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;
94 struct address_space *mapping = file->f_mapping;
95 spin_lock(&mapping->i_mmap_lock);
96 __remove_shared_vm_struct(vma, file, mapping);
97 spin_unlock(&mapping->i_mmap_lock);
99 if (vma->vm_ops && vma->vm_ops->close)
100 vma->vm_ops->close(vma);
103 anon_vma_unlink(vma);
104 mpol_free(vma_policy(vma));
105 kmem_cache_free(vm_area_cachep, vma);
109 * sys_brk() for the most part doesn't need the global kernel
110 * lock, except when an application is doing something nasty
111 * like trying to un-brk an area that has already been mapped
112 * to a regular file. in this case, the unmapping will need
113 * to invoke file system routines that need the global lock.
115 asmlinkage unsigned long sys_brk(unsigned long brk)
117 unsigned long rlim, retval;
118 unsigned long newbrk, oldbrk;
119 struct mm_struct *mm = current->mm;
121 down_write(&mm->mmap_sem);
123 if (brk < mm->end_code)
125 newbrk = PAGE_ALIGN(brk);
126 oldbrk = PAGE_ALIGN(mm->brk);
127 if (oldbrk == newbrk)
130 /* Always allow shrinking brk. */
131 if (brk <= mm->brk) {
132 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
137 /* Check against rlimit.. */
138 rlim = current->rlim[RLIMIT_DATA].rlim_cur;
139 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
142 /* Check against existing mmap mappings. */
143 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
146 /* Ok, looks good - let it rip. */
147 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
153 up_write(&mm->mmap_sem);
158 static int browse_rb(struct rb_root *root)
161 struct rb_node *nd, *pn = NULL;
162 unsigned long prev = 0, pend = 0;
164 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
165 struct vm_area_struct *vma;
166 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
167 if (vma->vm_start < prev)
168 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
169 if (vma->vm_start < pend)
170 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
171 if (vma->vm_start > vma->vm_end)
172 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
177 for (nd = pn; nd; nd = rb_prev(nd)) {
181 printk("backwards %d, forwards %d\n", j, i), i = 0;
185 void validate_mm(struct mm_struct *mm)
189 struct vm_area_struct *tmp = mm->mmap;
194 if (i != mm->map_count)
195 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
196 i = browse_rb(&mm->mm_rb);
197 if (i != mm->map_count)
198 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
203 #define validate_mm(mm) do { } while (0)
206 static struct vm_area_struct *
207 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
208 struct vm_area_struct **pprev, struct rb_node ***rb_link,
209 struct rb_node ** rb_parent)
211 struct vm_area_struct * vma;
212 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
214 __rb_link = &mm->mm_rb.rb_node;
215 rb_prev = __rb_parent = NULL;
219 struct vm_area_struct *vma_tmp;
221 __rb_parent = *__rb_link;
222 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
224 if (vma_tmp->vm_end > addr) {
226 if (vma_tmp->vm_start <= addr)
228 __rb_link = &__rb_parent->rb_left;
230 rb_prev = __rb_parent;
231 __rb_link = &__rb_parent->rb_right;
237 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
238 *rb_link = __rb_link;
239 *rb_parent = __rb_parent;
244 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
245 struct vm_area_struct *prev, struct rb_node *rb_parent)
247 if (vma->vm_flags & VM_EXEC)
248 arch_add_exec_range(mm, vma->vm_end);
250 vma->vm_next = prev->vm_next;
255 vma->vm_next = rb_entry(rb_parent,
256 struct vm_area_struct, vm_rb);
262 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
263 struct rb_node **rb_link, struct rb_node *rb_parent)
265 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
266 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
269 static inline void __vma_link_file(struct vm_area_struct *vma)
275 struct address_space *mapping = file->f_mapping;
277 if (vma->vm_flags & VM_DENYWRITE)
278 atomic_dec(&file->f_dentry->d_inode->i_writecount);
279 if (vma->vm_flags & VM_SHARED)
280 mapping->i_mmap_writable++;
282 flush_dcache_mmap_lock(mapping);
283 if (unlikely(vma->vm_flags & VM_NONLINEAR))
284 list_add_tail(&vma->shared.vm_set.list,
285 &mapping->i_mmap_nonlinear);
287 vma_prio_tree_insert(vma, &mapping->i_mmap);
288 flush_dcache_mmap_unlock(mapping);
293 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
294 struct vm_area_struct *prev, struct rb_node **rb_link,
295 struct rb_node *rb_parent)
297 __vma_link_list(mm, vma, prev, rb_parent);
298 __vma_link_rb(mm, vma, rb_link, rb_parent);
299 __anon_vma_link(vma);
302 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
303 struct vm_area_struct *prev, struct rb_node **rb_link,
304 struct rb_node *rb_parent)
306 struct address_space *mapping = NULL;
309 mapping = vma->vm_file->f_mapping;
312 spin_lock(&mapping->i_mmap_lock);
315 __vma_link(mm, vma, prev, rb_link, rb_parent);
316 __vma_link_file(vma);
318 anon_vma_unlock(vma);
320 spin_unlock(&mapping->i_mmap_lock);
322 mark_mm_hugetlb(mm, vma);
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 address_space *mapping = NULL;
370 struct prio_tree_root *root = NULL;
371 struct file *file = vma->vm_file;
372 struct anon_vma *anon_vma = NULL;
373 long adjust_next = 0;
376 if (next && !insert) {
377 if (end >= next->vm_end) {
379 * vma expands, overlapping all the next, and
380 * perhaps the one after too (mprotect case 6).
382 again: remove_next = 1 + (end > next->vm_end);
384 anon_vma = next->anon_vma;
385 } else if (end > next->vm_start) {
387 * vma expands, overlapping part of the next:
388 * mprotect case 5 shifting the boundary up.
390 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
391 anon_vma = next->anon_vma;
392 } else if (end < vma->vm_end) {
394 * vma shrinks, and !insert tells it's not
395 * split_vma inserting another: so it must be
396 * mprotect case 4 shifting the boundary down.
398 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
399 anon_vma = next->anon_vma;
404 mapping = file->f_mapping;
405 if (!(vma->vm_flags & VM_NONLINEAR))
406 root = &mapping->i_mmap;
407 spin_lock(&mapping->i_mmap_lock);
410 * Put into prio_tree now, so instantiated pages
411 * are visible to arm/parisc __flush_dcache_page
412 * throughout; but we cannot insert into address
413 * space until vma start or end is updated.
415 __vma_link_file(insert);
420 * When changing only vma->vm_end, we don't really need
421 * anon_vma lock: but is that case worth optimizing out?
424 anon_vma = vma->anon_vma;
426 spin_lock(&anon_vma->lock);
429 flush_dcache_mmap_lock(mapping);
430 vma_prio_tree_remove(vma, root);
432 vma_prio_tree_remove(next, root);
435 vma->vm_start = start;
437 vma->vm_pgoff = pgoff;
439 next->vm_start += adjust_next << PAGE_SHIFT;
440 next->vm_pgoff += adjust_next;
445 vma_prio_tree_init(next);
446 vma_prio_tree_insert(next, root);
448 vma_prio_tree_init(vma);
449 vma_prio_tree_insert(vma, root);
450 flush_dcache_mmap_unlock(mapping);
455 * vma_merge has merged next into vma, and needs
456 * us to remove next before dropping the locks.
458 __vma_unlink(mm, next, vma);
460 __remove_shared_vm_struct(next, file, mapping);
462 __anon_vma_merge(vma, next);
465 * split_vma has split insert from vma, and needs
466 * us to insert it before dropping the locks
467 * (it may either follow vma or precede it).
469 __insert_vm_struct(mm, insert);
473 spin_unlock(&anon_vma->lock);
475 spin_unlock(&mapping->i_mmap_lock);
481 mpol_free(vma_policy(next));
482 kmem_cache_free(vm_area_cachep, next);
484 * In mprotect's case 6 (see comments on vma_merge),
485 * we must remove another next too. It would clutter
486 * up the code too much to do both in one go.
488 if (remove_next == 2) {
498 * If the vma has a ->close operation then the driver probably needs to release
499 * per-vma resources, so we don't attempt to merge those.
501 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
503 static inline int is_mergeable_vma(struct vm_area_struct *vma,
504 struct file *file, unsigned long vm_flags)
506 if (vma->vm_flags != vm_flags)
508 if (vma->vm_file != file)
510 if (vma->vm_ops && vma->vm_ops->close)
515 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
516 struct anon_vma *anon_vma2)
518 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
522 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
523 * in front of (at a lower virtual address and file offset than) the vma.
525 * We cannot merge two vmas if they have differently assigned (non-NULL)
526 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
528 * We don't check here for the merged mmap wrapping around the end of pagecache
529 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
530 * wrap, nor mmaps which cover the final page at index -1UL.
533 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
534 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
536 if (is_mergeable_vma(vma, file, vm_flags) &&
537 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
538 if (vma->vm_pgoff == vm_pgoff)
545 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
546 * beyond (at a higher virtual address and file offset than) the vma.
548 * We cannot merge two vmas if they have differently assigned (non-NULL)
549 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
552 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
553 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
555 if (is_mergeable_vma(vma, file, vm_flags) &&
556 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
558 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
559 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
566 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
567 * whether that can be merged with its predecessor or its successor.
568 * Or both (it neatly fills a hole).
570 * In most cases - when called for mmap, brk or mremap - [addr,end) is
571 * certain not to be mapped by the time vma_merge is called; but when
572 * called for mprotect, it is certain to be already mapped (either at
573 * an offset within prev, or at the start of next), and the flags of
574 * this area are about to be changed to vm_flags - and the no-change
575 * case has already been eliminated.
577 * The following mprotect cases have to be considered, where AAAA is
578 * the area passed down from mprotect_fixup, never extending beyond one
579 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
581 * AAAA AAAA AAAA AAAA
582 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
583 * cannot merge might become might become might become
584 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
585 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
586 * mremap move: PPPPNNNNNNNN 8
588 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
589 * might become case 1 below case 2 below case 3 below
591 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
592 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
594 struct vm_area_struct *vma_merge(struct mm_struct *mm,
595 struct vm_area_struct *prev, unsigned long addr,
596 unsigned long end, unsigned long vm_flags,
597 struct anon_vma *anon_vma, struct file *file,
598 pgoff_t pgoff, struct mempolicy *policy)
600 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
601 struct vm_area_struct *area, *next;
604 * We later require that vma->vm_flags == vm_flags,
605 * so this tests vma->vm_flags & VM_SPECIAL, too.
607 if (vm_flags & VM_SPECIAL)
611 next = prev->vm_next;
615 if (next && next->vm_end == end) /* cases 6, 7, 8 */
616 next = next->vm_next;
619 * Can it merge with the predecessor?
621 if (prev && prev->vm_end == addr &&
622 mpol_equal(vma_policy(prev), policy) &&
623 can_vma_merge_after(prev, vm_flags,
624 anon_vma, file, pgoff)) {
626 * OK, it can. Can we now merge in the successor as well?
628 if (next && end == next->vm_start &&
629 mpol_equal(policy, vma_policy(next)) &&
630 can_vma_merge_before(next, vm_flags,
631 anon_vma, file, pgoff+pglen) &&
632 is_mergeable_anon_vma(prev->anon_vma,
635 vma_adjust(prev, prev->vm_start,
636 next->vm_end, prev->vm_pgoff, NULL);
637 } else /* cases 2, 5, 7 */
638 vma_adjust(prev, prev->vm_start,
639 end, prev->vm_pgoff, NULL);
640 if (prev->vm_flags & VM_EXEC)
641 arch_add_exec_range(mm, prev->vm_end);
646 * Can this new request be merged in front of next?
648 if (next && end == next->vm_start &&
649 mpol_equal(policy, vma_policy(next)) &&
650 can_vma_merge_before(next, vm_flags,
651 anon_vma, file, pgoff+pglen)) {
652 if (prev && addr < prev->vm_end) /* case 4 */
653 vma_adjust(prev, prev->vm_start,
654 addr, prev->vm_pgoff, NULL);
655 else /* cases 3, 8 */
656 vma_adjust(area, addr, next->vm_end,
657 next->vm_pgoff - pglen, NULL);
665 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
666 * neighbouring vmas for a suitable anon_vma, before it goes off
667 * to allocate a new anon_vma. It checks because a repetitive
668 * sequence of mprotects and faults may otherwise lead to distinct
669 * anon_vmas being allocated, preventing vma merge in subsequent
672 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
674 struct vm_area_struct *near;
675 unsigned long vm_flags;
682 * Since only mprotect tries to remerge vmas, match flags
683 * which might be mprotected into each other later on.
684 * Neither mlock nor madvise tries to remerge at present,
685 * so leave their flags as obstructing a merge.
687 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
688 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
690 if (near->anon_vma && vma->vm_end == near->vm_start &&
691 mpol_equal(vma_policy(vma), vma_policy(near)) &&
692 can_vma_merge_before(near, vm_flags,
693 NULL, vma->vm_file, vma->vm_pgoff +
694 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
695 return near->anon_vma;
698 * It is potentially slow to have to call find_vma_prev here.
699 * But it's only on the first write fault on the vma, not
700 * every time, and we could devise a way to avoid it later
701 * (e.g. stash info in next's anon_vma_node when assigning
702 * an anon_vma, or when trying vma_merge). Another time.
704 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
709 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
710 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
712 if (near->anon_vma && near->vm_end == vma->vm_start &&
713 mpol_equal(vma_policy(near), vma_policy(vma)) &&
714 can_vma_merge_after(near, vm_flags,
715 NULL, vma->vm_file, vma->vm_pgoff))
716 return near->anon_vma;
719 * There's no absolute need to look only at touching neighbours:
720 * we could search further afield for "compatible" anon_vmas.
721 * But it would probably just be a waste of time searching,
722 * or lead to too many vmas hanging off the same anon_vma.
723 * We're trying to allow mprotect remerging later on,
724 * not trying to minimize memory used for anon_vmas.
730 * The caller must hold down_write(current->mm->mmap_sem).
733 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
734 unsigned long len, unsigned long prot,
735 unsigned long flags, unsigned long pgoff)
737 struct mm_struct * mm = current->mm;
738 struct vm_area_struct * vma, * prev;
740 unsigned int vm_flags;
741 int correct_wcount = 0;
743 struct rb_node ** rb_link, * rb_parent;
745 unsigned long charged = 0;
748 if (is_file_hugepages(file))
751 if (!file->f_op || !file->f_op->mmap)
754 if ((prot & PROT_EXEC) &&
755 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
762 /* Careful about overflows.. */
763 len = PAGE_ALIGN(len);
764 if (!len || len > TASK_SIZE)
767 /* offset overflow? */
768 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
771 /* Too many mappings? */
772 if (mm->map_count > sysctl_max_map_count)
775 /* Obtain the address to map to. we verify (or select) it and ensure
776 * that it represents a valid section of the address space.
778 addr = get_unmapped_area(file, addr, len, pgoff, flags, prot & PROT_EXEC);
779 if (addr & ~PAGE_MASK)
782 /* Do simple checking here so the lower-level routines won't have
783 * to. we assume access permissions have been handled by the open
784 * of the memory object, so we don't do any here.
786 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
787 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
789 if (flags & MAP_LOCKED) {
790 if (!capable(CAP_IPC_LOCK))
792 vm_flags |= VM_LOCKED;
794 /* mlock MCL_FUTURE? */
795 if (vm_flags & VM_LOCKED) {
796 unsigned long locked = mm->locked_vm << PAGE_SHIFT;
798 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
802 inode = file ? file->f_dentry->d_inode : NULL;
805 switch (flags & MAP_TYPE) {
807 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
811 * Make sure we don't allow writing to an append-only
814 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
818 * Make sure there are no mandatory locks on the file.
820 if (locks_verify_locked(inode))
823 vm_flags |= VM_SHARED | VM_MAYSHARE;
824 if (!(file->f_mode & FMODE_WRITE))
825 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
829 if (!(file->f_mode & FMODE_READ))
837 switch (flags & MAP_TYPE) {
839 vm_flags |= VM_SHARED | VM_MAYSHARE;
843 * Set pgoff according to addr for anon_vma.
845 pgoff = addr >> PAGE_SHIFT;
852 error = security_file_mmap(file, prot, flags);
859 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
860 if (vma && vma->vm_start < addr + len) {
861 if (do_munmap(mm, addr, len))
866 /* Check against address space limit. */
867 if ((mm->total_vm << PAGE_SHIFT) + len
868 > current->rlim[RLIMIT_AS].rlim_cur)
871 if (accountable && (!(flags & MAP_NORESERVE) ||
872 sysctl_overcommit_memory > 1)) {
873 if (vm_flags & VM_SHARED) {
874 /* Check memory availability in shmem_file_setup? */
875 vm_flags |= VM_ACCOUNT;
876 } else if (vm_flags & VM_WRITE) {
878 * Private writable mapping: check memory availability
880 charged = len >> PAGE_SHIFT;
881 if (security_vm_enough_memory(charged))
883 vm_flags |= VM_ACCOUNT;
888 * Can we just expand an old private anonymous mapping?
889 * The VM_SHARED test is necessary because shmem_zero_setup
890 * will create the file object for a shared anonymous map below.
892 if (!file && !(vm_flags & VM_SHARED) &&
893 vma_merge(mm, prev, addr, addr + len, vm_flags,
894 NULL, NULL, pgoff, NULL))
898 * Determine the object being mapped and call the appropriate
899 * specific mapper. the address has already been validated, but
900 * not unmapped, but the maps are removed from the list.
902 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
907 memset(vma, 0, sizeof(*vma));
910 vma->vm_start = addr;
911 vma->vm_end = addr + len;
912 vma->vm_flags = vm_flags;
913 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
914 vma->vm_pgoff = pgoff;
918 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
920 if (vm_flags & VM_DENYWRITE) {
921 error = deny_write_access(file);
928 error = file->f_op->mmap(file, vma);
930 goto unmap_and_free_vma;
931 } else if (vm_flags & VM_SHARED) {
932 error = shmem_zero_setup(vma);
937 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
938 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
939 * that memory reservation must be checked; but that reservation
940 * belongs to shared memory object, not to vma: so now clear it.
942 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
943 vma->vm_flags &= ~VM_ACCOUNT;
945 /* Can addr have changed??
947 * Answer: Yes, several device drivers can do it in their
948 * f_op->mmap method. -DaveM
950 addr = vma->vm_start;
952 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
953 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
954 vma_link(mm, vma, prev, rb_link, rb_parent);
956 atomic_inc(&inode->i_writecount);
960 atomic_inc(&inode->i_writecount);
963 mpol_free(vma_policy(vma));
964 kmem_cache_free(vm_area_cachep, vma);
967 mm->total_vm += len >> PAGE_SHIFT;
968 if (vm_flags & VM_LOCKED) {
969 mm->locked_vm += len >> PAGE_SHIFT;
970 make_pages_present(addr, addr + len);
972 if (flags & MAP_POPULATE) {
973 up_write(&mm->mmap_sem);
974 sys_remap_file_pages(addr, len, 0,
975 pgoff, flags & MAP_NONBLOCK);
976 down_write(&mm->mmap_sem);
982 atomic_inc(&inode->i_writecount);
986 /* Undo any partial mapping done by a device driver. */
987 zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
989 kmem_cache_free(vm_area_cachep, vma);
992 vm_unacct_memory(charged);
996 EXPORT_SYMBOL(do_mmap_pgoff);
998 /* Get an address range which is currently unmapped.
999 * For shmat() with addr=0.
1001 * Ugly calling convention alert:
1002 * Return value with the low bits set means error value,
1004 * if (ret & ~PAGE_MASK)
1007 * This function "knows" that -ENOMEM has the bits set.
1009 #ifndef HAVE_ARCH_UNMAPPED_AREA
1010 static inline unsigned long
1011 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1012 unsigned long len, unsigned long pgoff, unsigned long flags, unsigned long exec)
1014 struct mm_struct *mm = current->mm;
1015 struct vm_area_struct *vma;
1016 unsigned long start_addr;
1018 if (len > TASK_SIZE)
1022 addr = PAGE_ALIGN(addr);
1023 vma = find_vma(mm, addr);
1024 if (TASK_SIZE - len >= addr &&
1025 (!vma || addr + len <= vma->vm_start))
1028 start_addr = addr = mm->free_area_cache;
1031 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1032 /* At this point: (!vma || addr < vma->vm_end). */
1033 if (TASK_SIZE - len < addr) {
1035 * Start a new search - just in case we missed
1038 if (start_addr != TASK_UNMAPPED_BASE) {
1039 start_addr = addr = TASK_UNMAPPED_BASE;
1044 if (!vma || addr + len <= vma->vm_start) {
1046 * Remember the place where we stopped the search:
1048 mm->free_area_cache = addr + len;
1055 extern unsigned long
1056 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
1057 unsigned long, unsigned long, unsigned long);
1061 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1062 unsigned long pgoff, unsigned long flags, unsigned long exec)
1064 if (flags & MAP_FIXED) {
1067 if (addr > TASK_SIZE - len)
1069 if (addr & ~PAGE_MASK)
1071 if (file && is_file_hugepages(file)) {
1073 * Check if the given range is hugepage aligned, and
1074 * can be made suitable for hugepages.
1076 ret = prepare_hugepage_range(addr, len);
1079 * Ensure that a normal request is not falling in a
1080 * reserved hugepage range. For some archs like IA-64,
1081 * there is a separate region for hugepages.
1083 ret = is_hugepage_only_range(addr, len);
1090 if (file && file->f_op && file->f_op->get_unmapped_area)
1091 return file->f_op->get_unmapped_area(file, addr, len,
1094 return arch_get_unmapped_area(file, addr, len, pgoff, flags, exec);
1097 EXPORT_SYMBOL(get_unmapped_area);
1099 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1100 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1102 struct vm_area_struct *vma = NULL;
1105 /* Check the cache first. */
1106 /* (Cache hit rate is typically around 35%.) */
1107 vma = mm->mmap_cache;
1108 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1109 struct rb_node * rb_node;
1111 rb_node = mm->mm_rb.rb_node;
1115 struct vm_area_struct * vma_tmp;
1117 vma_tmp = rb_entry(rb_node,
1118 struct vm_area_struct, vm_rb);
1120 if (vma_tmp->vm_end > addr) {
1122 if (vma_tmp->vm_start <= addr)
1124 rb_node = rb_node->rb_left;
1126 rb_node = rb_node->rb_right;
1129 mm->mmap_cache = vma;
1135 EXPORT_SYMBOL(find_vma);
1137 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1138 struct vm_area_struct *
1139 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1140 struct vm_area_struct **pprev)
1142 struct vm_area_struct *vma = NULL, *prev = NULL;
1143 struct rb_node * rb_node;
1147 /* Guard against addr being lower than the first VMA */
1150 /* Go through the RB tree quickly. */
1151 rb_node = mm->mm_rb.rb_node;
1154 struct vm_area_struct *vma_tmp;
1155 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1157 if (addr < vma_tmp->vm_end) {
1158 rb_node = rb_node->rb_left;
1161 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1163 rb_node = rb_node->rb_right;
1169 return prev ? prev->vm_next : vma;
1173 static int over_stack_limit(unsigned long sz)
1175 if (sz < EXEC_STACK_BIAS)
1177 return (sz - EXEC_STACK_BIAS) > current->rlim[RLIMIT_STACK].rlim_cur;
1180 #ifdef CONFIG_STACK_GROWSUP
1182 * vma is the first one with address > vma->vm_end. Have to extend vma.
1184 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1188 if (!(vma->vm_flags & VM_GROWSUP))
1192 * We must make sure the anon_vma is allocated
1193 * so that the anon_vma locking is not a noop.
1195 if (unlikely(anon_vma_prepare(vma)))
1200 * vma->vm_start/vm_end cannot change under us because the caller
1201 * is required to hold the mmap_sem in read mode. We need the
1202 * anon_vma lock to serialize against concurrent expand_stacks.
1204 address += 4 + PAGE_SIZE - 1;
1205 address &= PAGE_MASK;
1206 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1209 if (security_vm_enough_memory(grow)) {
1210 anon_vma_unlock(vma);
1214 if (over_stack_limit(address - vma->vm_start) ||
1215 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1216 current->rlim[RLIMIT_AS].rlim_cur) {
1217 anon_vma_unlock(vma);
1218 vm_unacct_memory(grow);
1221 vma->vm_end = address;
1222 vma->vm_mm->total_vm += grow;
1223 if (vma->vm_flags & VM_LOCKED)
1224 vma->vm_mm->locked_vm += grow;
1225 anon_vma_unlock(vma);
1229 struct vm_area_struct *
1230 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1232 struct vm_area_struct *vma, *prev;
1235 vma = find_vma_prev(mm, addr, &prev);
1236 if (vma && (vma->vm_start <= addr))
1238 if (!prev || expand_stack(prev, addr))
1240 if (prev->vm_flags & VM_LOCKED) {
1241 make_pages_present(addr, prev->vm_end);
1247 * vma is the first one with address < vma->vm_start. Have to extend vma.
1249 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1254 * We must make sure the anon_vma is allocated
1255 * so that the anon_vma locking is not a noop.
1257 if (unlikely(anon_vma_prepare(vma)))
1262 * vma->vm_start/vm_end cannot change under us because the caller
1263 * is required to hold the mmap_sem in read mode. We need the
1264 * anon_vma lock to serialize against concurrent expand_stacks.
1266 address &= PAGE_MASK;
1267 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1270 if (security_vm_enough_memory(grow)) {
1271 anon_vma_unlock(vma);
1275 if (over_stack_limit(vma->vm_end - address) ||
1276 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1277 current->rlim[RLIMIT_AS].rlim_cur) {
1278 anon_vma_unlock(vma);
1279 vm_unacct_memory(grow);
1282 vma->vm_start = address;
1283 vma->vm_pgoff -= grow;
1284 vma->vm_mm->total_vm += grow;
1285 if (vma->vm_flags & VM_LOCKED)
1286 vma->vm_mm->locked_vm += grow;
1287 anon_vma_unlock(vma);
1291 struct vm_area_struct *
1292 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1294 struct vm_area_struct * vma;
1295 unsigned long start;
1298 vma = find_vma(mm,addr);
1301 if (vma->vm_start <= addr)
1303 if (!(vma->vm_flags & VM_GROWSDOWN))
1305 start = vma->vm_start;
1306 if (expand_stack(vma, addr))
1308 if (vma->vm_flags & VM_LOCKED) {
1309 make_pages_present(addr, start);
1316 * Try to free as many page directory entries as we can,
1317 * without having to work very hard at actually scanning
1318 * the page tables themselves.
1320 * Right now we try to free page tables if we have a nice
1321 * PGDIR-aligned area that got free'd up. We could be more
1322 * granular if we want to, but this is fast and simple,
1323 * and covers the bad cases.
1325 * "prev", if it exists, points to a vma before the one
1326 * we just free'd - but there's no telling how much before.
1328 static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev,
1329 unsigned long start, unsigned long end)
1331 unsigned long first = start & PGDIR_MASK;
1332 unsigned long last = end + PGDIR_SIZE - 1;
1333 unsigned long start_index, end_index;
1334 struct mm_struct *mm = tlb->mm;
1340 if (prev->vm_end > start) {
1341 if (last > prev->vm_start)
1342 last = prev->vm_start;
1347 struct vm_area_struct *next = prev->vm_next;
1350 if (next->vm_start < start) {
1354 if (last > next->vm_start)
1355 last = next->vm_start;
1357 if (prev->vm_end > first)
1358 first = prev->vm_end + PGDIR_SIZE - 1;
1362 if (last < first) /* for arches with discontiguous pgd indices */
1365 * If the PGD bits are not consecutive in the virtual address, the
1366 * old method of shifting the VA >> by PGDIR_SHIFT doesn't work.
1368 start_index = pgd_index(first);
1369 if (start_index < FIRST_USER_PGD_NR)
1370 start_index = FIRST_USER_PGD_NR;
1371 end_index = pgd_index(last);
1372 if (end_index > start_index) {
1373 clear_page_tables(tlb, start_index, end_index - start_index);
1374 flush_tlb_pgtables(mm, first & PGDIR_MASK, last & PGDIR_MASK);
1378 /* Normal function to fix up a mapping
1379 * This function is the default for when an area has no specific
1380 * function. This may be used as part of a more specific routine.
1382 * By the time this function is called, the area struct has been
1383 * removed from the process mapping list.
1385 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1387 size_t len = area->vm_end - area->vm_start;
1388 unsigned long old_end = area->vm_end;
1390 area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1391 if (area->vm_flags & VM_LOCKED)
1392 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1394 * Is this a new hole at the lowest possible address?
1396 if (area->vm_start >= TASK_UNMAPPED_BASE &&
1397 area->vm_start < area->vm_mm->free_area_cache)
1398 area->vm_mm->free_area_cache = area->vm_start;
1400 * Is this a new hole at the highest possible address?
1402 if (area->vm_start > area->vm_mm->non_executable_cache)
1403 area->vm_mm->non_executable_cache = area->vm_start;
1404 remove_vm_struct(area);
1405 if (unlikely(area->vm_flags & VM_EXEC))
1406 arch_remove_exec_range(mm, old_end);
1410 * Update the VMA and inode share lists.
1412 * Ok - we have the memory areas we should free on the 'free' list,
1413 * so release them, and do the vma updates.
1415 static void unmap_vma_list(struct mm_struct *mm,
1416 struct vm_area_struct *mpnt)
1419 struct vm_area_struct *next = mpnt->vm_next;
1420 unmap_vma(mm, mpnt);
1422 } while (mpnt != NULL);
1427 * Get rid of page table information in the indicated region.
1429 * Called with the page table lock held.
1431 static void unmap_region(struct mm_struct *mm,
1432 struct vm_area_struct *vma,
1433 struct vm_area_struct *prev,
1434 unsigned long start,
1437 struct mmu_gather *tlb;
1438 unsigned long nr_accounted = 0;
1441 tlb = tlb_gather_mmu(mm, 0);
1442 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1443 vm_unacct_memory(nr_accounted);
1445 if (is_hugepage_only_range(start, end - start))
1446 hugetlb_free_pgtables(tlb, prev, start, end);
1448 free_pgtables(tlb, prev, start, end);
1449 tlb_finish_mmu(tlb, start, end);
1453 * Create a list of vma's touched by the unmap, removing them from the mm's
1454 * vma list as we go..
1457 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1458 struct vm_area_struct *prev, unsigned long end)
1460 struct vm_area_struct **insertion_point;
1461 struct vm_area_struct *tail_vma = NULL;
1463 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1465 rb_erase(&vma->vm_rb, &mm->mm_rb);
1469 } while (vma && vma->vm_start < end);
1470 *insertion_point = vma;
1471 tail_vma->vm_next = NULL;
1472 mm->mmap_cache = NULL; /* Kill the cache. */
1476 * Split a vma into two pieces at address 'addr', a new vma is allocated
1477 * either for the first part or the the tail.
1479 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1480 unsigned long addr, int new_below)
1482 struct mempolicy *pol;
1483 struct vm_area_struct *new;
1485 if (mm->map_count >= sysctl_max_map_count)
1488 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1492 /* most fields are the same, copy all, and then fixup */
1494 vma_prio_tree_init(new);
1499 new->vm_start = addr;
1500 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1503 pol = mpol_copy(vma_policy(vma));
1505 kmem_cache_free(vm_area_cachep, new);
1506 return PTR_ERR(pol);
1508 vma_set_policy(new, pol);
1511 get_file(new->vm_file);
1513 if (new->vm_ops && new->vm_ops->open)
1514 new->vm_ops->open(new);
1517 unsigned long old_end = vma->vm_end;
1519 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1520 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1521 if (vma->vm_flags & VM_EXEC)
1522 arch_remove_exec_range(mm, old_end);
1524 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1529 /* Munmap is split into 2 main parts -- this part which finds
1530 * what needs doing, and the areas themselves, which do the
1531 * work. This now handles partial unmappings.
1532 * Jeremy Fitzhardinge <jeremy@goop.org>
1534 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1537 struct vm_area_struct *mpnt, *prev, *last;
1539 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1542 if ((len = PAGE_ALIGN(len)) == 0)
1545 /* Find the first overlapping VMA */
1546 mpnt = find_vma_prev(mm, start, &prev);
1549 /* we have start < mpnt->vm_end */
1551 if (is_vm_hugetlb_page(mpnt)) {
1552 int ret = is_aligned_hugepage_range(start, len);
1558 /* if it doesn't overlap, we have nothing.. */
1560 if (mpnt->vm_start >= end)
1563 /* Something will probably happen, so notify. */
1564 if (mpnt->vm_file && (mpnt->vm_flags & VM_EXEC))
1565 profile_exec_unmap(mm);
1568 * If we need to split any vma, do it now to save pain later.
1570 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1571 * unmapped vm_area_struct will remain in use: so lower split_vma
1572 * places tmp vma above, and higher split_vma places tmp vma below.
1574 if (start > mpnt->vm_start) {
1575 if (split_vma(mm, mpnt, start, 0))
1580 /* Does it split the last one? */
1581 last = find_vma(mm, end);
1582 if (last && end > last->vm_start) {
1583 if (split_vma(mm, last, end, 1))
1586 mpnt = prev? prev->vm_next: mm->mmap;
1589 * Remove the vma's, and unmap the actual pages
1591 detach_vmas_to_be_unmapped(mm, mpnt, prev, end);
1592 spin_lock(&mm->page_table_lock);
1593 unmap_region(mm, mpnt, prev, start, end);
1594 spin_unlock(&mm->page_table_lock);
1596 /* Fix up all other VM information */
1597 unmap_vma_list(mm, mpnt);
1602 EXPORT_SYMBOL(do_munmap);
1604 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1607 struct mm_struct *mm = current->mm;
1609 down_write(&mm->mmap_sem);
1610 ret = do_munmap(mm, addr, len);
1611 up_write(&mm->mmap_sem);
1616 * this is really a simplified "do_mmap". it only handles
1617 * anonymous maps. eventually we may be able to do some
1618 * brk-specific accounting here.
1620 unsigned long do_brk(unsigned long addr, unsigned long len)
1622 struct mm_struct * mm = current->mm;
1623 struct vm_area_struct * vma, * prev;
1624 unsigned long flags;
1625 struct rb_node ** rb_link, * rb_parent;
1626 pgoff_t pgoff = addr >> PAGE_SHIFT;
1628 len = PAGE_ALIGN(len);
1632 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1638 if (mm->def_flags & VM_LOCKED) {
1639 unsigned long locked = mm->locked_vm << PAGE_SHIFT;
1641 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
1646 * Clear old maps. this also does some error checking for us
1649 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1650 if (vma && vma->vm_start < addr + len) {
1651 if (do_munmap(mm, addr, len))
1656 /* Check against address space limits *after* clearing old maps... */
1657 if ((mm->total_vm << PAGE_SHIFT) + len
1658 > current->rlim[RLIMIT_AS].rlim_cur)
1661 if (mm->map_count > sysctl_max_map_count)
1664 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1667 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1669 /* Can we just expand an old private anonymous mapping? */
1670 if (vma_merge(mm, prev, addr, addr + len, flags,
1671 NULL, NULL, pgoff, NULL))
1675 * create a vma struct for an anonymous mapping
1677 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1679 vm_unacct_memory(len >> PAGE_SHIFT);
1682 memset(vma, 0, sizeof(*vma));
1685 vma->vm_start = addr;
1686 vma->vm_end = addr + len;
1687 vma->vm_pgoff = pgoff;
1688 vma->vm_flags = flags;
1689 vma->vm_page_prot = protection_map[flags & 0x0f];
1690 vma_link(mm, vma, prev, rb_link, rb_parent);
1692 mm->total_vm += len >> PAGE_SHIFT;
1693 if (flags & VM_LOCKED) {
1694 mm->locked_vm += len >> PAGE_SHIFT;
1695 make_pages_present(addr, addr + len);
1700 EXPORT_SYMBOL(do_brk);
1702 /* Release all mmaps. */
1703 void exit_mmap(struct mm_struct *mm)
1705 struct mmu_gather *tlb;
1706 struct vm_area_struct *vma;
1707 unsigned long nr_accounted = 0;
1709 profile_exit_mmap(mm);
1713 spin_lock(&mm->page_table_lock);
1715 tlb = tlb_gather_mmu(mm, 1);
1717 /* Use ~0UL here to ensure all VMAs in the mm are unmapped */
1718 mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0,
1719 ~0UL, &nr_accounted, NULL);
1720 vm_unacct_memory(nr_accounted);
1721 BUG_ON(mm->map_count); /* This is just debugging */
1722 clear_page_tables(tlb, FIRST_USER_PGD_NR, USER_PTRS_PER_PGD);
1723 tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm));
1726 mm->mmap = mm->mmap_cache = NULL;
1727 mm->mm_rb = RB_ROOT;
1731 arch_flush_exec_range(mm);
1733 spin_unlock(&mm->page_table_lock);
1736 * Walk the list again, actually closing and freeing it
1737 * without holding any MM locks.
1740 struct vm_area_struct *next = vma->vm_next;
1741 remove_vm_struct(vma);
1746 /* Insert vm structure into process list sorted by address
1747 * and into the inode's i_mmap tree. If vm_file is non-NULL
1748 * then i_mmap_lock is taken here.
1750 void insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1752 struct vm_area_struct * __vma, * prev;
1753 struct rb_node ** rb_link, * rb_parent;
1756 * The vm_pgoff of a purely anonymous vma should be irrelevant
1757 * until its first write fault, when page's anon_vma and index
1758 * are set. But now set the vm_pgoff it will almost certainly
1759 * end up with (unless mremap moves it elsewhere before that
1760 * first wfault), so /proc/pid/maps tells a consistent story.
1762 * By setting it to reflect the virtual start address of the
1763 * vma, merges and splits can happen in a seamless way, just
1764 * using the existing file pgoff checks and manipulations.
1765 * Similarly in do_mmap_pgoff and in do_brk.
1767 if (!vma->vm_file) {
1768 BUG_ON(vma->anon_vma);
1769 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1771 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1772 if (__vma && __vma->vm_start < vma->vm_end)
1774 vma_link(mm, vma, prev, rb_link, rb_parent);
1778 * Copy the vma structure to a new location in the same mm,
1779 * prior to moving page table entries, to effect an mremap move.
1781 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
1782 unsigned long addr, unsigned long len, pgoff_t pgoff)
1784 struct vm_area_struct *vma = *vmap;
1785 unsigned long vma_start = vma->vm_start;
1786 struct mm_struct *mm = vma->vm_mm;
1787 struct vm_area_struct *new_vma, *prev;
1788 struct rb_node **rb_link, *rb_parent;
1789 struct mempolicy *pol;
1792 * If anonymous vma has not yet been faulted, update new pgoff
1793 * to match new location, to increase its chance of merging.
1795 if (!vma->vm_file && !vma->anon_vma)
1796 pgoff = addr >> PAGE_SHIFT;
1798 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1799 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
1800 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
1803 * Source vma may have been merged into new_vma
1805 if (vma_start >= new_vma->vm_start &&
1806 vma_start < new_vma->vm_end)
1809 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1812 vma_prio_tree_init(new_vma);
1813 pol = mpol_copy(vma_policy(vma));
1815 kmem_cache_free(vm_area_cachep, new_vma);
1818 vma_set_policy(new_vma, pol);
1819 new_vma->vm_start = addr;
1820 new_vma->vm_end = addr + len;
1821 new_vma->vm_pgoff = pgoff;
1822 if (new_vma->vm_file)
1823 get_file(new_vma->vm_file);
1824 if (new_vma->vm_ops && new_vma->vm_ops->open)
1825 new_vma->vm_ops->open(new_vma);
1826 vma_link(mm, new_vma, prev, rb_link, rb_parent);