr=0.
1155 * Ugly calling convention alert:
1156 * Return value with the low bits set means error value,
1158 * if (ret & ~PAGE_MASK)
1161 * This function "knows" that -ENOMEM has the bits set.
1163 #ifndef HAVE_ARCH_UNMAPPED_AREA
1165 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1166 unsigned long len, unsigned long pgoff, unsigned long flags)
1168 struct mm_struct *mm = current->mm;
1169 struct vm_area_struct *vma;
1170 unsigned long start_addr;
1172 if (len > TASK_SIZE)
1176 addr = PAGE_ALIGN(addr);
1177 vma = find_vma(mm, addr);
1178 if (TASK_SIZE - len >= addr &&
1179 (!vma || addr + len <= vma->vm_start))
1182 start_addr = addr = mm->free_area_cache;
1185 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1186 /* At this point: (!vma || addr < vma->vm_end). */
1187 if (TASK_SIZE - len < addr) {
1189 * Start a new search - just in case we missed
1192 if (start_addr != TASK_UNMAPPED_BASE) {
1193 start_addr = addr = TASK_UNMAPPED_BASE;
1198 if (!vma || addr + len <= vma->vm_start) {
1200 * Remember the place where we stopped the search:
1202 mm->free_area_cache = addr + len;
1210 void arch_unmap_area(struct vm_area_struct *area)
1213 * Is this a new hole at the lowest possible address?
1215 if (area->vm_start >= TASK_UNMAPPED_BASE &&
1216 area->vm_start < area->vm_mm->free_area_cache)
1217 area->vm_mm->free_area_cache = area->vm_start;
1221 * This mmap-allocator allocates new areas top-down from below the
1222 * stack's low limit (the base):
1224 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1226 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1227 const unsigned long len, const unsigned long pgoff,
1228 const unsigned long flags)
1230 struct vm_area_struct *vma, *prev_vma;
1231 struct mm_struct *mm = current->mm;
1232 unsigned long base = mm->mmap_base, addr = addr0;
1235 /* requested length too big for entire address space */
1236 if (len > TASK_SIZE)
1239 /* dont allow allocations above current base */
1240 if (mm->free_area_cache > base)
1241 mm->free_area_cache = base;
1243 /* requesting a specific address */
1245 addr = PAGE_ALIGN(addr);
1246 vma = find_vma(mm, addr);
1247 if (TASK_SIZE - len >= addr &&
1248 (!vma || addr + len <= vma->vm_start))
1253 /* make sure it can fit in the remaining address space */
1254 if (mm->free_area_cache < len)
1257 /* either no address requested or cant fit in requested address hole */
1258 addr = (mm->free_area_cache - len) & PAGE_MASK;
1261 * Lookup failure means no vma is above this address,
1262 * i.e. return with success:
1264 if (!(vma = find_vma_prev(mm, addr, &prev_vma)))
1268 * new region fits between prev_vma->vm_end and
1269 * vma->vm_start, use it:
1271 if (addr+len <= vma->vm_start &&
1272 (!prev_vma || (addr >= prev_vma->vm_end)))
1273 /* remember the address as a hint for next time */
1274 return (mm->free_area_cache = addr);
1276 /* pull free_area_cache down to the first hole */
1277 if (mm->free_area_cache == vma->vm_end)
1278 mm->free_area_cache = vma->vm_start;
1280 /* try just below the current vma->vm_start */
1281 addr = vma->vm_start-len;
1282 } while (len <= vma->vm_start);
1286 * if hint left us with no space for the requested
1287 * mapping then try again:
1290 mm->free_area_cache = base;
1295 * A failed mmap() very likely causes application failure,
1296 * so fall back to the bottom-up function here. This scenario
1297 * can happen with large stack limits and large mmap()
1300 mm->free_area_cache = TASK_UNMAPPED_BASE;
1301 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1303 * Restore the topdown base:
1305 mm->free_area_cache = base;
1311 void arch_unmap_area_topdown(struct vm_area_struct *area)
1314 * Is this a new hole at the highest possible address?
1316 if (area->vm_end > area->vm_mm->free_area_cache)
1317 area->vm_mm->free_area_cache = area->vm_end;
1321 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1322 unsigned long pgoff, unsigned long flags)
1324 if (flags & MAP_FIXED) {
1327 if (addr > TASK_SIZE - len)
1329 if (addr & ~PAGE_MASK)
1331 if (file && is_file_hugepages(file)) {
1333 * Check if the given range is hugepage aligned, and
1334 * can be made suitable for hugepages.
1336 ret = prepare_hugepage_range(addr, len);
1339 * Ensure that a normal request is not falling in a
1340 * reserved hugepage range. For some archs like IA-64,
1341 * there is a separate region for hugepages.
1343 ret = is_hugepage_only_range(addr, len);
1350 if (file && file->f_op && file->f_op->get_unmapped_area)
1351 return file->f_op->get_unmapped_area(file, addr, len,
1354 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
1357 EXPORT_SYMBOL(get_unmapped_area);
1359 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1360 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1362 struct vm_area_struct *vma = NULL;
1365 /* Check the cache first. */
1366 /* (Cache hit rate is typically around 35%.) */
1367 vma = mm->mmap_cache;
1368 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1369 struct rb_node * rb_node;
1371 rb_node = mm->mm_rb.rb_node;
1375 struct vm_area_struct * vma_tmp;
1377 vma_tmp = rb_entry(rb_node,
1378 struct vm_area_struct, vm_rb);
1380 if (vma_tmp->vm_end > addr) {
1382 if (vma_tmp->vm_start <= addr)
1384 rb_node = rb_node->rb_left;
1386 rb_node = rb_node->rb_right;
1389 mm->mmap_cache = vma;
1395 EXPORT_SYMBOL(find_vma);
1397 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1398 struct vm_area_struct *
1399 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1400 struct vm_area_struct **pprev)
1402 struct vm_area_struct *vma = NULL, *prev = NULL;
1403 struct rb_node * rb_node;
1407 /* Guard against addr being lower than the first VMA */
1410 /* Go through the RB tree quickly. */
1411 rb_node = mm->mm_rb.rb_node;
1414 struct vm_area_struct *vma_tmp;
1415 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1417 if (addr < vma_tmp->vm_end) {
1418 rb_node = rb_node->rb_left;
1421 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1423 rb_node = rb_node->rb_right;
1429 return prev ? prev->vm_next : vma;
1433 * Verify that the stack growth is acceptable and
1434 * update accounting. This is shared with both the
1435 * grow-up and grow-down cases.
1437 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1439 struct mm_struct *mm = vma->vm_mm;
1440 struct rlimit *rlim = current->signal->rlim;
1442 /* address space limit tests */
1443 if (mm->total_vm + grow > rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT)
1446 /* Stack limit test */
1447 if (size > rlim[RLIMIT_STACK].rlim_cur)
1450 /* mlock limit tests */
1451 if (vma->vm_flags & VM_LOCKED) {
1452 unsigned long locked;
1453 unsigned long limit;
1454 locked = mm->locked_vm + grow;
1455 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1456 if (locked > limit && !capable(CAP_IPC_LOCK))
1460 if (!vx_vmpages_avail(vma->vm_mm, grow))
1464 * Overcommit.. This must be the final test, as it will
1465 * update security statistics.
1467 if (security_vm_enough_memory(grow))
1470 /* Ok, everything looks good - let it rip */
1471 // mm->total_vm += grow;
1472 vx_vmpages_add(mm, grow);
1473 if (vma->vm_flags & VM_LOCKED)
1474 // mm->locked_vm += grow;
1475 vx_vmlocked_add(mm, grow);
1476 __vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1477 acct_update_integrals();
1478 update_mem_hiwater();
1482 #ifdef CONFIG_STACK_GROWSUP
1484 * vma is the first one with address > vma->vm_end. Have to extend vma.
1486 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1490 if (!(vma->vm_flags & VM_GROWSUP))
1494 * We must make sure the anon_vma is allocated
1495 * so that the anon_vma locking is not a noop.
1497 if (unlikely(anon_vma_prepare(vma)))
1502 * vma->vm_start/vm_end cannot change under us because the caller
1503 * is required to hold the mmap_sem in read mode. We need the
1504 * anon_vma lock to serialize against concurrent expand_stacks.
1506 address += 4 + PAGE_SIZE - 1;
1507 address &= PAGE_MASK;
1510 /* Somebody else might have raced and expanded it already */
1511 if (address > vma->vm_end) {
1512 unsigned long size, grow;
1514 size = address - vma->vm_start;
1515 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1517 error = acct_stack_growth(vma, size, grow);
1519 vma->vm_end = address;
1521 anon_vma_unlock(vma);
1525 struct vm_area_struct *
1526 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1528 struct vm_area_struct *vma, *prev;
1531 vma = find_vma_prev(mm, addr, &prev);
1532 if (vma && (vma->vm_start <= addr))
1534 if (!prev || expand_stack(prev, addr))
1536 if (prev->vm_flags & VM_LOCKED) {
1537 make_pages_present(addr, prev->vm_end);
1543 * vma is the first one with address < vma->vm_start. Have to extend vma.
1545 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1550 * We must make sure the anon_vma is allocated
1551 * so that the anon_vma locking is not a noop.
1553 if (unlikely(anon_vma_prepare(vma)))
1558 * vma->vm_start/vm_end cannot change under us because the caller
1559 * is required to hold the mmap_sem in read mode. We need the
1560 * anon_vma lock to serialize against concurrent expand_stacks.
1562 address &= PAGE_MASK;
1565 /* Somebody else might have raced and expanded it already */
1566 if (address < vma->vm_start) {
1567 unsigned long size, grow;
1569 size = vma->vm_end - address;
1570 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1572 error = acct_stack_growth(vma, size, grow);
1574 vma->vm_start = address;
1575 vma->vm_pgoff -= grow;
1578 anon_vma_unlock(vma);
1582 struct vm_area_struct *
1583 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1585 struct vm_area_struct * vma;
1586 unsigned long start;
1589 vma = find_vma(mm,addr);
1592 if (vma->vm_start <= addr)
1594 if (!(vma->vm_flags & VM_GROWSDOWN))
1596 start = vma->vm_start;
1597 if (expand_stack(vma, addr))
1599 if (vma->vm_flags & VM_LOCKED) {
1600 make_pages_present(addr, start);
1607 * Try to free as many page directory entries as we can,
1608 * without having to work very hard at actually scanning
1609 * the page tables themselves.
1611 * Right now we try to free page tables if we have a nice
1612 * PGDIR-aligned area that got free'd up. We could be more
1613 * granular if we want to, but this is fast and simple,
1614 * and covers the bad cases.
1616 * "prev", if it exists, points to a vma before the one
1617 * we just free'd - but there's no telling how much before.
1619 static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev,
1620 unsigned long start, unsigned long end)
1622 unsigned long first = start & PGDIR_MASK;
1623 unsigned long last = end + PGDIR_SIZE - 1;
1624 struct mm_struct *mm = tlb->mm;
1626 if (last > MM_VM_SIZE(mm) || last < end)
1627 last = MM_VM_SIZE(mm);
1633 if (prev->vm_end > start) {
1634 if (last > prev->vm_start)
1635 last = prev->vm_start;
1640 struct vm_area_struct *next = prev->vm_next;
1643 if (next->vm_start < start) {
1647 if (last > next->vm_start)
1648 last = next->vm_start;
1650 if (prev->vm_end > first)
1651 first = prev->vm_end;
1655 if (last < first) /* for arches with discontiguous pgd indices */
1657 if (first < FIRST_USER_PGD_NR * PGDIR_SIZE)
1658 first = FIRST_USER_PGD_NR * PGDIR_SIZE;
1659 /* No point trying to free anything if we're in the same pte page */
1660 if ((first & PMD_MASK) < (last & PMD_MASK)) {
1661 clear_page_range(tlb, first, last);
1662 flush_tlb_pgtables(mm, first, last);
1666 /* Normal function to fix up a mapping
1667 * This function is the default for when an area has no specific
1668 * function. This may be used as part of a more specific routine.
1670 * By the time this function is called, the area struct has been
1671 * removed from the process mapping list.
1673 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1675 size_t len = area->vm_end - area->vm_start;
1677 // area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1678 vx_vmpages_sub(area->vm_mm, len >> PAGE_SHIFT);
1680 if (area->vm_flags & VM_LOCKED)
1681 // area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1682 vx_vmlocked_sub(area->vm_mm, len >> PAGE_SHIFT);
1683 vm_stat_unaccount(area);
1684 area->vm_mm->unmap_area(area);
1685 remove_vm_struct(area);
1689 * Update the VMA and inode share lists.
1691 * Ok - we have the memory areas we should free on the 'free' list,
1692 * so release them, and do the vma updates.
1694 static void unmap_vma_list(struct mm_struct *mm,
1695 struct vm_area_struct *mpnt)
1698 struct vm_area_struct *next = mpnt->vm_next;
1699 unmap_vma(mm, mpnt);
1701 } while (mpnt != NULL);
1706 * Get rid of page table information in the indicated region.
1708 * Called with the page table lock held.
1710 static void unmap_region(struct mm_struct *mm,
1711 struct vm_area_struct *vma,
1712 struct vm_area_struct *prev,
1713 unsigned long start,
1716 struct mmu_gather *tlb;
1717 unsigned long nr_accounted = 0;
1720 tlb = tlb_gather_mmu(mm, 0);
1721 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1722 vm_unacct_memory(nr_accounted);
1724 if (is_hugepage_only_range(start, end - start))
1725 hugetlb_free_pgtables(tlb, prev, start, end);
1727 free_pgtables(tlb, prev, start, end);
1728 tlb_finish_mmu(tlb, start, end);
1732 * Create a list of vma's touched by the unmap, removing them from the mm's
1733 * vma list as we go..
1736 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1737 struct vm_area_struct *prev, unsigned long end)
1739 struct vm_area_struct **insertion_point;
1740 struct vm_area_struct *tail_vma = NULL;
1742 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1744 rb_erase(&vma->vm_rb, &mm->mm_rb);
1748 } while (vma && vma->vm_start < end);
1749 *insertion_point = vma;
1750 tail_vma->vm_next = NULL;
1751 mm->mmap_cache = NULL; /* Kill the cache. */
1755 * Split a vma into two pieces at address 'addr', a new vma is allocated
1756 * either for the first part or the the tail.
1758 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1759 unsigned long addr, int new_below)
1761 struct mempolicy *pol;
1762 struct vm_area_struct *new;
1764 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1767 if (mm->map_count >= sysctl_max_map_count)
1770 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1774 /* most fields are the same, copy all, and then fixup */
1780 new->vm_start = addr;
1781 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1784 pol = mpol_copy(vma_policy(vma));
1786 kmem_cache_free(vm_area_cachep, new);
1787 return PTR_ERR(pol);
1789 vma_set_policy(new, pol);
1792 get_file(new->vm_file);
1794 if (new->vm_ops && new->vm_ops->open)
1795 new->vm_ops->open(new);
1798 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1799 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1801 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1806 /* Munmap is split into 2 main parts -- this part which finds
1807 * what needs doing, and the areas themselves, which do the
1808 * work. This now handles partial unmappings.
1809 * Jeremy Fitzhardinge <jeremy@goop.org>
1811 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1814 struct vm_area_struct *mpnt, *prev, *last;
1816 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1819 if ((len = PAGE_ALIGN(len)) == 0)
1822 /* Find the first overlapping VMA */
1823 mpnt = find_vma_prev(mm, start, &prev);
1826 /* we have start < mpnt->vm_end */
1828 /* if it doesn't overlap, we have nothing.. */
1830 if (mpnt->vm_start >= end)
1834 * If we need to split any vma, do it now to save pain later.
1836 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1837 * unmapped vm_area_struct will remain in use: so lower split_vma
1838 * places tmp vma above, and higher split_vma places tmp vma below.
1840 if (start > mpnt->vm_start) {
1841 int error = split_vma(mm, mpnt, start, 0);
1847 /* Does it split the last one? */
1848 last = find_vma(mm, end);
1849 if (last && end > last->vm_start) {
1850 int error = split_vma(mm, last, end, 1);
1854 mpnt = prev? prev->vm_next: mm->mmap;
1857 * Remove the vma's, and unmap the actual pages
1859 detach_vmas_to_be_unmapped(mm, mpnt, prev, end);
1860 spin_lock(&mm->page_table_lock);
1861 unmap_region(mm, mpnt, prev, start, end);
1862 spin_unlock(&mm->page_table_lock);
1864 /* Fix up all other VM information */
1865 unmap_vma_list(mm, mpnt);
1870 EXPORT_SYMBOL(do_munmap);
1872 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1875 struct mm_struct *mm = current->mm;
1877 profile_munmap(addr);
1879 down_write(&mm->mmap_sem);
1880 ret = do_munmap(mm, addr, len);
1881 up_write(&mm->mmap_sem);
1885 static inline void verify_mm_writelocked(struct mm_struct *mm)
1887 #ifdef CONFIG_DEBUG_KERNEL
1888 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1890 up_read(&mm->mmap_sem);
1896 * this is really a simplified "do_mmap". it only handles
1897 * anonymous maps. eventually we may be able to do some
1898 * brk-specific accounting here.
1900 unsigned long do_brk(unsigned long addr, unsigned long len)
1902 struct mm_struct * mm = current->mm;
1903 struct vm_area_struct * vma, * prev;
1904 unsigned long flags;
1905 struct rb_node ** rb_link, * rb_parent;
1906 pgoff_t pgoff = addr >> PAGE_SHIFT;
1908 len = PAGE_ALIGN(len);
1912 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1918 if (mm->def_flags & VM_LOCKED) {
1919 unsigned long locked, lock_limit;
1920 locked = mm->locked_vm << PAGE_SHIFT;
1921 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1923 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1925 if (!vx_vmlocked_avail(mm, len >> PAGE_SHIFT))
1930 * mm->mmap_sem is required to protect against another thread
1931 * changing the mappings in case we sleep.
1933 verify_mm_writelocked(mm);
1936 * Clear old maps. this also does some error checking for us
1939 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1940 if (vma && vma->vm_start < addr + len) {
1941 if (do_munmap(mm, addr, len))
1946 /* Check against address space limits *after* clearing old maps... */
1947 if ((mm->total_vm << PAGE_SHIFT) + len
1948 > current->signal->rlim[RLIMIT_AS].rlim_cur)
1951 if (mm->map_count > sysctl_max_map_count)
1954 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
1955 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
1958 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1960 /* Can we just expand an old private anonymous mapping? */
1961 if (vma_merge(mm, prev, addr, addr + len, flags,
1962 NULL, NULL, pgoff, NULL))
1966 * create a vma struct for an anonymous mapping
1968 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1970 vm_unacct_memory(len >> PAGE_SHIFT);
1973 memset(vma, 0, sizeof(*vma));
1976 vma->vm_start = addr;
1977 vma->vm_end = addr + len;
1978 vma->vm_pgoff = pgoff;
1979 vma->vm_flags = flags;
1980 vma->vm_page_prot = protection_map[flags & 0x0f];
1981 vma_link(mm, vma, prev, rb_link, rb_parent);
1983 // mm->total_vm += len >> PAGE_SHIFT;
1984 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1985 if (flags & VM_LOCKED) {
1986 // mm->locked_vm += len >> PAGE_SHIFT;
1987 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1988 make_pages_present(addr, addr + len);
1990 acct_update_integrals();
1991 update_mem_hiwater();
1995 EXPORT_SYMBOL(do_brk);
1997 /* Release all mmaps. */
1998 void exit_mmap(struct mm_struct *mm)
2000 struct mmu_gather *tlb;
2001 struct vm_area_struct *vma;
2002 unsigned long nr_accounted = 0;
2006 spin_lock(&mm->page_table_lock);
2008 tlb = tlb_gather_mmu(mm, 1);
2010 /* Use ~0UL here to ensure all VMAs in the mm are unmapped */
2011 mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0,
2012 ~0UL, &nr_accounted, NULL);
2013 vm_unacct_memory(nr_accounted);
2014 BUG_ON(mm->map_count); /* This is just debugging */
2015 clear_page_range(tlb, FIRST_USER_PGD_NR * PGDIR_SIZE, MM_VM_SIZE(mm));
2017 tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm));
2020 mm->mmap = mm->mmap_cache = NULL;
2021 mm->mm_rb = RB_ROOT;
2023 vx_rsspages_sub(mm, mm->rss);
2024 // mm->total_vm = 0;
2025 vx_vmpages_sub(mm, mm->total_vm);
2026 // mm->locked_vm = 0;
2027 vx_vmlocked_sub(mm, mm->locked_vm);
2029 spin_unlock(&mm->page_table_lock);
2032 * Walk the list again, actually closing and freeing it
2033 * without holding any MM locks.
2036 struct vm_area_struct *next = vma->vm_next;
2037 remove_vm_struct(vma);
2042 /* Insert vm structure into process list sorted by address
2043 * and into the inode's i_mmap tree. If vm_file is non-NULL
2044 * then i_mmap_lock is taken here.
2046 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2048 struct vm_area_struct * __vma, * prev;
2049 struct rb_node ** rb_link, * rb_parent;
2052 * The vm_pgoff of a purely anonymous vma should be irrelevant
2053 * until its first write fault, when page's anon_vma and index
2054 * are set. But now set the vm_pgoff it will almost certainly
2055 * end up with (unless mremap moves it elsewhere before that
2056 * first wfault), so /proc/pid/maps tells a consistent story.
2058 * By setting it to reflect the virtual start address of the
2059 * vma, merges and splits can happen in a seamless way, just
2060 * using the existing file pgoff checks and manipulations.
2061 * Similarly in do_mmap_pgoff and in do_brk.
2063 if (!vma->vm_file) {
2064 BUG_ON(vma->anon_vma);
2065 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2067 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2068 if (__vma && __vma->vm_start < vma->vm_end)
2070 vma_link(mm, vma, prev, rb_link, rb_parent);
2075 * Copy the vma structure to a new location in the same mm,
2076 * prior to moving page table entries, to effect an mremap move.
2078 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2079 unsigned long addr, unsigned long len, pgoff_t pgoff)
2081 struct vm_area_struct *vma = *vmap;
2082 unsigned long vma_start = vma->vm_start;
2083 struct mm_struct *mm = vma->vm_mm;
2084 struct vm_area_struct *new_vma, *prev;
2085 struct rb_node **rb_link, *rb_parent;
2086 struct mempolicy *pol;
2089 * If anonymous vma has not yet been faulted, update new pgoff
2090 * to match new location, to increase its chance of merging.
2092 if (!vma->vm_file && !vma->anon_vma)
2093 pgoff = addr >> PAGE_SHIFT;
2095 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2096 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2097 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2100 * Source vma may have been merged into new_vma
2102 if (vma_start >= new_vma->vm_start &&
2103 vma_start < new_vma->vm_end)
2106 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2109 pol = mpol_copy(vma_policy(vma));
2111 kmem_cache_free(vm_area_cachep, new_vma);
2114 vma_set_policy(new_vma, pol);
2115 new_vma->vm_start = addr;
2116 new_vma->vm_end = addr + len;
2117 new_vma->vm_pgoff = pgoff;
2118 if (new_vma->vm_file)
2119 get_file(new_vma->vm_file);
2120 if (new_vma->vm_ops && new_vma->vm_ops->open)
2121 new_vma->vm_ops->open(new_vma);
2122 vma_link(mm, new_vma, prev, rb_link, rb_parent);