6 * Address space accounting code <alan@redhat.com>
9 #include <linux/slab.h>
10 #include <linux/shm.h>
11 #include <linux/mman.h>
12 #include <linux/pagemap.h>
13 #include <linux/swap.h>
14 #include <linux/syscalls.h>
15 #include <linux/init.h>
16 #include <linux/file.h>
18 #include <linux/personality.h>
19 #include <linux/security.h>
20 #include <linux/hugetlb.h>
21 #include <linux/profile.h>
22 #include <linux/module.h>
23 #include <linux/mount.h>
24 #include <linux/mempolicy.h>
25 #include <linux/rmap.h>
27 #include <asm/uaccess.h>
28 #include <asm/cacheflush.h>
32 * WARNING: the debugging will use recursive algorithms so never enable this
33 * unless you know what you are doing.
37 /* description of effects of mapping type and prot in current implementation.
38 * this is due to the limited x86 page protection hardware. The expected
39 * behavior is in parens:
42 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
43 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
44 * w: (no) no w: (no) no w: (yes) yes w: (no) no
45 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
47 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
48 * w: (no) no w: (no) no w: (copy) copy w: (no) no
49 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
52 pgprot_t protection_map[16] = {
53 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
54 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
57 int sysctl_overcommit_memory = 0; /* default is heuristic overcommit */
58 int sysctl_overcommit_ratio = 50; /* default is 50% */
59 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
60 atomic_t vm_committed_space = ATOMIC_INIT(0);
62 EXPORT_SYMBOL(sysctl_overcommit_memory);
63 EXPORT_SYMBOL(sysctl_overcommit_ratio);
64 EXPORT_SYMBOL(sysctl_max_map_count);
65 EXPORT_SYMBOL(vm_committed_space);
68 * Requires inode->i_mapping->i_mmap_lock
70 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
71 struct file *file, struct address_space *mapping)
73 if (vma->vm_flags & VM_DENYWRITE)
74 atomic_inc(&file->f_dentry->d_inode->i_writecount);
75 if (vma->vm_flags & VM_SHARED)
76 mapping->i_mmap_writable--;
78 flush_dcache_mmap_lock(mapping);
79 if (unlikely(vma->vm_flags & VM_NONLINEAR))
80 list_del_init(&vma->shared.vm_set.list);
82 vma_prio_tree_remove(vma, &mapping->i_mmap);
83 flush_dcache_mmap_unlock(mapping);
87 * Remove one vm structure and free it.
89 static void remove_vm_struct(struct vm_area_struct *vma)
91 struct file *file = vma->vm_file;
95 struct address_space *mapping = file->f_mapping;
96 spin_lock(&mapping->i_mmap_lock);
97 __remove_shared_vm_struct(vma, file, mapping);
98 spin_unlock(&mapping->i_mmap_lock);
100 if (vma->vm_ops && vma->vm_ops->close)
101 vma->vm_ops->close(vma);
104 anon_vma_unlink(vma);
105 mpol_free(vma_policy(vma));
106 kmem_cache_free(vm_area_cachep, vma);
110 * sys_brk() for the most part doesn't need the global kernel
111 * lock, except when an application is doing something nasty
112 * like trying to un-brk an area that has already been mapped
113 * to a regular file. in this case, the unmapping will need
114 * to invoke file system routines that need the global lock.
116 asmlinkage unsigned long sys_brk(unsigned long brk)
118 unsigned long rlim, retval;
119 unsigned long newbrk, oldbrk;
120 struct mm_struct *mm = current->mm;
122 down_write(&mm->mmap_sem);
124 if (brk < mm->end_code)
126 newbrk = PAGE_ALIGN(brk);
127 oldbrk = PAGE_ALIGN(mm->brk);
128 if (oldbrk == newbrk)
131 /* Always allow shrinking brk. */
132 if (brk <= mm->brk) {
133 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
138 /* Check against rlimit.. */
139 rlim = current->rlim[RLIMIT_DATA].rlim_cur;
140 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
143 /* Check against existing mmap mappings. */
144 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
147 /* Ok, looks good - let it rip. */
148 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
154 up_write(&mm->mmap_sem);
159 static int browse_rb(struct rb_root *root)
162 struct rb_node *nd, *pn = NULL;
163 unsigned long prev = 0, pend = 0;
165 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
166 struct vm_area_struct *vma;
167 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
168 if (vma->vm_start < prev)
169 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
170 if (vma->vm_start < pend)
171 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
172 if (vma->vm_start > vma->vm_end)
173 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
178 for (nd = pn; nd; nd = rb_prev(nd)) {
182 printk("backwards %d, forwards %d\n", j, i), i = 0;
186 void validate_mm(struct mm_struct *mm)
190 struct vm_area_struct *tmp = mm->mmap;
195 if (i != mm->map_count)
196 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
197 i = browse_rb(&mm->mm_rb);
198 if (i != mm->map_count)
199 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
204 #define validate_mm(mm) do { } while (0)
207 static struct vm_area_struct *
208 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
209 struct vm_area_struct **pprev, struct rb_node ***rb_link,
210 struct rb_node ** rb_parent)
212 struct vm_area_struct * vma;
213 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
215 __rb_link = &mm->mm_rb.rb_node;
216 rb_prev = __rb_parent = NULL;
220 struct vm_area_struct *vma_tmp;
222 __rb_parent = *__rb_link;
223 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
225 if (vma_tmp->vm_end > addr) {
227 if (vma_tmp->vm_start <= addr)
229 __rb_link = &__rb_parent->rb_left;
231 rb_prev = __rb_parent;
232 __rb_link = &__rb_parent->rb_right;
238 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
239 *rb_link = __rb_link;
240 *rb_parent = __rb_parent;
245 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
246 struct vm_area_struct *prev, struct rb_node *rb_parent)
248 if (vma->vm_flags & VM_EXEC)
249 arch_add_exec_range(mm, vma->vm_end);
251 vma->vm_next = prev->vm_next;
256 vma->vm_next = rb_entry(rb_parent,
257 struct vm_area_struct, vm_rb);
263 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
264 struct rb_node **rb_link, struct rb_node *rb_parent)
266 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
267 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
270 static inline void __vma_link_file(struct vm_area_struct *vma)
276 struct address_space *mapping = file->f_mapping;
278 if (vma->vm_flags & VM_DENYWRITE)
279 atomic_dec(&file->f_dentry->d_inode->i_writecount);
280 if (vma->vm_flags & VM_SHARED)
281 mapping->i_mmap_writable++;
283 flush_dcache_mmap_lock(mapping);
284 if (unlikely(vma->vm_flags & VM_NONLINEAR))
285 list_add_tail(&vma->shared.vm_set.list,
286 &mapping->i_mmap_nonlinear);
288 vma_prio_tree_insert(vma, &mapping->i_mmap);
289 flush_dcache_mmap_unlock(mapping);
294 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
295 struct vm_area_struct *prev, struct rb_node **rb_link,
296 struct rb_node *rb_parent)
298 __vma_link_list(mm, vma, prev, rb_parent);
299 __vma_link_rb(mm, vma, rb_link, rb_parent);
300 __anon_vma_link(vma);
303 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
304 struct vm_area_struct *prev, struct rb_node **rb_link,
305 struct rb_node *rb_parent)
307 struct address_space *mapping = NULL;
310 mapping = vma->vm_file->f_mapping;
313 spin_lock(&mapping->i_mmap_lock);
316 __vma_link(mm, vma, prev, rb_link, rb_parent);
317 __vma_link_file(vma);
319 anon_vma_unlock(vma);
321 spin_unlock(&mapping->i_mmap_lock);
328 * Helper for vma_adjust in the split_vma insert case:
329 * insert vm structure into list and rbtree and anon_vma,
330 * but it has already been inserted into prio_tree earlier.
333 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
335 struct vm_area_struct * __vma, * prev;
336 struct rb_node ** rb_link, * rb_parent;
338 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
339 if (__vma && __vma->vm_start < vma->vm_end)
341 __vma_link(mm, vma, prev, rb_link, rb_parent);
346 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
347 struct vm_area_struct *prev)
349 prev->vm_next = vma->vm_next;
350 rb_erase(&vma->vm_rb, &mm->mm_rb);
351 if (mm->mmap_cache == vma)
352 mm->mmap_cache = prev;
353 if (vma->vm_flags & VM_EXEC)
354 arch_remove_exec_range(mm, vma->vm_end);
358 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
359 * is already present in an i_mmap tree without adjusting the tree.
360 * The following helper function should be used when such adjustments
361 * are necessary. The "insert" vma (if any) is to be inserted
362 * before we drop the necessary locks.
364 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
365 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
367 struct mm_struct *mm = vma->vm_mm;
368 struct vm_area_struct *next = vma->vm_next;
369 struct vm_area_struct *importer = NULL;
370 struct address_space *mapping = NULL;
371 struct prio_tree_root *root = NULL;
372 struct file *file = vma->vm_file;
373 struct anon_vma *anon_vma = NULL;
374 long adjust_next = 0;
377 if (next && !insert) {
378 if (end >= next->vm_end) {
380 * vma expands, overlapping all the next, and
381 * perhaps the one after too (mprotect case 6).
383 again: remove_next = 1 + (end > next->vm_end);
385 anon_vma = next->anon_vma;
386 } else if (end > next->vm_start) {
388 * vma expands, overlapping part of the next:
389 * mprotect case 5 shifting the boundary up.
391 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
392 anon_vma = next->anon_vma;
394 } else if (end < vma->vm_end) {
396 * vma shrinks, and !insert tells it's not
397 * split_vma inserting another: so it must be
398 * mprotect case 4 shifting the boundary down.
400 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
401 anon_vma = next->anon_vma;
407 mapping = file->f_mapping;
408 if (!(vma->vm_flags & VM_NONLINEAR))
409 root = &mapping->i_mmap;
410 spin_lock(&mapping->i_mmap_lock);
413 * Put into prio_tree now, so instantiated pages
414 * are visible to arm/parisc __flush_dcache_page
415 * throughout; but we cannot insert into address
416 * space until vma start or end is updated.
418 __vma_link_file(insert);
423 * When changing only vma->vm_end, we don't really need
424 * anon_vma lock: but is that case worth optimizing out?
427 anon_vma = vma->anon_vma;
429 spin_lock(&anon_vma->lock);
431 * Easily overlooked: when mprotect shifts the boundary,
432 * make sure the expanding vma has anon_vma set if the
433 * shrinking vma had, to cover any anon pages imported.
435 if (importer && !importer->anon_vma) {
436 importer->anon_vma = anon_vma;
437 __anon_vma_link(importer);
442 flush_dcache_mmap_lock(mapping);
443 vma_prio_tree_remove(vma, root);
445 vma_prio_tree_remove(next, root);
448 vma->vm_start = start;
450 vma->vm_pgoff = pgoff;
452 next->vm_start += adjust_next << PAGE_SHIFT;
453 next->vm_pgoff += adjust_next;
458 vma_prio_tree_init(next);
459 vma_prio_tree_insert(next, root);
461 vma_prio_tree_init(vma);
462 vma_prio_tree_insert(vma, root);
463 flush_dcache_mmap_unlock(mapping);
468 * vma_merge has merged next into vma, and needs
469 * us to remove next before dropping the locks.
471 __vma_unlink(mm, next, vma);
473 __remove_shared_vm_struct(next, file, mapping);
475 __anon_vma_merge(vma, next);
478 * split_vma has split insert from vma, and needs
479 * us to insert it before dropping the locks
480 * (it may either follow vma or precede it).
482 __insert_vm_struct(mm, insert);
486 spin_unlock(&anon_vma->lock);
488 spin_unlock(&mapping->i_mmap_lock);
494 mpol_free(vma_policy(next));
495 kmem_cache_free(vm_area_cachep, next);
497 * In mprotect's case 6 (see comments on vma_merge),
498 * we must remove another next too. It would clutter
499 * up the code too much to do both in one go.
501 if (remove_next == 2) {
511 * If the vma has a ->close operation then the driver probably needs to release
512 * per-vma resources, so we don't attempt to merge those.
514 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
516 static inline int is_mergeable_vma(struct vm_area_struct *vma,
517 struct file *file, unsigned long vm_flags)
519 if (vma->vm_flags != vm_flags)
521 if (vma->vm_file != file)
523 if (vma->vm_ops && vma->vm_ops->close)
528 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
529 struct anon_vma *anon_vma2)
531 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
535 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
536 * in front of (at a lower virtual address and file offset than) the vma.
538 * We cannot merge two vmas if they have differently assigned (non-NULL)
539 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
541 * We don't check here for the merged mmap wrapping around the end of pagecache
542 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
543 * wrap, nor mmaps which cover the final page at index -1UL.
546 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
547 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
549 if (is_mergeable_vma(vma, file, vm_flags) &&
550 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
551 if (vma->vm_pgoff == vm_pgoff)
558 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
559 * beyond (at a higher virtual address and file offset than) the vma.
561 * We cannot merge two vmas if they have differently assigned (non-NULL)
562 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
565 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
566 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
568 if (is_mergeable_vma(vma, file, vm_flags) &&
569 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
571 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
572 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
579 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
580 * whether that can be merged with its predecessor or its successor.
581 * Or both (it neatly fills a hole).
583 * In most cases - when called for mmap, brk or mremap - [addr,end) is
584 * certain not to be mapped by the time vma_merge is called; but when
585 * called for mprotect, it is certain to be already mapped (either at
586 * an offset within prev, or at the start of next), and the flags of
587 * this area are about to be changed to vm_flags - and the no-change
588 * case has already been eliminated.
590 * The following mprotect cases have to be considered, where AAAA is
591 * the area passed down from mprotect_fixup, never extending beyond one
592 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
594 * AAAA AAAA AAAA AAAA
595 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
596 * cannot merge might become might become might become
597 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
598 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
599 * mremap move: PPPPNNNNNNNN 8
601 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
602 * might become case 1 below case 2 below case 3 below
604 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
605 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
607 struct vm_area_struct *vma_merge(struct mm_struct *mm,
608 struct vm_area_struct *prev, unsigned long addr,
609 unsigned long end, unsigned long vm_flags,
610 struct anon_vma *anon_vma, struct file *file,
611 pgoff_t pgoff, struct mempolicy *policy)
613 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
614 struct vm_area_struct *area, *next;
617 * We later require that vma->vm_flags == vm_flags,
618 * so this tests vma->vm_flags & VM_SPECIAL, too.
620 if (vm_flags & VM_SPECIAL)
624 next = prev->vm_next;
628 if (next && next->vm_end == end) /* cases 6, 7, 8 */
629 next = next->vm_next;
632 * Can it merge with the predecessor?
634 if (prev && prev->vm_end == addr &&
635 mpol_equal(vma_policy(prev), policy) &&
636 can_vma_merge_after(prev, vm_flags,
637 anon_vma, file, pgoff)) {
639 * OK, it can. Can we now merge in the successor as well?
641 if (next && end == next->vm_start &&
642 mpol_equal(policy, vma_policy(next)) &&
643 can_vma_merge_before(next, vm_flags,
644 anon_vma, file, pgoff+pglen) &&
645 is_mergeable_anon_vma(prev->anon_vma,
648 vma_adjust(prev, prev->vm_start,
649 next->vm_end, prev->vm_pgoff, NULL);
650 } else /* cases 2, 5, 7 */
651 vma_adjust(prev, prev->vm_start,
652 end, prev->vm_pgoff, NULL);
653 if (prev->vm_flags & VM_EXEC)
654 arch_add_exec_range(mm, prev->vm_end);
659 * Can this new request be merged in front of next?
661 if (next && end == next->vm_start &&
662 mpol_equal(policy, vma_policy(next)) &&
663 can_vma_merge_before(next, vm_flags,
664 anon_vma, file, pgoff+pglen)) {
665 if (prev && addr < prev->vm_end) /* case 4 */
666 vma_adjust(prev, prev->vm_start,
667 addr, prev->vm_pgoff, NULL);
668 else /* cases 3, 8 */
669 vma_adjust(area, addr, next->vm_end,
670 next->vm_pgoff - pglen, NULL);
678 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
679 * neighbouring vmas for a suitable anon_vma, before it goes off
680 * to allocate a new anon_vma. It checks because a repetitive
681 * sequence of mprotects and faults may otherwise lead to distinct
682 * anon_vmas being allocated, preventing vma merge in subsequent
685 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
687 struct vm_area_struct *near;
688 unsigned long vm_flags;
695 * Since only mprotect tries to remerge vmas, match flags
696 * which might be mprotected into each other later on.
697 * Neither mlock nor madvise tries to remerge at present,
698 * so leave their flags as obstructing a merge.
700 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
701 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
703 if (near->anon_vma && vma->vm_end == near->vm_start &&
704 mpol_equal(vma_policy(vma), vma_policy(near)) &&
705 can_vma_merge_before(near, vm_flags,
706 NULL, vma->vm_file, vma->vm_pgoff +
707 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
708 return near->anon_vma;
711 * It is potentially slow to have to call find_vma_prev here.
712 * But it's only on the first write fault on the vma, not
713 * every time, and we could devise a way to avoid it later
714 * (e.g. stash info in next's anon_vma_node when assigning
715 * an anon_vma, or when trying vma_merge). Another time.
717 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
722 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
723 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
725 if (near->anon_vma && near->vm_end == vma->vm_start &&
726 mpol_equal(vma_policy(near), vma_policy(vma)) &&
727 can_vma_merge_after(near, vm_flags,
728 NULL, vma->vm_file, vma->vm_pgoff))
729 return near->anon_vma;
732 * There's no absolute need to look only at touching neighbours:
733 * we could search further afield for "compatible" anon_vmas.
734 * But it would probably just be a waste of time searching,
735 * or lead to too many vmas hanging off the same anon_vma.
736 * We're trying to allow mprotect remerging later on,
737 * not trying to minimize memory used for anon_vmas.
743 * The caller must hold down_write(current->mm->mmap_sem).
746 unsigned long do_mmap_pgoff(struct mm_struct *mm, struct file * file,
747 unsigned long addr, unsigned long len,
748 unsigned long prot, unsigned long flags,
751 struct vm_area_struct * vma, * prev;
753 unsigned int vm_flags;
754 int correct_wcount = 0;
756 struct rb_node ** rb_link, * rb_parent;
758 unsigned long charged = 0;
761 if (is_file_hugepages(file))
764 if (!file->f_op || !file->f_op->mmap)
767 if ((prot & PROT_EXEC) &&
768 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
775 /* Careful about overflows.. */
776 len = PAGE_ALIGN(len);
777 if (!len || len > TASK_SIZE)
780 /* offset overflow? */
781 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
784 /* Too many mappings? */
785 if (mm->map_count > sysctl_max_map_count)
788 /* Obtain the address to map to. we verify (or select) it and ensure
789 * that it represents a valid section of the address space.
791 addr = get_unmapped_area(file, addr, len, pgoff, flags, prot & PROT_EXEC);
792 if (addr & ~PAGE_MASK)
795 /* Do simple checking here so the lower-level routines won't have
796 * to. we assume access permissions have been handled by the open
797 * of the memory object, so we don't do any here.
799 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
800 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
802 if (flags & MAP_LOCKED) {
805 vm_flags |= VM_LOCKED;
807 /* mlock MCL_FUTURE? */
808 if (vm_flags & VM_LOCKED) {
809 unsigned long locked, lock_limit;
810 locked = mm->locked_vm << PAGE_SHIFT;
811 lock_limit = current->rlim[RLIMIT_MEMLOCK].rlim_cur;
813 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
817 inode = file ? file->f_dentry->d_inode : NULL;
820 switch (flags & MAP_TYPE) {
822 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
826 * Make sure we don't allow writing to an append-only
829 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
833 * Make sure there are no mandatory locks on the file.
835 if (locks_verify_locked(inode))
838 vm_flags |= VM_SHARED | VM_MAYSHARE;
839 if (!(file->f_mode & FMODE_WRITE))
840 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
844 if (!(file->f_mode & FMODE_READ))
852 switch (flags & MAP_TYPE) {
854 vm_flags |= VM_SHARED | VM_MAYSHARE;
858 * Set pgoff according to addr for anon_vma.
860 pgoff = addr >> PAGE_SHIFT;
867 error = security_file_mmap(file, prot, flags);
874 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
875 if (vma && vma->vm_start < addr + len) {
876 if (do_munmap(mm, addr, len))
881 /* Check against address space limit. */
882 if ((mm->total_vm << PAGE_SHIFT) + len
883 > current->rlim[RLIMIT_AS].rlim_cur)
886 /* check context space, maybe only Private writable mapping? */
887 if (!vx_vmpages_avail(mm, len >> PAGE_SHIFT))
890 if (accountable && (!(flags & MAP_NORESERVE) ||
891 sysctl_overcommit_memory > 1)) {
892 if (vm_flags & VM_SHARED) {
893 /* Check memory availability in shmem_file_setup? */
894 vm_flags |= VM_ACCOUNT;
895 } else if (vm_flags & VM_WRITE) {
897 * Private writable mapping: check memory availability
899 charged = len >> PAGE_SHIFT;
900 if (security_vm_enough_memory(charged))
902 vm_flags |= VM_ACCOUNT;
907 * Can we just expand an old private anonymous mapping?
908 * The VM_SHARED test is necessary because shmem_zero_setup
909 * will create the file object for a shared anonymous map below.
911 if (!file && !(vm_flags & VM_SHARED) &&
912 vma_merge(mm, prev, addr, addr + len, vm_flags,
913 NULL, NULL, pgoff, NULL))
917 * Determine the object being mapped and call the appropriate
918 * specific mapper. the address has already been validated, but
919 * not unmapped, but the maps are removed from the list.
921 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
926 memset(vma, 0, sizeof(*vma));
929 vma->vm_start = addr;
930 vma->vm_end = addr + len;
931 vma->vm_flags = vm_flags;
932 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
933 vma->vm_pgoff = pgoff;
937 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
939 if (vm_flags & VM_DENYWRITE) {
940 error = deny_write_access(file);
947 error = file->f_op->mmap(file, vma);
949 goto unmap_and_free_vma;
950 } else if (vm_flags & VM_SHARED) {
951 error = shmem_zero_setup(vma);
956 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
957 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
958 * that memory reservation must be checked; but that reservation
959 * belongs to shared memory object, not to vma: so now clear it.
961 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
962 vma->vm_flags &= ~VM_ACCOUNT;
964 /* Can addr have changed??
966 * Answer: Yes, several device drivers can do it in their
967 * f_op->mmap method. -DaveM
969 addr = vma->vm_start;
971 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
972 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
973 vma_link(mm, vma, prev, rb_link, rb_parent);
975 atomic_inc(&inode->i_writecount);
979 atomic_inc(&inode->i_writecount);
982 mpol_free(vma_policy(vma));
983 kmem_cache_free(vm_area_cachep, vma);
986 // mm->total_vm += len >> PAGE_SHIFT;
987 vx_vmpages_add(mm, len >> PAGE_SHIFT);
988 if (vm_flags & VM_LOCKED) {
989 // mm->locked_vm += len >> PAGE_SHIFT;
990 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
991 make_pages_present(addr, addr + len);
993 if (flags & MAP_POPULATE) {
994 up_write(&mm->mmap_sem);
995 sys_remap_file_pages(addr, len, 0,
996 pgoff, flags & MAP_NONBLOCK);
997 down_write(&mm->mmap_sem);
1003 atomic_inc(&inode->i_writecount);
1004 vma->vm_file = NULL;
1007 /* Undo any partial mapping done by a device driver. */
1008 zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1010 kmem_cache_free(vm_area_cachep, vma);
1013 vm_unacct_memory(charged);
1017 EXPORT_SYMBOL(do_mmap_pgoff);
1019 /* Get an address range which is currently unmapped.
1020 * For shmat() with addr=0.
1022 * Ugly calling convention alert:
1023 * Return value with the low bits set means error value,
1025 * if (ret & ~PAGE_MASK)
1028 * This function "knows" that -ENOMEM has the bits set.
1030 #ifndef HAVE_ARCH_UNMAPPED_AREA
1031 static inline unsigned long
1032 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1033 unsigned long len, unsigned long pgoff, unsigned long flags, unsigned long exec)
1035 struct mm_struct *mm = current->mm;
1036 struct vm_area_struct *vma;
1037 unsigned long start_addr;
1039 if (len > TASK_SIZE)
1043 addr = PAGE_ALIGN(addr);
1044 vma = find_vma(mm, addr);
1045 if (TASK_SIZE - len >= addr &&
1046 (!vma || addr + len <= vma->vm_start))
1049 start_addr = addr = mm->free_area_cache;
1052 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1053 /* At this point: (!vma || addr < vma->vm_end). */
1054 if (TASK_SIZE - len < addr) {
1056 * Start a new search - just in case we missed
1059 if (start_addr != TASK_UNMAPPED_BASE) {
1060 start_addr = addr = TASK_UNMAPPED_BASE;
1065 if (!vma || addr + len <= vma->vm_start) {
1067 * Remember the place where we stopped the search:
1069 mm->free_area_cache = addr + len;
1076 extern unsigned long
1077 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
1078 unsigned long, unsigned long, unsigned long);
1082 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1083 unsigned long pgoff, unsigned long flags, unsigned long exec)
1085 if (flags & MAP_FIXED) {
1088 if (addr > TASK_SIZE - len)
1090 if (addr & ~PAGE_MASK)
1092 if (file && is_file_hugepages(file)) {
1094 * Check if the given range is hugepage aligned, and
1095 * can be made suitable for hugepages.
1097 ret = prepare_hugepage_range(addr, len);
1100 * Ensure that a normal request is not falling in a
1101 * reserved hugepage range. For some archs like IA-64,
1102 * there is a separate region for hugepages.
1104 ret = is_hugepage_only_range(addr, len);
1111 if (file && file->f_op && file->f_op->get_unmapped_area)
1112 return file->f_op->get_unmapped_area(file, addr, len,
1115 return arch_get_unmapped_area(file, addr, len, pgoff, flags, exec);
1118 EXPORT_SYMBOL(get_unmapped_area);
1120 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1121 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1123 struct vm_area_struct *vma = NULL;
1126 /* Check the cache first. */
1127 /* (Cache hit rate is typically around 35%.) */
1128 vma = mm->mmap_cache;
1129 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1130 struct rb_node * rb_node;
1132 rb_node = mm->mm_rb.rb_node;
1136 struct vm_area_struct * vma_tmp;
1138 vma_tmp = rb_entry(rb_node,
1139 struct vm_area_struct, vm_rb);
1141 if (vma_tmp->vm_end > addr) {
1143 if (vma_tmp->vm_start <= addr)
1145 rb_node = rb_node->rb_left;
1147 rb_node = rb_node->rb_right;
1150 mm->mmap_cache = vma;
1156 EXPORT_SYMBOL(find_vma);
1158 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1159 struct vm_area_struct *
1160 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1161 struct vm_area_struct **pprev)
1163 struct vm_area_struct *vma = NULL, *prev = NULL;
1164 struct rb_node * rb_node;
1168 /* Guard against addr being lower than the first VMA */
1171 /* Go through the RB tree quickly. */
1172 rb_node = mm->mm_rb.rb_node;
1175 struct vm_area_struct *vma_tmp;
1176 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1178 if (addr < vma_tmp->vm_end) {
1179 rb_node = rb_node->rb_left;
1182 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1184 rb_node = rb_node->rb_right;
1190 return prev ? prev->vm_next : vma;
1194 static int over_stack_limit(unsigned long sz)
1196 if (sz < EXEC_STACK_BIAS)
1198 return (sz - EXEC_STACK_BIAS) > current->rlim[RLIMIT_STACK].rlim_cur;
1201 #ifdef CONFIG_STACK_GROWSUP
1203 * vma is the first one with address > vma->vm_end. Have to extend vma.
1205 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1209 if (!(vma->vm_flags & VM_GROWSUP))
1213 * We must make sure the anon_vma is allocated
1214 * so that the anon_vma locking is not a noop.
1216 if (unlikely(anon_vma_prepare(vma)))
1221 * vma->vm_start/vm_end cannot change under us because the caller
1222 * is required to hold the mmap_sem in read mode. We need the
1223 * anon_vma lock to serialize against concurrent expand_stacks.
1225 address += 4 + PAGE_SIZE - 1;
1226 address &= PAGE_MASK;
1227 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1230 if (security_vm_enough_memory(grow) ||
1231 !vx_vmpages_avail(vma->vm_mm, grow)) {
1232 anon_vma_unlock(vma);
1236 if (over_stack_limit(address - vma->vm_start) ||
1237 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1238 current->rlim[RLIMIT_AS].rlim_cur) {
1239 anon_vma_unlock(vma);
1240 vm_unacct_memory(grow);
1244 vma->vm_end = address;
1245 // vma->vm_mm->total_vm += grow;
1246 vx_vmpages_add(vma->vm_mm, grow);
1247 if (vma->vm_flags & VM_LOCKED)
1248 // vma->vm_mm->locked_vm += grow;
1249 vx_vmlocked_add(vma->vm_mm, grow);
1250 anon_vma_unlock(vma);
1254 struct vm_area_struct *
1255 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1257 struct vm_area_struct *vma, *prev;
1260 vma = find_vma_prev(mm, addr, &prev);
1261 if (vma && (vma->vm_start <= addr))
1263 if (!prev || expand_stack(prev, addr))
1265 if (prev->vm_flags & VM_LOCKED) {
1266 make_pages_present(addr, prev->vm_end);
1272 * vma is the first one with address < vma->vm_start. Have to extend vma.
1274 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1279 * We must make sure the anon_vma is allocated
1280 * so that the anon_vma locking is not a noop.
1282 if (unlikely(anon_vma_prepare(vma)))
1287 * vma->vm_start/vm_end cannot change under us because the caller
1288 * is required to hold the mmap_sem in read mode. We need the
1289 * anon_vma lock to serialize against concurrent expand_stacks.
1291 address &= PAGE_MASK;
1292 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1295 if (security_vm_enough_memory(grow) ||
1296 !vx_vmpages_avail(vma->vm_mm, grow)) {
1297 anon_vma_unlock(vma);
1301 if (over_stack_limit(vma->vm_end - address) ||
1302 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1303 current->rlim[RLIMIT_AS].rlim_cur) {
1304 anon_vma_unlock(vma);
1305 vm_unacct_memory(grow);
1309 vma->vm_start = address;
1310 vma->vm_pgoff -= grow;
1311 // vma->vm_mm->total_vm += grow;
1312 vx_vmpages_add(vma->vm_mm, grow);
1313 if (vma->vm_flags & VM_LOCKED)
1314 // vma->vm_mm->locked_vm += grow;
1315 vx_vmlocked_add(vma->vm_mm, grow);
1316 anon_vma_unlock(vma);
1320 struct vm_area_struct *
1321 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1323 struct vm_area_struct * vma;
1324 unsigned long start;
1327 vma = find_vma(mm,addr);
1330 if (vma->vm_start <= addr)
1332 if (!(vma->vm_flags & VM_GROWSDOWN))
1334 start = vma->vm_start;
1335 if (expand_stack(vma, addr))
1337 if (vma->vm_flags & VM_LOCKED) {
1338 make_pages_present(addr, start);
1345 * Try to free as many page directory entries as we can,
1346 * without having to work very hard at actually scanning
1347 * the page tables themselves.
1349 * Right now we try to free page tables if we have a nice
1350 * PGDIR-aligned area that got free'd up. We could be more
1351 * granular if we want to, but this is fast and simple,
1352 * and covers the bad cases.
1354 * "prev", if it exists, points to a vma before the one
1355 * we just free'd - but there's no telling how much before.
1357 static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev,
1358 unsigned long start, unsigned long end)
1360 unsigned long first = start & PGDIR_MASK;
1361 unsigned long last = end + PGDIR_SIZE - 1;
1362 unsigned long start_index, end_index;
1363 struct mm_struct *mm = tlb->mm;
1369 if (prev->vm_end > start) {
1370 if (last > prev->vm_start)
1371 last = prev->vm_start;
1376 struct vm_area_struct *next = prev->vm_next;
1379 if (next->vm_start < start) {
1383 if (last > next->vm_start)
1384 last = next->vm_start;
1386 if (prev->vm_end > first)
1387 first = prev->vm_end + PGDIR_SIZE - 1;
1391 if (last < first) /* for arches with discontiguous pgd indices */
1394 * If the PGD bits are not consecutive in the virtual address, the
1395 * old method of shifting the VA >> by PGDIR_SHIFT doesn't work.
1397 start_index = pgd_index(first);
1398 if (start_index < FIRST_USER_PGD_NR)
1399 start_index = FIRST_USER_PGD_NR;
1400 end_index = pgd_index(last);
1401 if (end_index > start_index) {
1402 clear_page_tables(tlb, start_index, end_index - start_index);
1403 flush_tlb_pgtables(mm, first & PGDIR_MASK, last & PGDIR_MASK);
1407 /* Normal function to fix up a mapping
1408 * This function is the default for when an area has no specific
1409 * function. This may be used as part of a more specific routine.
1411 * By the time this function is called, the area struct has been
1412 * removed from the process mapping list.
1414 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1416 size_t len = area->vm_end - area->vm_start;
1417 unsigned long old_end = area->vm_end;
1419 // area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1420 vx_vmpages_sub(area->vm_mm, len >> PAGE_SHIFT);
1422 if (area->vm_flags & VM_LOCKED)
1423 // area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1424 vx_vmlocked_sub(area->vm_mm, len >> PAGE_SHIFT);
1426 * Is this a new hole at the lowest possible address?
1428 if (area->vm_start >= TASK_UNMAPPED_BASE &&
1429 area->vm_start < area->vm_mm->free_area_cache)
1430 area->vm_mm->free_area_cache = area->vm_start;
1432 * Is this a new hole at the highest possible address?
1434 if (area->vm_start > area->vm_mm->non_executable_cache)
1435 area->vm_mm->non_executable_cache = area->vm_start;
1436 remove_vm_struct(area);
1437 if (unlikely(area->vm_flags & VM_EXEC))
1438 arch_remove_exec_range(mm, old_end);
1442 * Update the VMA and inode share lists.
1444 * Ok - we have the memory areas we should free on the 'free' list,
1445 * so release them, and do the vma updates.
1447 static void unmap_vma_list(struct mm_struct *mm,
1448 struct vm_area_struct *mpnt)
1451 struct vm_area_struct *next = mpnt->vm_next;
1452 unmap_vma(mm, mpnt);
1454 } while (mpnt != NULL);
1459 * Get rid of page table information in the indicated region.
1461 * Called with the page table lock held.
1463 static void unmap_region(struct mm_struct *mm,
1464 struct vm_area_struct *vma,
1465 struct vm_area_struct *prev,
1466 unsigned long start,
1469 struct mmu_gather *tlb;
1470 unsigned long nr_accounted = 0;
1473 tlb = tlb_gather_mmu(mm, 0);
1474 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1475 vm_unacct_memory(nr_accounted);
1477 if (is_hugepage_only_range(start, end - start))
1478 hugetlb_free_pgtables(tlb, prev, start, end);
1480 free_pgtables(tlb, prev, start, end);
1481 tlb_finish_mmu(tlb, start, end);
1485 * Create a list of vma's touched by the unmap, removing them from the mm's
1486 * vma list as we go..
1489 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1490 struct vm_area_struct *prev, unsigned long end)
1492 struct vm_area_struct **insertion_point;
1493 struct vm_area_struct *tail_vma = NULL;
1495 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1497 rb_erase(&vma->vm_rb, &mm->mm_rb);
1501 } while (vma && vma->vm_start < end);
1502 *insertion_point = vma;
1503 tail_vma->vm_next = NULL;
1504 mm->mmap_cache = NULL; /* Kill the cache. */
1508 * Split a vma into two pieces at address 'addr', a new vma is allocated
1509 * either for the first part or the the tail.
1511 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1512 unsigned long addr, int new_below)
1514 struct mempolicy *pol;
1515 struct vm_area_struct *new;
1517 if (mm->map_count >= sysctl_max_map_count)
1520 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1524 /* most fields are the same, copy all, and then fixup */
1526 vma_prio_tree_init(new);
1531 new->vm_start = addr;
1532 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1535 pol = mpol_copy(vma_policy(vma));
1537 kmem_cache_free(vm_area_cachep, new);
1538 return PTR_ERR(pol);
1540 vma_set_policy(new, pol);
1543 get_file(new->vm_file);
1545 if (new->vm_ops && new->vm_ops->open)
1546 new->vm_ops->open(new);
1549 unsigned long old_end = vma->vm_end;
1551 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1552 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1553 if (vma->vm_flags & VM_EXEC)
1554 arch_remove_exec_range(mm, old_end);
1556 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1561 /* Munmap is split into 2 main parts -- this part which finds
1562 * what needs doing, and the areas themselves, which do the
1563 * work. This now handles partial unmappings.
1564 * Jeremy Fitzhardinge <jeremy@goop.org>
1566 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1569 struct vm_area_struct *mpnt, *prev, *last;
1571 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1574 if ((len = PAGE_ALIGN(len)) == 0)
1577 /* Find the first overlapping VMA */
1578 mpnt = find_vma_prev(mm, start, &prev);
1581 /* we have start < mpnt->vm_end */
1583 if (is_vm_hugetlb_page(mpnt)) {
1584 int ret = is_aligned_hugepage_range(start, len);
1590 /* if it doesn't overlap, we have nothing.. */
1592 if (mpnt->vm_start >= end)
1595 /* Something will probably happen, so notify. */
1596 if (mpnt->vm_file && (mpnt->vm_flags & VM_EXEC))
1597 profile_exec_unmap(mm);
1600 * If we need to split any vma, do it now to save pain later.
1602 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1603 * unmapped vm_area_struct will remain in use: so lower split_vma
1604 * places tmp vma above, and higher split_vma places tmp vma below.
1606 if (start > mpnt->vm_start) {
1607 if (split_vma(mm, mpnt, start, 0))
1612 /* Does it split the last one? */
1613 last = find_vma(mm, end);
1614 if (last && end > last->vm_start) {
1615 if (split_vma(mm, last, end, 1))
1618 mpnt = prev? prev->vm_next: mm->mmap;
1621 * Remove the vma's, and unmap the actual pages
1623 detach_vmas_to_be_unmapped(mm, mpnt, prev, end);
1624 spin_lock(&mm->page_table_lock);
1625 unmap_region(mm, mpnt, prev, start, end);
1626 spin_unlock(&mm->page_table_lock);
1628 /* Fix up all other VM information */
1629 unmap_vma_list(mm, mpnt);
1634 EXPORT_SYMBOL(do_munmap);
1636 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1639 struct mm_struct *mm = current->mm;
1641 down_write(&mm->mmap_sem);
1642 ret = do_munmap(mm, addr, len);
1643 up_write(&mm->mmap_sem);
1648 * this is really a simplified "do_mmap". it only handles
1649 * anonymous maps. eventually we may be able to do some
1650 * brk-specific accounting here.
1652 unsigned long do_brk(unsigned long addr, unsigned long len)
1654 struct mm_struct * mm = current->mm;
1655 struct vm_area_struct * vma, * prev;
1656 unsigned long flags;
1657 struct rb_node ** rb_link, * rb_parent;
1658 pgoff_t pgoff = addr >> PAGE_SHIFT;
1660 len = PAGE_ALIGN(len);
1664 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1670 if (mm->def_flags & VM_LOCKED) {
1671 unsigned long locked, lock_limit;
1672 locked = mm->locked_vm << PAGE_SHIFT;
1673 lock_limit = current->rlim[RLIMIT_MEMLOCK].rlim_cur;
1675 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1677 /* vserver checks ? */
1681 * Clear old maps. this also does some error checking for us
1684 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1685 if (vma && vma->vm_start < addr + len) {
1686 if (do_munmap(mm, addr, len))
1691 /* Check against address space limits *after* clearing old maps... */
1692 if ((mm->total_vm << PAGE_SHIFT) + len
1693 > current->rlim[RLIMIT_AS].rlim_cur)
1696 if (mm->map_count > sysctl_max_map_count)
1699 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
1700 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
1703 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1705 /* Can we just expand an old private anonymous mapping? */
1706 if (vma_merge(mm, prev, addr, addr + len, flags,
1707 NULL, NULL, pgoff, NULL))
1711 * create a vma struct for an anonymous mapping
1713 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1715 vm_unacct_memory(len >> PAGE_SHIFT);
1718 memset(vma, 0, sizeof(*vma));
1721 vma->vm_start = addr;
1722 vma->vm_end = addr + len;
1723 vma->vm_pgoff = pgoff;
1724 vma->vm_flags = flags;
1725 vma->vm_page_prot = protection_map[flags & 0x0f];
1726 vma_link(mm, vma, prev, rb_link, rb_parent);
1728 // mm->total_vm += len >> PAGE_SHIFT;
1729 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1730 if (flags & VM_LOCKED) {
1731 // mm->locked_vm += len >> PAGE_SHIFT;
1732 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1733 make_pages_present(addr, addr + len);
1738 EXPORT_SYMBOL(do_brk);
1740 /* Release all mmaps. */
1741 void exit_mmap(struct mm_struct *mm)
1743 struct mmu_gather *tlb;
1744 struct vm_area_struct *vma;
1745 unsigned long nr_accounted = 0;
1747 profile_exit_mmap(mm);
1751 spin_lock(&mm->page_table_lock);
1753 tlb = tlb_gather_mmu(mm, 1);
1755 /* Use ~0UL here to ensure all VMAs in the mm are unmapped */
1756 mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0,
1757 ~0UL, &nr_accounted, NULL);
1758 vm_unacct_memory(nr_accounted);
1759 BUG_ON(mm->map_count); /* This is just debugging */
1760 clear_page_tables(tlb, FIRST_USER_PGD_NR, USER_PTRS_PER_PGD);
1761 tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm));
1764 mm->mmap = mm->mmap_cache = NULL;
1765 mm->mm_rb = RB_ROOT;
1767 vx_rsspages_sub(mm, mm->rss);
1768 // mm->total_vm = 0;
1769 vx_vmpages_sub(mm, mm->total_vm);
1770 // mm->locked_vm = 0;
1771 vx_vmlocked_sub(mm, mm->locked_vm);
1772 arch_flush_exec_range(mm);
1774 spin_unlock(&mm->page_table_lock);
1777 * Walk the list again, actually closing and freeing it
1778 * without holding any MM locks.
1781 struct vm_area_struct *next = vma->vm_next;
1782 remove_vm_struct(vma);
1787 /* Insert vm structure into process list sorted by address
1788 * and into the inode's i_mmap tree. If vm_file is non-NULL
1789 * then i_mmap_lock is taken here.
1791 void insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1793 struct vm_area_struct * __vma, * prev;
1794 struct rb_node ** rb_link, * rb_parent;
1797 * The vm_pgoff of a purely anonymous vma should be irrelevant
1798 * until its first write fault, when page's anon_vma and index
1799 * are set. But now set the vm_pgoff it will almost certainly
1800 * end up with (unless mremap moves it elsewhere before that
1801 * first wfault), so /proc/pid/maps tells a consistent story.
1803 * By setting it to reflect the virtual start address of the
1804 * vma, merges and splits can happen in a seamless way, just
1805 * using the existing file pgoff checks and manipulations.
1806 * Similarly in do_mmap_pgoff and in do_brk.
1808 if (!vma->vm_file) {
1809 BUG_ON(vma->anon_vma);
1810 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1812 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1813 if (__vma && __vma->vm_start < vma->vm_end)
1815 vma_link(mm, vma, prev, rb_link, rb_parent);
1819 * Copy the vma structure to a new location in the same mm,
1820 * prior to moving page table entries, to effect an mremap move.
1822 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
1823 unsigned long addr, unsigned long len, pgoff_t pgoff)
1825 struct vm_area_struct *vma = *vmap;
1826 unsigned long vma_start = vma->vm_start;
1827 struct mm_struct *mm = vma->vm_mm;
1828 struct vm_area_struct *new_vma, *prev;
1829 struct rb_node **rb_link, *rb_parent;
1830 struct mempolicy *pol;
1833 * If anonymous vma has not yet been faulted, update new pgoff
1834 * to match new location, to increase its chance of merging.
1836 if (!vma->vm_file && !vma->anon_vma)
1837 pgoff = addr >> PAGE_SHIFT;
1839 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1840 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
1841 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
1844 * Source vma may have been merged into new_vma
1846 if (vma_start >= new_vma->vm_start &&
1847 vma_start < new_vma->vm_end)
1850 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1853 vma_prio_tree_init(new_vma);
1854 pol = mpol_copy(vma_policy(vma));
1856 kmem_cache_free(vm_area_cachep, new_vma);
1859 vma_set_policy(new_vma, pol);
1860 new_vma->vm_start = addr;
1861 new_vma->vm_end = addr + len;
1862 new_vma->vm_pgoff = pgoff;
1863 if (new_vma->vm_file)
1864 get_file(new_vma->vm_file);
1865 if (new_vma->vm_ops && new_vma->vm_ops->open)
1866 new_vma->vm_ops->open(new_vma);
1867 vma_link(mm, new_vma, prev, rb_link, rb_parent);