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/pgalloc.h>
29 #include <asm/cacheflush.h>
33 * WARNING: the debugging will use recursive algorithms so never enable this
34 * unless you know what you are doing.
38 /* description of effects of mapping type and prot in current implementation.
39 * this is due to the limited x86 page protection hardware. The expected
40 * behavior is in parens:
43 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
44 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
45 * w: (no) no w: (no) no w: (yes) yes w: (no) no
46 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
48 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
49 * w: (no) no w: (no) no w: (copy) copy w: (no) no
50 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
53 pgprot_t protection_map[16] = {
54 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
55 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
58 int sysctl_overcommit_memory = 0; /* default is heuristic overcommit */
59 int sysctl_overcommit_ratio = 50; /* default is 50% */
60 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
61 atomic_t vm_committed_space = ATOMIC_INIT(0);
63 EXPORT_SYMBOL(sysctl_overcommit_memory);
64 EXPORT_SYMBOL(sysctl_overcommit_ratio);
65 EXPORT_SYMBOL(sysctl_max_map_count);
66 EXPORT_SYMBOL(vm_committed_space);
69 * Requires inode->i_mapping->i_mmap_lock
71 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
72 struct file *file, struct address_space *mapping)
74 if (vma->vm_flags & VM_DENYWRITE)
75 atomic_inc(&file->f_dentry->d_inode->i_writecount);
76 if (vma->vm_flags & VM_SHARED)
77 mapping->i_mmap_writable--;
79 flush_dcache_mmap_lock(mapping);
80 if (unlikely(vma->vm_flags & VM_NONLINEAR))
81 list_del_init(&vma->shared.vm_set.list);
83 vma_prio_tree_remove(vma, &mapping->i_mmap);
84 flush_dcache_mmap_unlock(mapping);
88 * Remove one vm structure and free it.
90 static void remove_vm_struct(struct vm_area_struct *vma)
92 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_prio_tree_init(vma);
299 __vma_link_list(mm, vma, prev, rb_parent);
300 __vma_link_rb(mm, vma, rb_link, rb_parent);
301 __vma_link_file(vma);
302 __anon_vma_link(vma);
305 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
306 struct vm_area_struct *prev, struct rb_node **rb_link,
307 struct rb_node *rb_parent)
309 struct address_space *mapping = NULL;
312 mapping = vma->vm_file->f_mapping;
315 spin_lock(&mapping->i_mmap_lock);
317 __vma_link(mm, vma, prev, rb_link, rb_parent);
318 anon_vma_unlock(vma);
320 spin_unlock(&mapping->i_mmap_lock);
322 mark_mm_hugetlb(mm, vma);
328 * Insert vm structure into process list sorted by address and into the
329 * inode's i_mmap tree. The caller should hold mm->mmap_sem and
330 * ->f_mappping->i_mmap_lock if vm_file is non-NULL.
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);
342 mark_mm_hugetlb(mm, vma);
347 static inline void __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
348 struct vm_area_struct *prev)
350 prev->vm_next = vma->vm_next;
351 rb_erase(&vma->vm_rb, &mm->mm_rb);
352 if (mm->mmap_cache == vma)
353 mm->mmap_cache = prev;
354 if (vma->vm_flags & VM_EXEC)
355 arch_remove_exec_range(mm, vma->vm_end);
359 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
360 * is already present in an i_mmap tree without adjusting the tree.
361 * The following helper function should be used when such adjustments
362 * are necessary. The "insert" vma (if any) is to be inserted
363 * before we drop the necessary locks.
365 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
366 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
368 struct mm_struct *mm = vma->vm_mm;
369 struct vm_area_struct *next = vma->vm_next;
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) {
379 again: remove_next = 1 + (end > next->vm_end);
381 anon_vma = next->anon_vma;
382 } else if (end < vma->vm_end || end > next->vm_start) {
384 * vma shrinks, and !insert tells it's not
385 * split_vma inserting another: so it must
386 * be mprotect shifting the boundary down.
388 * vma expands, overlapping part of the next:
389 * must be mprotect shifting the boundary up.
391 BUG_ON(vma->vm_end != next->vm_start);
392 adjust_next = end - next->vm_start;
393 anon_vma = next->anon_vma;
398 mapping = file->f_mapping;
399 if (!(vma->vm_flags & VM_NONLINEAR))
400 root = &mapping->i_mmap;
401 spin_lock(&mapping->i_mmap_lock);
405 * When changing only vma->vm_end, we don't really need
406 * anon_vma lock: but is that case worth optimizing out?
409 anon_vma = vma->anon_vma;
411 spin_lock(&anon_vma->lock);
414 flush_dcache_mmap_lock(mapping);
415 vma_prio_tree_remove(vma, root);
417 vma_prio_tree_remove(next, root);
420 vma->vm_start = start;
422 vma->vm_pgoff = pgoff;
424 next->vm_start += adjust_next;
425 next->vm_pgoff += adjust_next >> PAGE_SHIFT;
430 vma_prio_tree_init(next);
431 vma_prio_tree_insert(next, root);
433 vma_prio_tree_init(vma);
434 vma_prio_tree_insert(vma, root);
435 flush_dcache_mmap_unlock(mapping);
440 * vma_merge has merged next into vma, and needs
441 * us to remove next before dropping the locks.
443 __vma_unlink(mm, next, vma);
445 __remove_shared_vm_struct(next, file, mapping);
447 __anon_vma_merge(vma, next);
450 * split_vma has split insert from vma, and needs
451 * us to insert it before dropping the locks
452 * (it may either follow vma or precede it).
454 __insert_vm_struct(mm, insert);
458 spin_unlock(&anon_vma->lock);
460 spin_unlock(&mapping->i_mmap_lock);
466 mpol_free(vma_policy(next));
467 kmem_cache_free(vm_area_cachep, next);
469 * In mprotect's case 6 (see comments on vma_merge),
470 * we must remove another next too. It would clutter
471 * up the code too much to do both in one go.
473 if (remove_next == 2) {
483 * If the vma has a ->close operation then the driver probably needs to release
484 * per-vma resources, so we don't attempt to merge those.
486 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
488 static inline int is_mergeable_vma(struct vm_area_struct *vma,
489 struct file *file, unsigned long vm_flags)
491 if (vma->vm_flags != vm_flags)
493 if (vma->vm_file != file)
495 if (vma->vm_ops && vma->vm_ops->close)
500 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
501 struct anon_vma *anon_vma2)
503 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
507 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
508 * in front of (at a lower virtual address and file offset than) the vma.
510 * We cannot merge two vmas if they have differently assigned (non-NULL)
511 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
513 * We don't check here for the merged mmap wrapping around the end of pagecache
514 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
515 * wrap, nor mmaps which cover the final page at index -1UL.
518 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
519 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
521 if (is_mergeable_vma(vma, file, vm_flags) &&
522 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
523 if (vma->vm_pgoff == vm_pgoff)
530 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
531 * beyond (at a higher virtual address and file offset than) the vma.
533 * We cannot merge two vmas if they have differently assigned (non-NULL)
534 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
537 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
538 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
540 if (is_mergeable_vma(vma, file, vm_flags) &&
541 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
543 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
544 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
551 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
552 * whether that can be merged with its predecessor or its successor.
553 * Or both (it neatly fills a hole).
555 * In most cases - when called for mmap, brk or mremap - [addr,end) is
556 * certain not to be mapped by the time vma_merge is called; but when
557 * called for mprotect, it is certain to be already mapped (either at
558 * an offset within prev, or at the start of next), and the flags of
559 * this area are about to be changed to vm_flags - and the no-change
560 * case has already been eliminated.
562 * The following mprotect cases have to be considered, where AAAA is
563 * the area passed down from mprotect_fixup, never extending beyond one
564 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
566 * AAAA AAAA AAAA AAAA
567 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
568 * cannot merge might become might become might become
569 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
570 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
571 * mremap move: PPPPNNNNNNNN 8
573 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
574 * might become case 1 below case 2 below case 3 below
576 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
577 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
579 struct vm_area_struct *vma_merge(struct mm_struct *mm,
580 struct vm_area_struct *prev, unsigned long addr,
581 unsigned long end, unsigned long vm_flags,
582 struct anon_vma *anon_vma, struct file *file,
583 pgoff_t pgoff, struct mempolicy *policy)
585 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
586 struct vm_area_struct *area, *next;
589 * We later require that vma->vm_flags == vm_flags,
590 * so this tests vma->vm_flags & VM_SPECIAL, too.
592 if (vm_flags & VM_SPECIAL)
596 next = prev->vm_next;
600 if (next && next->vm_end == end) /* cases 6, 7, 8 */
601 next = next->vm_next;
604 * Can it merge with the predecessor?
606 if (prev && prev->vm_end == addr &&
607 mpol_equal(vma_policy(prev), policy) &&
608 can_vma_merge_after(prev, vm_flags,
609 anon_vma, file, pgoff)) {
611 * OK, it can. Can we now merge in the successor as well?
613 if (next && end == next->vm_start &&
614 mpol_equal(policy, vma_policy(next)) &&
615 can_vma_merge_before(next, vm_flags,
616 anon_vma, file, pgoff+pglen) &&
617 is_mergeable_anon_vma(prev->anon_vma,
620 vma_adjust(prev, prev->vm_start,
621 next->vm_end, prev->vm_pgoff, NULL);
622 } else /* cases 2, 5, 7 */
623 vma_adjust(prev, prev->vm_start,
624 end, prev->vm_pgoff, NULL);
625 if (prev->vm_flags & VM_EXEC)
626 arch_add_exec_range(mm, prev->vm_end);
631 * Can this new request be merged in front of next?
633 if (next && end == next->vm_start &&
634 mpol_equal(policy, vma_policy(next)) &&
635 can_vma_merge_before(next, vm_flags,
636 anon_vma, file, pgoff+pglen)) {
637 if (prev && addr < prev->vm_end) /* case 4 */
638 vma_adjust(prev, prev->vm_start,
639 addr, prev->vm_pgoff, NULL);
640 else /* cases 3, 8 */
641 vma_adjust(area, addr, next->vm_end,
642 next->vm_pgoff - pglen, NULL);
650 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
651 * neighbouring vmas for a suitable anon_vma, before it goes off
652 * to allocate a new anon_vma. It checks because a repetitive
653 * sequence of mprotects and faults may otherwise lead to distinct
654 * anon_vmas being allocated, preventing vma merge in subsequent
657 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
659 struct vm_area_struct *near;
660 unsigned long vm_flags;
667 * Since only mprotect tries to remerge vmas, match flags
668 * which might be mprotected into each other later on.
669 * Neither mlock nor madvise tries to remerge at present,
670 * so leave their flags as obstructing a merge.
672 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
673 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
675 if (near->anon_vma && vma->vm_end == near->vm_start &&
676 mpol_equal(vma_policy(vma), vma_policy(near)) &&
677 can_vma_merge_before(near, vm_flags,
678 NULL, vma->vm_file, vma->vm_pgoff +
679 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
680 return near->anon_vma;
683 * It is potentially slow to have to call find_vma_prev here.
684 * But it's only on the first write fault on the vma, not
685 * every time, and we could devise a way to avoid it later
686 * (e.g. stash info in next's anon_vma_node when assigning
687 * an anon_vma, or when trying vma_merge). Another time.
689 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
694 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
695 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
697 if (near->anon_vma && near->vm_end == vma->vm_start &&
698 mpol_equal(vma_policy(near), vma_policy(vma)) &&
699 can_vma_merge_after(near, vm_flags,
700 NULL, vma->vm_file, vma->vm_pgoff))
701 return near->anon_vma;
704 * There's no absolute need to look only at touching neighbours:
705 * we could search further afield for "compatible" anon_vmas.
706 * But it would probably just be a waste of time searching,
707 * or lead to too many vmas hanging off the same anon_vma.
708 * We're trying to allow mprotect remerging later on,
709 * not trying to minimize memory used for anon_vmas.
715 * The caller must hold down_write(current->mm->mmap_sem).
718 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
719 unsigned long len, unsigned long prot,
720 unsigned long flags, unsigned long pgoff)
722 struct mm_struct * mm = current->mm;
723 struct vm_area_struct * vma, * prev;
725 unsigned int vm_flags;
726 int correct_wcount = 0;
728 struct rb_node ** rb_link, * rb_parent;
730 unsigned long charged = 0;
733 if (is_file_hugepages(file))
736 if (!file->f_op || !file->f_op->mmap)
739 if ((prot & PROT_EXEC) &&
740 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
747 /* Careful about overflows.. */
748 len = PAGE_ALIGN(len);
749 if (!len || len > TASK_SIZE)
752 /* offset overflow? */
753 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
756 /* Too many mappings? */
757 if (mm->map_count > sysctl_max_map_count)
760 /* Obtain the address to map to. we verify (or select) it and ensure
761 * that it represents a valid section of the address space.
763 addr = get_unmapped_area(file, addr, len, pgoff, flags, prot & PROT_EXEC);
764 if (addr & ~PAGE_MASK)
767 /* Do simple checking here so the lower-level routines won't have
768 * to. we assume access permissions have been handled by the open
769 * of the memory object, so we don't do any here.
771 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
772 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
774 if (flags & MAP_LOCKED) {
775 if (!capable(CAP_IPC_LOCK))
777 vm_flags |= VM_LOCKED;
779 /* mlock MCL_FUTURE? */
780 if (vm_flags & VM_LOCKED) {
781 unsigned long locked = mm->locked_vm << PAGE_SHIFT;
783 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
787 inode = file ? file->f_dentry->d_inode : NULL;
790 switch (flags & MAP_TYPE) {
792 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
796 * Make sure we don't allow writing to an append-only
799 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
803 * Make sure there are no mandatory locks on the file.
805 if (locks_verify_locked(inode))
808 vm_flags |= VM_SHARED | VM_MAYSHARE;
809 if (!(file->f_mode & FMODE_WRITE))
810 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
814 if (!(file->f_mode & FMODE_READ))
822 switch (flags & MAP_TYPE) {
824 vm_flags |= VM_SHARED | VM_MAYSHARE;
828 * Set pgoff according to addr for anon_vma.
830 pgoff = addr >> PAGE_SHIFT;
837 error = security_file_mmap(file, prot, flags);
844 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
845 if (vma && vma->vm_start < addr + len) {
846 if (do_munmap(mm, addr, len))
851 /* Check against address space limit. */
852 if ((mm->total_vm << PAGE_SHIFT) + len
853 > current->rlim[RLIMIT_AS].rlim_cur)
856 if (accountable && (!(flags & MAP_NORESERVE) ||
857 sysctl_overcommit_memory > 1)) {
858 if (vm_flags & VM_SHARED) {
859 /* Check memory availability in shmem_file_setup? */
860 vm_flags |= VM_ACCOUNT;
861 } else if (vm_flags & VM_WRITE) {
863 * Private writable mapping: check memory availability
865 charged = len >> PAGE_SHIFT;
866 if (security_vm_enough_memory(charged))
868 vm_flags |= VM_ACCOUNT;
873 * Can we just expand an old private anonymous mapping?
874 * The VM_SHARED test is necessary because shmem_zero_setup
875 * will create the file object for a shared anonymous map below.
877 if (!file && !(vm_flags & VM_SHARED) &&
878 vma_merge(mm, prev, addr, addr + len, vm_flags,
879 NULL, NULL, pgoff, NULL))
883 * Determine the object being mapped and call the appropriate
884 * specific mapper. the address has already been validated, but
885 * not unmapped, but the maps are removed from the list.
887 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
892 memset(vma, 0, sizeof(*vma));
895 vma->vm_start = addr;
896 vma->vm_end = addr + len;
897 vma->vm_flags = vm_flags;
898 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
899 vma->vm_pgoff = pgoff;
903 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
905 if (vm_flags & VM_DENYWRITE) {
906 error = deny_write_access(file);
913 error = file->f_op->mmap(file, vma);
915 goto unmap_and_free_vma;
916 } else if (vm_flags & VM_SHARED) {
917 error = shmem_zero_setup(vma);
922 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
923 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
924 * that memory reservation must be checked; but that reservation
925 * belongs to shared memory object, not to vma: so now clear it.
927 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
928 vma->vm_flags &= ~VM_ACCOUNT;
930 /* Can addr have changed??
932 * Answer: Yes, several device drivers can do it in their
933 * f_op->mmap method. -DaveM
935 addr = vma->vm_start;
937 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
938 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
939 vma_link(mm, vma, prev, rb_link, rb_parent);
941 atomic_inc(&inode->i_writecount);
945 atomic_inc(&inode->i_writecount);
948 mpol_free(vma_policy(vma));
949 kmem_cache_free(vm_area_cachep, vma);
952 mm->total_vm += len >> PAGE_SHIFT;
953 if (vm_flags & VM_LOCKED) {
954 mm->locked_vm += len >> PAGE_SHIFT;
955 make_pages_present(addr, addr + len);
957 if (flags & MAP_POPULATE) {
958 up_write(&mm->mmap_sem);
959 sys_remap_file_pages(addr, len, 0,
960 pgoff, flags & MAP_NONBLOCK);
961 down_write(&mm->mmap_sem);
967 atomic_inc(&inode->i_writecount);
971 /* Undo any partial mapping done by a device driver. */
972 zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
974 kmem_cache_free(vm_area_cachep, vma);
977 vm_unacct_memory(charged);
981 EXPORT_SYMBOL(do_mmap_pgoff);
983 /* Get an address range which is currently unmapped.
984 * For shmat() with addr=0.
986 * Ugly calling convention alert:
987 * Return value with the low bits set means error value,
989 * if (ret & ~PAGE_MASK)
992 * This function "knows" that -ENOMEM has the bits set.
994 #ifndef HAVE_ARCH_UNMAPPED_AREA
995 static inline unsigned long
996 arch_get_unmapped_area(struct file *filp, unsigned long addr,
997 unsigned long len, unsigned long pgoff, unsigned long flags, unsigned long exec)
999 struct mm_struct *mm = current->mm;
1000 struct vm_area_struct *vma;
1001 unsigned long start_addr;
1003 if (len > TASK_SIZE)
1007 addr = PAGE_ALIGN(addr);
1008 vma = find_vma(mm, addr);
1009 if (TASK_SIZE - len >= addr &&
1010 (!vma || addr + len <= vma->vm_start))
1013 start_addr = addr = mm->free_area_cache;
1016 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1017 /* At this point: (!vma || addr < vma->vm_end). */
1018 if (TASK_SIZE - len < addr) {
1020 * Start a new search - just in case we missed
1023 if (start_addr != TASK_UNMAPPED_BASE) {
1024 start_addr = addr = TASK_UNMAPPED_BASE;
1029 if (!vma || addr + len <= vma->vm_start) {
1031 * Remember the place where we stopped the search:
1033 mm->free_area_cache = addr + len;
1040 extern unsigned long
1041 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
1042 unsigned long, unsigned long, unsigned long);
1046 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1047 unsigned long pgoff, unsigned long flags, unsigned long exec)
1049 if (flags & MAP_FIXED) {
1052 if (addr > TASK_SIZE - len)
1054 if (addr & ~PAGE_MASK)
1056 if (file && is_file_hugepages(file)) {
1058 * Check if the given range is hugepage aligned, and
1059 * can be made suitable for hugepages.
1061 ret = prepare_hugepage_range(addr, len);
1064 * Ensure that a normal request is not falling in a
1065 * reserved hugepage range. For some archs like IA-64,
1066 * there is a separate region for hugepages.
1068 ret = is_hugepage_only_range(addr, len);
1075 if (file && file->f_op && file->f_op->get_unmapped_area)
1076 return file->f_op->get_unmapped_area(file, addr, len,
1079 return arch_get_unmapped_area(file, addr, len, pgoff, flags, exec);
1082 EXPORT_SYMBOL(get_unmapped_area);
1084 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1085 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1087 struct vm_area_struct *vma = NULL;
1090 /* Check the cache first. */
1091 /* (Cache hit rate is typically around 35%.) */
1092 vma = mm->mmap_cache;
1093 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1094 struct rb_node * rb_node;
1096 rb_node = mm->mm_rb.rb_node;
1100 struct vm_area_struct * vma_tmp;
1102 vma_tmp = rb_entry(rb_node,
1103 struct vm_area_struct, vm_rb);
1105 if (vma_tmp->vm_end > addr) {
1107 if (vma_tmp->vm_start <= addr)
1109 rb_node = rb_node->rb_left;
1111 rb_node = rb_node->rb_right;
1114 mm->mmap_cache = vma;
1120 EXPORT_SYMBOL(find_vma);
1122 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1123 struct vm_area_struct *
1124 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1125 struct vm_area_struct **pprev)
1127 struct vm_area_struct *vma = NULL, *prev = NULL;
1128 struct rb_node * rb_node;
1132 /* Guard against addr being lower than the first VMA */
1135 /* Go through the RB tree quickly. */
1136 rb_node = mm->mm_rb.rb_node;
1139 struct vm_area_struct *vma_tmp;
1140 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1142 if (addr < vma_tmp->vm_end) {
1143 rb_node = rb_node->rb_left;
1146 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1148 rb_node = rb_node->rb_right;
1154 return prev ? prev->vm_next : vma;
1158 static int over_stack_limit(unsigned long sz)
1160 if (sz < EXEC_STACK_BIAS)
1162 return (sz - EXEC_STACK_BIAS) > current->rlim[RLIMIT_STACK].rlim_cur;
1165 #ifdef CONFIG_STACK_GROWSUP
1167 * vma is the first one with address > vma->vm_end. Have to extend vma.
1169 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1173 if (!(vma->vm_flags & VM_GROWSUP))
1177 * We must make sure the anon_vma is allocated
1178 * so that the anon_vma locking is not a noop.
1180 if (unlikely(anon_vma_prepare(vma)))
1185 * vma->vm_start/vm_end cannot change under us because the caller
1186 * is required to hold the mmap_sem in read mode. We need the
1187 * anon_vma lock to serialize against concurrent expand_stacks.
1189 address += 4 + PAGE_SIZE - 1;
1190 address &= PAGE_MASK;
1191 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1194 if (security_vm_enough_memory(grow)) {
1195 anon_vma_unlock(vma);
1199 if (over_stack_limit(address - vma->vm_start) ||
1200 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1201 current->rlim[RLIMIT_AS].rlim_cur) {
1202 anon_vma_unlock(vma);
1203 vm_unacct_memory(grow);
1206 vma->vm_end = address;
1207 vma->vm_mm->total_vm += grow;
1208 if (vma->vm_flags & VM_LOCKED)
1209 vma->vm_mm->locked_vm += grow;
1210 anon_vma_unlock(vma);
1214 struct vm_area_struct *
1215 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1217 struct vm_area_struct *vma, *prev;
1220 vma = find_vma_prev(mm, addr, &prev);
1221 if (vma && (vma->vm_start <= addr))
1223 if (!prev || expand_stack(prev, addr))
1225 if (prev->vm_flags & VM_LOCKED) {
1226 make_pages_present(addr, prev->vm_end);
1232 * vma is the first one with address < vma->vm_start. Have to extend vma.
1234 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1239 * We must make sure the anon_vma is allocated
1240 * so that the anon_vma locking is not a noop.
1242 if (unlikely(anon_vma_prepare(vma)))
1247 * vma->vm_start/vm_end cannot change under us because the caller
1248 * is required to hold the mmap_sem in read mode. We need the
1249 * anon_vma lock to serialize against concurrent expand_stacks.
1251 address &= PAGE_MASK;
1252 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1255 if (security_vm_enough_memory(grow)) {
1256 anon_vma_unlock(vma);
1260 if (over_stack_limit(vma->vm_end - address) ||
1261 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1262 current->rlim[RLIMIT_AS].rlim_cur) {
1263 anon_vma_unlock(vma);
1264 vm_unacct_memory(grow);
1267 vma->vm_start = address;
1268 vma->vm_pgoff -= grow;
1269 vma->vm_mm->total_vm += grow;
1270 if (vma->vm_flags & VM_LOCKED)
1271 vma->vm_mm->locked_vm += grow;
1272 anon_vma_unlock(vma);
1276 struct vm_area_struct *
1277 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1279 struct vm_area_struct * vma;
1280 unsigned long start;
1283 vma = find_vma(mm,addr);
1286 if (vma->vm_start <= addr)
1288 if (!(vma->vm_flags & VM_GROWSDOWN))
1290 start = vma->vm_start;
1291 if (expand_stack(vma, addr))
1293 if (vma->vm_flags & VM_LOCKED) {
1294 make_pages_present(addr, start);
1301 * Try to free as many page directory entries as we can,
1302 * without having to work very hard at actually scanning
1303 * the page tables themselves.
1305 * Right now we try to free page tables if we have a nice
1306 * PGDIR-aligned area that got free'd up. We could be more
1307 * granular if we want to, but this is fast and simple,
1308 * and covers the bad cases.
1310 * "prev", if it exists, points to a vma before the one
1311 * we just free'd - but there's no telling how much before.
1313 static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev,
1314 unsigned long start, unsigned long end)
1316 unsigned long first = start & PGDIR_MASK;
1317 unsigned long last = end + PGDIR_SIZE - 1;
1318 unsigned long start_index, end_index;
1319 struct mm_struct *mm = tlb->mm;
1325 if (prev->vm_end > start) {
1326 if (last > prev->vm_start)
1327 last = prev->vm_start;
1332 struct vm_area_struct *next = prev->vm_next;
1335 if (next->vm_start < start) {
1339 if (last > next->vm_start)
1340 last = next->vm_start;
1342 if (prev->vm_end > first)
1343 first = prev->vm_end + PGDIR_SIZE - 1;
1347 if (last < first) /* for arches with discontiguous pgd indices */
1350 * If the PGD bits are not consecutive in the virtual address, the
1351 * old method of shifting the VA >> by PGDIR_SHIFT doesn't work.
1353 start_index = pgd_index(first);
1354 if (start_index < FIRST_USER_PGD_NR)
1355 start_index = FIRST_USER_PGD_NR;
1356 end_index = pgd_index(last);
1357 if (end_index > start_index) {
1358 clear_page_tables(tlb, start_index, end_index - start_index);
1359 flush_tlb_pgtables(mm, first & PGDIR_MASK, last & PGDIR_MASK);
1363 /* Normal function to fix up a mapping
1364 * This function is the default for when an area has no specific
1365 * function. This may be used as part of a more specific routine.
1367 * By the time this function is called, the area struct has been
1368 * removed from the process mapping list.
1370 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1372 size_t len = area->vm_end - area->vm_start;
1374 area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1375 if (area->vm_flags & VM_LOCKED)
1376 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1378 * Is this a new hole at the lowest possible address?
1380 if (area->vm_start >= TASK_UNMAPPED_BASE &&
1381 area->vm_start < area->vm_mm->free_area_cache)
1382 area->vm_mm->free_area_cache = area->vm_start;
1384 * Is this a new hole at the highest possible address?
1386 if (area->vm_start > area->vm_mm->non_executable_cache)
1387 area->vm_mm->non_executable_cache = area->vm_start;
1389 remove_vm_struct(area);
1390 if (unlikely(area->vm_flags & VM_EXEC))
1391 arch_remove_exec_range(mm, area->vm_end);
1395 * Update the VMA and inode share lists.
1397 * Ok - we have the memory areas we should free on the 'free' list,
1398 * so release them, and do the vma updates.
1400 static void unmap_vma_list(struct mm_struct *mm,
1401 struct vm_area_struct *mpnt)
1404 struct vm_area_struct *next = mpnt->vm_next;
1405 unmap_vma(mm, mpnt);
1407 } while (mpnt != NULL);
1412 * Get rid of page table information in the indicated region.
1414 * Called with the page table lock held.
1416 static void unmap_region(struct mm_struct *mm,
1417 struct vm_area_struct *vma,
1418 struct vm_area_struct *prev,
1419 unsigned long start,
1422 struct mmu_gather *tlb;
1423 unsigned long nr_accounted = 0;
1426 tlb = tlb_gather_mmu(mm, 0);
1427 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1428 vm_unacct_memory(nr_accounted);
1430 if (is_hugepage_only_range(start, end - start))
1431 hugetlb_free_pgtables(tlb, prev, start, end);
1433 free_pgtables(tlb, prev, start, end);
1434 tlb_finish_mmu(tlb, start, end);
1438 * Create a list of vma's touched by the unmap, removing them from the mm's
1439 * vma list as we go..
1442 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1443 struct vm_area_struct *prev, unsigned long end)
1445 struct vm_area_struct **insertion_point;
1446 struct vm_area_struct *tail_vma = NULL;
1448 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1450 rb_erase(&vma->vm_rb, &mm->mm_rb);
1454 } while (vma && vma->vm_start < end);
1455 *insertion_point = vma;
1456 tail_vma->vm_next = NULL;
1457 mm->mmap_cache = NULL; /* Kill the cache. */
1461 * Split a vma into two pieces at address 'addr', a new vma is allocated
1462 * either for the first part or the the tail.
1464 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1465 unsigned long addr, int new_below)
1467 struct mempolicy *pol;
1468 struct vm_area_struct *new;
1470 if (mm->map_count >= sysctl_max_map_count)
1473 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1477 /* most fields are the same, copy all, and then fixup */
1481 if (vma->vm_flags & VM_EXEC)
1482 arch_remove_exec_range(mm, new->vm_end);
1486 new->vm_start = addr;
1487 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1490 pol = mpol_copy(vma_policy(vma));
1492 kmem_cache_free(vm_area_cachep, new);
1493 return PTR_ERR(pol);
1495 vma_set_policy(new, pol);
1498 get_file(new->vm_file);
1500 if (new->vm_ops && new->vm_ops->open)
1501 new->vm_ops->open(new);
1504 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1505 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1507 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1512 /* Munmap is split into 2 main parts -- this part which finds
1513 * what needs doing, and the areas themselves, which do the
1514 * work. This now handles partial unmappings.
1515 * Jeremy Fitzhardinge <jeremy@goop.org>
1517 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1520 struct vm_area_struct *mpnt, *prev, *last;
1522 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1525 if ((len = PAGE_ALIGN(len)) == 0)
1528 /* Find the first overlapping VMA */
1529 mpnt = find_vma_prev(mm, start, &prev);
1532 /* we have start < mpnt->vm_end */
1534 if (is_vm_hugetlb_page(mpnt)) {
1535 int ret = is_aligned_hugepage_range(start, len);
1541 /* if it doesn't overlap, we have nothing.. */
1543 if (mpnt->vm_start >= end)
1546 /* Something will probably happen, so notify. */
1547 if (mpnt->vm_file && (mpnt->vm_flags & VM_EXEC))
1548 profile_exec_unmap(mm);
1551 * If we need to split any vma, do it now to save pain later.
1553 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1554 * unmapped vm_area_struct will remain in use: so lower split_vma
1555 * places tmp vma above, and higher split_vma places tmp vma below.
1557 if (start > mpnt->vm_start) {
1558 if (split_vma(mm, mpnt, start, 0))
1563 /* Does it split the last one? */
1564 last = find_vma(mm, end);
1565 if (last && end > last->vm_start) {
1566 if (split_vma(mm, last, end, 1))
1569 mpnt = prev? prev->vm_next: mm->mmap;
1572 * Remove the vma's, and unmap the actual pages
1574 detach_vmas_to_be_unmapped(mm, mpnt, prev, end);
1575 spin_lock(&mm->page_table_lock);
1576 unmap_region(mm, mpnt, prev, start, end);
1577 spin_unlock(&mm->page_table_lock);
1579 /* Fix up all other VM information */
1580 unmap_vma_list(mm, mpnt);
1585 EXPORT_SYMBOL(do_munmap);
1587 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1590 struct mm_struct *mm = current->mm;
1592 down_write(&mm->mmap_sem);
1593 ret = do_munmap(mm, addr, len);
1594 up_write(&mm->mmap_sem);
1599 * this is really a simplified "do_mmap". it only handles
1600 * anonymous maps. eventually we may be able to do some
1601 * brk-specific accounting here.
1603 unsigned long do_brk(unsigned long addr, unsigned long len)
1605 struct mm_struct * mm = current->mm;
1606 struct vm_area_struct * vma, * prev;
1607 unsigned long flags;
1608 struct rb_node ** rb_link, * rb_parent;
1609 pgoff_t pgoff = addr >> PAGE_SHIFT;
1611 len = PAGE_ALIGN(len);
1615 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1621 if (mm->def_flags & VM_LOCKED) {
1622 unsigned long locked = mm->locked_vm << PAGE_SHIFT;
1624 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
1629 * Clear old maps. this also does some error checking for us
1632 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1633 if (vma && vma->vm_start < addr + len) {
1634 if (do_munmap(mm, addr, len))
1639 /* Check against address space limits *after* clearing old maps... */
1640 if ((mm->total_vm << PAGE_SHIFT) + len
1641 > current->rlim[RLIMIT_AS].rlim_cur)
1644 if (mm->map_count > sysctl_max_map_count)
1647 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1650 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1652 /* Can we just expand an old private anonymous mapping? */
1653 if (vma_merge(mm, prev, addr, addr + len, flags,
1654 NULL, NULL, pgoff, NULL))
1658 * create a vma struct for an anonymous mapping
1660 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1662 vm_unacct_memory(len >> PAGE_SHIFT);
1665 memset(vma, 0, sizeof(*vma));
1668 vma->vm_start = addr;
1669 vma->vm_end = addr + len;
1670 vma->vm_pgoff = pgoff;
1671 vma->vm_flags = flags;
1672 vma->vm_page_prot = protection_map[flags & 0x0f];
1673 vma_link(mm, vma, prev, rb_link, rb_parent);
1675 mm->total_vm += len >> PAGE_SHIFT;
1676 if (flags & VM_LOCKED) {
1677 mm->locked_vm += len >> PAGE_SHIFT;
1678 make_pages_present(addr, addr + len);
1683 EXPORT_SYMBOL(do_brk);
1685 /* Release all mmaps. */
1686 void exit_mmap(struct mm_struct *mm)
1688 struct mmu_gather *tlb;
1689 struct vm_area_struct *vma;
1690 unsigned long nr_accounted = 0;
1692 profile_exit_mmap(mm);
1696 spin_lock(&mm->page_table_lock);
1698 tlb = tlb_gather_mmu(mm, 1);
1700 /* Use ~0UL here to ensure all VMAs in the mm are unmapped */
1701 mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0,
1702 ~0UL, &nr_accounted, NULL);
1703 vm_unacct_memory(nr_accounted);
1704 BUG_ON(mm->map_count); /* This is just debugging */
1705 clear_page_tables(tlb, FIRST_USER_PGD_NR, USER_PTRS_PER_PGD);
1706 tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm));
1709 mm->mmap = mm->mmap_cache = NULL;
1710 mm->mm_rb = RB_ROOT;
1714 arch_flush_exec_range(mm);
1716 spin_unlock(&mm->page_table_lock);
1719 * Walk the list again, actually closing and freeing it
1720 * without holding any MM locks.
1723 struct vm_area_struct *next = vma->vm_next;
1724 remove_vm_struct(vma);
1729 /* Insert vm structure into process list sorted by address
1730 * and into the inode's i_mmap tree. If vm_file is non-NULL
1731 * then i_mmap_lock is taken here.
1733 void insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1735 struct vm_area_struct * __vma, * prev;
1736 struct rb_node ** rb_link, * rb_parent;
1739 * The vm_pgoff of a purely anonymous vma should be irrelevant
1740 * until its first write fault, when page's anon_vma and index
1741 * are set. But now set the vm_pgoff it will almost certainly
1742 * end up with (unless mremap moves it elsewhere before that
1743 * first wfault), so /proc/pid/maps tells a consistent story.
1745 * By setting it to reflect the virtual start address of the
1746 * vma, merges and splits can happen in a seamless way, just
1747 * using the existing file pgoff checks and manipulations.
1748 * Similarly in do_mmap_pgoff and in do_brk.
1750 if (!vma->vm_file) {
1751 BUG_ON(vma->anon_vma);
1752 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1754 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1755 if (__vma && __vma->vm_start < vma->vm_end)
1757 vma_link(mm, vma, prev, rb_link, rb_parent);
1761 * Copy the vma structure to a new location in the same mm,
1762 * prior to moving page table entries, to effect an mremap move.
1764 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
1765 unsigned long addr, unsigned long len, pgoff_t pgoff)
1767 struct vm_area_struct *vma = *vmap;
1768 unsigned long vma_start = vma->vm_start;
1769 struct mm_struct *mm = vma->vm_mm;
1770 struct vm_area_struct *new_vma, *prev;
1771 struct rb_node **rb_link, *rb_parent;
1772 struct mempolicy *pol;
1775 * If anonymous vma has not yet been faulted, update new pgoff
1776 * to match new location, to increase its chance of merging.
1778 if (!vma->vm_file && !vma->anon_vma)
1779 pgoff = addr >> PAGE_SHIFT;
1781 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1782 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
1783 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
1786 * Source vma may have been merged into new_vma
1788 if (vma_start >= new_vma->vm_start &&
1789 vma_start < new_vma->vm_end)
1792 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1795 pol = mpol_copy(vma_policy(vma));
1797 kmem_cache_free(vm_area_cachep, new_vma);
1800 vma_set_policy(new_vma, pol);
1801 new_vma->vm_start = addr;
1802 new_vma->vm_end = addr + len;
1803 new_vma->vm_pgoff = pgoff;
1804 if (new_vma->vm_file)
1805 get_file(new_vma->vm_file);
1806 if (new_vma->vm_ops && new_vma->vm_ops->open)
1807 new_vma->vm_ops->open(new_vma);
1808 vma_link(mm, new_vma, prev, rb_link, rb_parent);