6 * Address space accounting code <alan@redhat.com>
9 #include <linux/slab.h>
11 #include <linux/shm.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swap.h>
15 #include <linux/syscalls.h>
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/file.h>
20 #include <linux/personality.h>
21 #include <linux/security.h>
22 #include <linux/hugetlb.h>
23 #include <linux/profile.h>
24 #include <linux/module.h>
25 #include <linux/mount.h>
26 #include <linux/mempolicy.h>
27 #include <linux/rmap.h>
28 #include <linux/random.h>
30 #include <asm/uaccess.h>
31 #include <asm/cacheflush.h>
34 #ifndef arch_mmap_check
35 #define arch_mmap_check(addr, len, flags) (0)
38 static void unmap_region(struct mm_struct *mm,
39 struct vm_area_struct *vma, struct vm_area_struct *prev,
40 unsigned long start, unsigned long end);
43 * WARNING: the debugging will use recursive algorithms so never enable this
44 * unless you know what you are doing.
48 /* description of effects of mapping type and prot in current implementation.
49 * this is due to the limited x86 page protection hardware. The expected
50 * behavior is in parens:
53 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
54 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
55 * w: (no) no w: (no) no w: (yes) yes w: (no) no
56 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
58 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
59 * w: (no) no w: (no) no w: (copy) copy w: (no) no
60 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
63 pgprot_t protection_map[16] = {
64 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
65 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
68 pgprot_t vm_get_page_prot(unsigned long vm_flags)
70 return protection_map[vm_flags &
71 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
73 EXPORT_SYMBOL(vm_get_page_prot);
75 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
76 int sysctl_overcommit_ratio = 50; /* default is 50% */
77 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
78 atomic_t vm_committed_space = ATOMIC_INIT(0);
81 * Check that a process has enough memory to allocate a new virtual
82 * mapping. 0 means there is enough memory for the allocation to
83 * succeed and -ENOMEM implies there is not.
85 * We currently support three overcommit policies, which are set via the
86 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
88 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
89 * Additional code 2002 Jul 20 by Robert Love.
91 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
93 * Note this is a helper function intended to be used by LSMs which
94 * wish to use this logic.
96 int __vm_enough_memory(long pages, int cap_sys_admin)
98 unsigned long free, allowed;
100 vm_acct_memory(pages);
103 * Sometimes we want to use more memory than we have
105 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
108 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
111 free = global_page_state(NR_FILE_PAGES);
112 free += nr_swap_pages;
115 * Any slabs which are created with the
116 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
117 * which are reclaimable, under pressure. The dentry
118 * cache and most inode caches should fall into this
120 free += atomic_read(&slab_reclaim_pages);
123 * Leave the last 3% for root
132 * nr_free_pages() is very expensive on large systems,
133 * only call if we're about to fail.
138 * Leave reserved pages. The pages are not for anonymous pages.
140 if (n <= totalreserve_pages)
143 n -= totalreserve_pages;
146 * Leave the last 3% for root
158 allowed = (totalram_pages - hugetlb_total_pages())
159 * sysctl_overcommit_ratio / 100;
161 * Leave the last 3% for root
164 allowed -= allowed / 32;
165 allowed += total_swap_pages;
167 /* Don't let a single process grow too big:
168 leave 3% of the size of this process for other processes */
169 allowed -= current->mm->total_vm / 32;
172 * cast `allowed' as a signed long because vm_committed_space
173 * sometimes has a negative value
175 if (atomic_read(&vm_committed_space) < (long)allowed)
178 vm_unacct_memory(pages);
183 EXPORT_SYMBOL(__vm_enough_memory);
186 * Requires inode->i_mapping->i_mmap_lock
188 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
189 struct file *file, struct address_space *mapping)
191 if (vma->vm_flags & VM_DENYWRITE)
192 atomic_inc(&file->f_dentry->d_inode->i_writecount);
193 if (vma->vm_flags & VM_SHARED)
194 mapping->i_mmap_writable--;
196 flush_dcache_mmap_lock(mapping);
197 if (unlikely(vma->vm_flags & VM_NONLINEAR))
198 list_del_init(&vma->shared.vm_set.list);
200 vma_prio_tree_remove(vma, &mapping->i_mmap);
201 flush_dcache_mmap_unlock(mapping);
205 * Unlink a file-based vm structure from its prio_tree, to hide
206 * vma from rmap and vmtruncate before freeing its page tables.
208 void unlink_file_vma(struct vm_area_struct *vma)
210 struct file *file = vma->vm_file;
213 struct address_space *mapping = file->f_mapping;
214 spin_lock(&mapping->i_mmap_lock);
215 __remove_shared_vm_struct(vma, file, mapping);
216 spin_unlock(&mapping->i_mmap_lock);
221 * Close a vm structure and free it, returning the next.
223 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
225 struct vm_area_struct *next = vma->vm_next;
228 if (vma->vm_ops && vma->vm_ops->close)
229 vma->vm_ops->close(vma);
232 mpol_free(vma_policy(vma));
233 kmem_cache_free(vm_area_cachep, vma);
237 asmlinkage unsigned long sys_brk(unsigned long brk)
239 unsigned long rlim, retval;
240 unsigned long newbrk, oldbrk;
241 struct mm_struct *mm = current->mm;
243 down_write(&mm->mmap_sem);
245 if (brk < mm->end_code)
249 * Check against rlimit here. If this check is done later after the test
250 * of oldbrk with newbrk then it can escape the test and let the data
251 * segment grow beyond its set limit the in case where the limit is
252 * not page aligned -Ram Gupta
254 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
255 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
258 newbrk = PAGE_ALIGN(brk);
259 oldbrk = PAGE_ALIGN(mm->brk);
260 if (oldbrk == newbrk)
263 /* Always allow shrinking brk. */
264 if (brk <= mm->brk) {
265 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
270 /* Check against existing mmap mappings. */
271 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
274 /* Ok, looks good - let it rip. */
275 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
281 up_write(&mm->mmap_sem);
286 static int browse_rb(struct rb_root *root)
289 struct rb_node *nd, *pn = NULL;
290 unsigned long prev = 0, pend = 0;
292 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
293 struct vm_area_struct *vma;
294 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
295 if (vma->vm_start < prev)
296 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
297 if (vma->vm_start < pend)
298 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
299 if (vma->vm_start > vma->vm_end)
300 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
305 for (nd = pn; nd; nd = rb_prev(nd)) {
309 printk("backwards %d, forwards %d\n", j, i), i = 0;
313 void validate_mm(struct mm_struct *mm)
317 struct vm_area_struct *tmp = mm->mmap;
322 if (i != mm->map_count)
323 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
324 i = browse_rb(&mm->mm_rb);
325 if (i != mm->map_count)
326 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
330 #define validate_mm(mm) do { } while (0)
333 static struct vm_area_struct *
334 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
335 struct vm_area_struct **pprev, struct rb_node ***rb_link,
336 struct rb_node ** rb_parent)
338 struct vm_area_struct * vma;
339 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
341 __rb_link = &mm->mm_rb.rb_node;
342 rb_prev = __rb_parent = NULL;
346 struct vm_area_struct *vma_tmp;
348 __rb_parent = *__rb_link;
349 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
351 if (vma_tmp->vm_end > addr) {
353 if (vma_tmp->vm_start <= addr)
355 __rb_link = &__rb_parent->rb_left;
357 rb_prev = __rb_parent;
358 __rb_link = &__rb_parent->rb_right;
364 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
365 *rb_link = __rb_link;
366 *rb_parent = __rb_parent;
371 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
372 struct vm_area_struct *prev, struct rb_node *rb_parent)
374 if (vma->vm_flags & VM_EXEC)
375 arch_add_exec_range(mm, vma->vm_end);
377 vma->vm_next = prev->vm_next;
382 vma->vm_next = rb_entry(rb_parent,
383 struct vm_area_struct, vm_rb);
389 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
390 struct rb_node **rb_link, struct rb_node *rb_parent)
392 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
393 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
396 static inline void __vma_link_file(struct vm_area_struct *vma)
402 struct address_space *mapping = file->f_mapping;
404 if (vma->vm_flags & VM_DENYWRITE)
405 atomic_dec(&file->f_dentry->d_inode->i_writecount);
406 if (vma->vm_flags & VM_SHARED)
407 mapping->i_mmap_writable++;
409 flush_dcache_mmap_lock(mapping);
410 if (unlikely(vma->vm_flags & VM_NONLINEAR))
411 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
413 vma_prio_tree_insert(vma, &mapping->i_mmap);
414 flush_dcache_mmap_unlock(mapping);
419 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
420 struct vm_area_struct *prev, struct rb_node **rb_link,
421 struct rb_node *rb_parent)
423 __vma_link_list(mm, vma, prev, rb_parent);
424 __vma_link_rb(mm, vma, rb_link, rb_parent);
425 __anon_vma_link(vma);
428 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
429 struct vm_area_struct *prev, struct rb_node **rb_link,
430 struct rb_node *rb_parent)
432 struct address_space *mapping = NULL;
435 mapping = vma->vm_file->f_mapping;
438 spin_lock(&mapping->i_mmap_lock);
439 vma->vm_truncate_count = mapping->truncate_count;
443 __vma_link(mm, vma, prev, rb_link, rb_parent);
444 __vma_link_file(vma);
446 anon_vma_unlock(vma);
448 spin_unlock(&mapping->i_mmap_lock);
455 * Helper for vma_adjust in the split_vma insert case:
456 * insert vm structure into list and rbtree and anon_vma,
457 * but it has already been inserted into prio_tree earlier.
460 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
462 struct vm_area_struct * __vma, * prev;
463 struct rb_node ** rb_link, * rb_parent;
465 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
466 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
467 __vma_link(mm, vma, prev, rb_link, rb_parent);
472 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
473 struct vm_area_struct *prev)
475 prev->vm_next = vma->vm_next;
476 rb_erase(&vma->vm_rb, &mm->mm_rb);
477 if (mm->mmap_cache == vma)
478 mm->mmap_cache = prev;
479 if (vma->vm_flags & VM_EXEC)
480 arch_remove_exec_range(mm, vma->vm_end);
484 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
485 * is already present in an i_mmap tree without adjusting the tree.
486 * The following helper function should be used when such adjustments
487 * are necessary. The "insert" vma (if any) is to be inserted
488 * before we drop the necessary locks.
490 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
491 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
493 struct mm_struct *mm = vma->vm_mm;
494 struct vm_area_struct *next = vma->vm_next;
495 struct vm_area_struct *importer = NULL;
496 struct address_space *mapping = NULL;
497 struct prio_tree_root *root = NULL;
498 struct file *file = vma->vm_file;
499 struct anon_vma *anon_vma = NULL;
500 long adjust_next = 0;
503 if (next && !insert) {
504 if (end >= next->vm_end) {
506 * vma expands, overlapping all the next, and
507 * perhaps the one after too (mprotect case 6).
509 again: remove_next = 1 + (end > next->vm_end);
511 anon_vma = next->anon_vma;
513 } else if (end > next->vm_start) {
515 * vma expands, overlapping part of the next:
516 * mprotect case 5 shifting the boundary up.
518 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
519 anon_vma = next->anon_vma;
521 } else if (end < vma->vm_end) {
523 * vma shrinks, and !insert tells it's not
524 * split_vma inserting another: so it must be
525 * mprotect case 4 shifting the boundary down.
527 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
528 anon_vma = next->anon_vma;
534 mapping = file->f_mapping;
535 if (!(vma->vm_flags & VM_NONLINEAR))
536 root = &mapping->i_mmap;
537 spin_lock(&mapping->i_mmap_lock);
539 vma->vm_truncate_count != next->vm_truncate_count) {
541 * unmap_mapping_range might be in progress:
542 * ensure that the expanding vma is rescanned.
544 importer->vm_truncate_count = 0;
547 insert->vm_truncate_count = vma->vm_truncate_count;
549 * Put into prio_tree now, so instantiated pages
550 * are visible to arm/parisc __flush_dcache_page
551 * throughout; but we cannot insert into address
552 * space until vma start or end is updated.
554 __vma_link_file(insert);
559 * When changing only vma->vm_end, we don't really need
560 * anon_vma lock: but is that case worth optimizing out?
563 anon_vma = vma->anon_vma;
565 spin_lock(&anon_vma->lock);
567 * Easily overlooked: when mprotect shifts the boundary,
568 * make sure the expanding vma has anon_vma set if the
569 * shrinking vma had, to cover any anon pages imported.
571 if (importer && !importer->anon_vma) {
572 importer->anon_vma = anon_vma;
573 __anon_vma_link(importer);
578 flush_dcache_mmap_lock(mapping);
579 vma_prio_tree_remove(vma, root);
581 vma_prio_tree_remove(next, root);
584 vma->vm_start = start;
586 vma->vm_pgoff = pgoff;
588 next->vm_start += adjust_next << PAGE_SHIFT;
589 next->vm_pgoff += adjust_next;
594 vma_prio_tree_insert(next, root);
595 vma_prio_tree_insert(vma, root);
596 flush_dcache_mmap_unlock(mapping);
601 * vma_merge has merged next into vma, and needs
602 * us to remove next before dropping the locks.
604 __vma_unlink(mm, next, vma);
606 __remove_shared_vm_struct(next, file, mapping);
608 __anon_vma_merge(vma, next);
611 * split_vma has split insert from vma, and needs
612 * us to insert it before dropping the locks
613 * (it may either follow vma or precede it).
615 __insert_vm_struct(mm, insert);
619 spin_unlock(&anon_vma->lock);
621 spin_unlock(&mapping->i_mmap_lock);
627 mpol_free(vma_policy(next));
628 kmem_cache_free(vm_area_cachep, next);
630 * In mprotect's case 6 (see comments on vma_merge),
631 * we must remove another next too. It would clutter
632 * up the code too much to do both in one go.
634 if (remove_next == 2) {
644 * If the vma has a ->close operation then the driver probably needs to release
645 * per-vma resources, so we don't attempt to merge those.
647 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
649 static inline int is_mergeable_vma(struct vm_area_struct *vma,
650 struct file *file, unsigned long vm_flags)
652 if (vma->vm_flags != vm_flags)
654 if (vma->vm_file != file)
656 if (vma->vm_ops && vma->vm_ops->close)
661 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
662 struct anon_vma *anon_vma2)
664 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
668 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
669 * in front of (at a lower virtual address and file offset than) the vma.
671 * We cannot merge two vmas if they have differently assigned (non-NULL)
672 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
674 * We don't check here for the merged mmap wrapping around the end of pagecache
675 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
676 * wrap, nor mmaps which cover the final page at index -1UL.
679 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
680 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
682 if (is_mergeable_vma(vma, file, vm_flags) &&
683 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
684 if (vma->vm_pgoff == vm_pgoff)
691 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
692 * beyond (at a higher virtual address and file offset than) the vma.
694 * We cannot merge two vmas if they have differently assigned (non-NULL)
695 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
698 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
699 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
701 if (is_mergeable_vma(vma, file, vm_flags) &&
702 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
704 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
705 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
712 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
713 * whether that can be merged with its predecessor or its successor.
714 * Or both (it neatly fills a hole).
716 * In most cases - when called for mmap, brk or mremap - [addr,end) is
717 * certain not to be mapped by the time vma_merge is called; but when
718 * called for mprotect, it is certain to be already mapped (either at
719 * an offset within prev, or at the start of next), and the flags of
720 * this area are about to be changed to vm_flags - and the no-change
721 * case has already been eliminated.
723 * The following mprotect cases have to be considered, where AAAA is
724 * the area passed down from mprotect_fixup, never extending beyond one
725 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
727 * AAAA AAAA AAAA AAAA
728 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
729 * cannot merge might become might become might become
730 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
731 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
732 * mremap move: PPPPNNNNNNNN 8
734 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
735 * might become case 1 below case 2 below case 3 below
737 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
738 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
740 struct vm_area_struct *vma_merge(struct mm_struct *mm,
741 struct vm_area_struct *prev, unsigned long addr,
742 unsigned long end, unsigned long vm_flags,
743 struct anon_vma *anon_vma, struct file *file,
744 pgoff_t pgoff, struct mempolicy *policy)
746 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
747 struct vm_area_struct *area, *next;
750 * We later require that vma->vm_flags == vm_flags,
751 * so this tests vma->vm_flags & VM_SPECIAL, too.
753 if (vm_flags & VM_SPECIAL)
757 next = prev->vm_next;
761 if (next && next->vm_end == end) /* cases 6, 7, 8 */
762 next = next->vm_next;
765 * Can it merge with the predecessor?
767 if (prev && prev->vm_end == addr &&
768 mpol_equal(vma_policy(prev), policy) &&
769 can_vma_merge_after(prev, vm_flags,
770 anon_vma, file, pgoff)) {
772 * OK, it can. Can we now merge in the successor as well?
774 if (next && end == next->vm_start &&
775 mpol_equal(policy, vma_policy(next)) &&
776 can_vma_merge_before(next, vm_flags,
777 anon_vma, file, pgoff+pglen) &&
778 is_mergeable_anon_vma(prev->anon_vma,
781 vma_adjust(prev, prev->vm_start,
782 next->vm_end, prev->vm_pgoff, NULL);
783 } else /* cases 2, 5, 7 */
784 vma_adjust(prev, prev->vm_start,
785 end, prev->vm_pgoff, NULL);
786 if (prev->vm_flags & VM_EXEC)
787 arch_add_exec_range(mm, prev->vm_end);
792 * Can this new request be merged in front of next?
794 if (next && end == next->vm_start &&
795 mpol_equal(policy, vma_policy(next)) &&
796 can_vma_merge_before(next, vm_flags,
797 anon_vma, file, pgoff+pglen)) {
798 if (prev && addr < prev->vm_end) /* case 4 */
799 vma_adjust(prev, prev->vm_start,
800 addr, prev->vm_pgoff, NULL);
801 else /* cases 3, 8 */
802 vma_adjust(area, addr, next->vm_end,
803 next->vm_pgoff - pglen, NULL);
811 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
812 * neighbouring vmas for a suitable anon_vma, before it goes off
813 * to allocate a new anon_vma. It checks because a repetitive
814 * sequence of mprotects and faults may otherwise lead to distinct
815 * anon_vmas being allocated, preventing vma merge in subsequent
818 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
820 struct vm_area_struct *near;
821 unsigned long vm_flags;
828 * Since only mprotect tries to remerge vmas, match flags
829 * which might be mprotected into each other later on.
830 * Neither mlock nor madvise tries to remerge at present,
831 * so leave their flags as obstructing a merge.
833 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
834 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
836 if (near->anon_vma && vma->vm_end == near->vm_start &&
837 mpol_equal(vma_policy(vma), vma_policy(near)) &&
838 can_vma_merge_before(near, vm_flags,
839 NULL, vma->vm_file, vma->vm_pgoff +
840 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
841 return near->anon_vma;
844 * It is potentially slow to have to call find_vma_prev here.
845 * But it's only on the first write fault on the vma, not
846 * every time, and we could devise a way to avoid it later
847 * (e.g. stash info in next's anon_vma_node when assigning
848 * an anon_vma, or when trying vma_merge). Another time.
850 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
854 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
855 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
857 if (near->anon_vma && near->vm_end == vma->vm_start &&
858 mpol_equal(vma_policy(near), vma_policy(vma)) &&
859 can_vma_merge_after(near, vm_flags,
860 NULL, vma->vm_file, vma->vm_pgoff))
861 return near->anon_vma;
864 * There's no absolute need to look only at touching neighbours:
865 * we could search further afield for "compatible" anon_vmas.
866 * But it would probably just be a waste of time searching,
867 * or lead to too many vmas hanging off the same anon_vma.
868 * We're trying to allow mprotect remerging later on,
869 * not trying to minimize memory used for anon_vmas.
874 #ifdef CONFIG_PROC_FS
875 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
876 struct file *file, long pages)
878 const unsigned long stack_flags
879 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
882 mm->shared_vm += pages;
883 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
884 mm->exec_vm += pages;
885 } else if (flags & stack_flags)
886 mm->stack_vm += pages;
887 if (flags & (VM_RESERVED|VM_IO))
888 mm->reserved_vm += pages;
890 #endif /* CONFIG_PROC_FS */
893 * The caller must hold down_write(current->mm->mmap_sem).
896 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
897 unsigned long len, unsigned long prot,
898 unsigned long flags, unsigned long pgoff)
900 struct mm_struct * mm = current->mm;
901 struct vm_area_struct * vma, * prev;
903 unsigned int vm_flags;
904 int correct_wcount = 0;
906 struct rb_node ** rb_link, * rb_parent;
908 unsigned long charged = 0, reqprot = prot;
911 if (is_file_hugepages(file))
914 if (!file->f_op || !file->f_op->mmap)
917 if ((prot & PROT_EXEC) &&
918 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
922 * Does the application expect PROT_READ to imply PROT_EXEC?
924 * (the exception is when the underlying filesystem is noexec
925 * mounted, in which case we dont add PROT_EXEC.)
927 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
928 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
934 error = arch_mmap_check(addr, len, flags);
938 /* Careful about overflows.. */
939 len = PAGE_ALIGN(len);
940 if (!len || len > TASK_SIZE)
943 /* offset overflow? */
944 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
947 /* Too many mappings? */
948 if (mm->map_count > sysctl_max_map_count)
951 /* Obtain the address to map to. we verify (or select) it and ensure
952 * that it represents a valid section of the address space.
954 addr = get_unmapped_area_prot(file, addr, len, pgoff, flags, prot & PROT_EXEC);
955 if (addr & ~PAGE_MASK)
958 /* Do simple checking here so the lower-level routines won't have
959 * to. we assume access permissions have been handled by the open
960 * of the memory object, so we don't do any here.
962 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
963 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
965 if (flags & MAP_LOCKED) {
968 vm_flags |= VM_LOCKED;
970 /* mlock MCL_FUTURE? */
971 if (vm_flags & VM_LOCKED) {
972 unsigned long locked, lock_limit;
973 locked = len >> PAGE_SHIFT;
974 locked += mm->locked_vm;
975 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
976 lock_limit >>= PAGE_SHIFT;
977 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
981 inode = file ? file->f_dentry->d_inode : NULL;
984 switch (flags & MAP_TYPE) {
986 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
990 * Make sure we don't allow writing to an append-only
993 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
997 * Make sure there are no mandatory locks on the file.
999 if (locks_verify_locked(inode))
1002 vm_flags |= VM_SHARED | VM_MAYSHARE;
1003 if (!(file->f_mode & FMODE_WRITE))
1004 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1008 if (!(file->f_mode & FMODE_READ))
1016 switch (flags & MAP_TYPE) {
1018 vm_flags |= VM_SHARED | VM_MAYSHARE;
1022 * Set pgoff according to addr for anon_vma.
1024 pgoff = addr >> PAGE_SHIFT;
1031 error = security_file_mmap(file, reqprot, prot, flags);
1035 /* Clear old maps */
1038 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1039 if (vma && vma->vm_start < addr + len) {
1040 if (do_munmap(mm, addr, len))
1045 /* Check against address space limit. */
1046 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1049 if (accountable && (!(flags & MAP_NORESERVE) ||
1050 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1051 if (vm_flags & VM_SHARED) {
1052 /* Check memory availability in shmem_file_setup? */
1053 vm_flags |= VM_ACCOUNT;
1054 } else if (vm_flags & VM_WRITE) {
1056 * Private writable mapping: check memory availability
1058 charged = len >> PAGE_SHIFT;
1059 if (security_vm_enough_memory(charged))
1061 vm_flags |= VM_ACCOUNT;
1066 * Can we just expand an old private anonymous mapping?
1067 * The VM_SHARED test is necessary because shmem_zero_setup
1068 * will create the file object for a shared anonymous map below.
1070 if (!file && !(vm_flags & VM_SHARED) &&
1071 vma_merge(mm, prev, addr, addr + len, vm_flags,
1072 NULL, NULL, pgoff, NULL))
1076 * Determine the object being mapped and call the appropriate
1077 * specific mapper. the address has already been validated, but
1078 * not unmapped, but the maps are removed from the list.
1080 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1087 vma->vm_start = addr;
1088 vma->vm_end = addr + len;
1089 vma->vm_flags = vm_flags;
1090 vma->vm_page_prot = protection_map[vm_flags &
1091 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
1092 vma->vm_pgoff = pgoff;
1096 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1098 if (vm_flags & VM_DENYWRITE) {
1099 error = deny_write_access(file);
1104 vma->vm_file = file;
1106 error = file->f_op->mmap(file, vma);
1108 goto unmap_and_free_vma;
1109 } else if (vm_flags & VM_SHARED) {
1110 error = shmem_zero_setup(vma);
1115 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1116 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1117 * that memory reservation must be checked; but that reservation
1118 * belongs to shared memory object, not to vma: so now clear it.
1120 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1121 vma->vm_flags &= ~VM_ACCOUNT;
1123 /* Can addr have changed??
1125 * Answer: Yes, several device drivers can do it in their
1126 * f_op->mmap method. -DaveM
1128 addr = vma->vm_start;
1129 pgoff = vma->vm_pgoff;
1130 vm_flags = vma->vm_flags;
1132 if (vma_wants_writenotify(vma))
1134 protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC)];
1136 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1137 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1138 file = vma->vm_file;
1139 vma_link(mm, vma, prev, rb_link, rb_parent);
1141 atomic_inc(&inode->i_writecount);
1145 atomic_inc(&inode->i_writecount);
1148 mpol_free(vma_policy(vma));
1149 kmem_cache_free(vm_area_cachep, vma);
1152 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1153 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1154 if (vm_flags & VM_LOCKED) {
1155 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1156 make_pages_present(addr, addr + len);
1158 if (flags & MAP_POPULATE) {
1159 up_write(&mm->mmap_sem);
1160 sys_remap_file_pages(addr, len, 0,
1161 pgoff, flags & MAP_NONBLOCK);
1162 down_write(&mm->mmap_sem);
1168 atomic_inc(&inode->i_writecount);
1169 vma->vm_file = NULL;
1172 /* Undo any partial mapping done by a device driver. */
1173 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1176 kmem_cache_free(vm_area_cachep, vma);
1179 vm_unacct_memory(charged);
1183 EXPORT_SYMBOL(do_mmap_pgoff);
1185 /* Get an address range which is currently unmapped.
1186 * For shmat() with addr=0.
1188 * Ugly calling convention alert:
1189 * Return value with the low bits set means error value,
1191 * if (ret & ~PAGE_MASK)
1194 * This function "knows" that -ENOMEM has the bits set.
1196 #ifndef HAVE_ARCH_UNMAPPED_AREA
1198 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1199 unsigned long len, unsigned long pgoff, unsigned long flags)
1201 struct mm_struct *mm = current->mm;
1202 struct vm_area_struct *vma;
1203 unsigned long start_addr;
1205 if (len > TASK_SIZE)
1209 addr = PAGE_ALIGN(addr);
1210 vma = find_vma(mm, addr);
1211 if (TASK_SIZE - len >= addr &&
1212 (!vma || addr + len <= vma->vm_start))
1215 if (len > mm->cached_hole_size) {
1216 start_addr = addr = mm->free_area_cache;
1218 start_addr = addr = TASK_UNMAPPED_BASE;
1219 mm->cached_hole_size = 0;
1223 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1224 /* At this point: (!vma || addr < vma->vm_end). */
1225 if (TASK_SIZE - len < addr) {
1227 * Start a new search - just in case we missed
1230 if (start_addr != TASK_UNMAPPED_BASE) {
1231 addr = TASK_UNMAPPED_BASE;
1233 mm->cached_hole_size = 0;
1238 if (!vma || addr + len <= vma->vm_start) {
1240 * Remember the place where we stopped the search:
1242 mm->free_area_cache = addr + len;
1245 if (addr + mm->cached_hole_size < vma->vm_start)
1246 mm->cached_hole_size = vma->vm_start - addr;
1252 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1255 * Is this a new hole at the lowest possible address?
1257 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1258 mm->free_area_cache = addr;
1259 mm->cached_hole_size = ~0UL;
1264 * This mmap-allocator allocates new areas top-down from below the
1265 * stack's low limit (the base):
1267 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1269 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1270 const unsigned long len, const unsigned long pgoff,
1271 const unsigned long flags)
1273 struct vm_area_struct *vma;
1274 struct mm_struct *mm = current->mm;
1275 unsigned long addr = addr0;
1277 /* requested length too big for entire address space */
1278 if (len > TASK_SIZE)
1281 /* requesting a specific address */
1283 addr = PAGE_ALIGN(addr);
1284 vma = find_vma(mm, addr);
1285 if (TASK_SIZE - len >= addr &&
1286 (!vma || addr + len <= vma->vm_start))
1290 /* check if free_area_cache is useful for us */
1291 if (len <= mm->cached_hole_size) {
1292 mm->cached_hole_size = 0;
1293 mm->free_area_cache = mm->mmap_base;
1296 /* either no address requested or can't fit in requested address hole */
1297 addr = mm->free_area_cache;
1299 /* make sure it can fit in the remaining address space */
1301 vma = find_vma(mm, addr-len);
1302 if (!vma || addr <= vma->vm_start)
1303 /* remember the address as a hint for next time */
1304 return (mm->free_area_cache = addr-len);
1307 if (mm->mmap_base < len)
1310 addr = mm->mmap_base-len;
1314 * Lookup failure means no vma is above this address,
1315 * else if new region fits below vma->vm_start,
1316 * return with success:
1318 vma = find_vma(mm, addr);
1319 if (!vma || addr+len <= vma->vm_start)
1320 /* remember the address as a hint for next time */
1321 return (mm->free_area_cache = addr);
1323 /* remember the largest hole we saw so far */
1324 if (addr + mm->cached_hole_size < vma->vm_start)
1325 mm->cached_hole_size = vma->vm_start - addr;
1327 /* try just below the current vma->vm_start */
1328 addr = vma->vm_start-len;
1329 } while (len < vma->vm_start);
1333 * A failed mmap() very likely causes application failure,
1334 * so fall back to the bottom-up function here. This scenario
1335 * can happen with large stack limits and large mmap()
1338 mm->cached_hole_size = ~0UL;
1339 mm->free_area_cache = TASK_UNMAPPED_BASE;
1340 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1342 * Restore the topdown base:
1344 mm->free_area_cache = mm->mmap_base;
1345 mm->cached_hole_size = ~0UL;
1351 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1354 * Is this a new hole at the highest possible address?
1356 if (addr > mm->free_area_cache)
1357 mm->free_area_cache = addr;
1359 /* dont allow allocations above current base */
1360 if (mm->free_area_cache > mm->mmap_base)
1361 mm->free_area_cache = mm->mmap_base;
1366 get_unmapped_area_prot(struct file *file, unsigned long addr, unsigned long len,
1367 unsigned long pgoff, unsigned long flags, int exec)
1371 if (!(flags & MAP_FIXED)) {
1372 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1374 if (exec && current->mm->get_unmapped_exec_area)
1375 get_area = current->mm->get_unmapped_exec_area;
1377 get_area = current->mm->get_unmapped_area;
1379 if (file && file->f_op && file->f_op->get_unmapped_area)
1380 get_area = file->f_op->get_unmapped_area;
1381 addr = get_area(file, addr, len, pgoff, flags);
1382 if (IS_ERR_VALUE(addr))
1386 if (addr > TASK_SIZE - len)
1388 if (addr & ~PAGE_MASK)
1390 if (file && is_file_hugepages(file)) {
1392 * Check if the given range is hugepage aligned, and
1393 * can be made suitable for hugepages.
1395 ret = prepare_hugepage_range(addr, len);
1398 * Ensure that a normal request is not falling in a
1399 * reserved hugepage range. For some archs like IA-64,
1400 * there is a separate region for hugepages.
1402 ret = is_hugepage_only_range(current->mm, addr, len);
1409 EXPORT_SYMBOL(get_unmapped_area_prot);
1411 #define SHLIB_BASE 0x00110000
1413 unsigned long arch_get_unmapped_exec_area(struct file *filp, unsigned long addr0,
1414 unsigned long len0, unsigned long pgoff, unsigned long flags)
1416 unsigned long addr = addr0, len = len0;
1417 struct mm_struct *mm = current->mm;
1418 struct vm_area_struct *vma;
1421 if (len > TASK_SIZE)
1424 if (!addr && !(flags & MAP_FIXED))
1425 addr = randomize_range(SHLIB_BASE, 0x01000000, len);
1428 addr = PAGE_ALIGN(addr);
1429 vma = find_vma(mm, addr);
1430 if (TASK_SIZE - len >= addr &&
1431 (!vma || addr + len <= vma->vm_start)) {
1437 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1438 /* At this point: (!vma || addr < vma->vm_end). */
1439 if (TASK_SIZE - len < addr)
1442 if (!vma || addr + len <= vma->vm_start) {
1444 * Must not let a PROT_EXEC mapping get into the
1447 if (addr + len > mm->brk)
1451 * Up until the brk area we randomize addresses
1452 * as much as possible:
1454 if (addr >= 0x01000000) {
1455 tmp = randomize_range(0x01000000, PAGE_ALIGN(max(mm->start_brk, (unsigned long)0x08000000)), len);
1456 vma = find_vma(mm, tmp);
1457 if (TASK_SIZE - len >= tmp &&
1458 (!vma || tmp + len <= vma->vm_start))
1462 * Ok, randomization didnt work out - return
1463 * the result of the linear search:
1471 return current->mm->get_unmapped_area(filp, addr0, len0, pgoff, flags);
1475 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1476 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1478 struct vm_area_struct *vma = NULL;
1481 /* Check the cache first. */
1482 /* (Cache hit rate is typically around 35%.) */
1483 vma = mm->mmap_cache;
1484 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1485 struct rb_node * rb_node;
1487 rb_node = mm->mm_rb.rb_node;
1491 struct vm_area_struct * vma_tmp;
1493 vma_tmp = rb_entry(rb_node,
1494 struct vm_area_struct, vm_rb);
1496 if (vma_tmp->vm_end > addr) {
1498 if (vma_tmp->vm_start <= addr)
1500 rb_node = rb_node->rb_left;
1502 rb_node = rb_node->rb_right;
1505 mm->mmap_cache = vma;
1511 EXPORT_SYMBOL(find_vma);
1513 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1514 struct vm_area_struct *
1515 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1516 struct vm_area_struct **pprev)
1518 struct vm_area_struct *vma = NULL, *prev = NULL;
1519 struct rb_node * rb_node;
1523 /* Guard against addr being lower than the first VMA */
1526 /* Go through the RB tree quickly. */
1527 rb_node = mm->mm_rb.rb_node;
1530 struct vm_area_struct *vma_tmp;
1531 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1533 if (addr < vma_tmp->vm_end) {
1534 rb_node = rb_node->rb_left;
1537 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1539 rb_node = rb_node->rb_right;
1545 return prev ? prev->vm_next : vma;
1548 static int over_stack_limit(unsigned long sz)
1550 if (sz < EXEC_STACK_BIAS)
1552 return (sz - EXEC_STACK_BIAS) >
1553 current->signal->rlim[RLIMIT_STACK].rlim_cur;
1557 * Verify that the stack growth is acceptable and
1558 * update accounting. This is shared with both the
1559 * grow-up and grow-down cases.
1561 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1563 struct mm_struct *mm = vma->vm_mm;
1564 struct rlimit *rlim = current->signal->rlim;
1566 /* address space limit tests */
1567 if (!may_expand_vm(mm, grow))
1570 /* Stack limit test */
1571 if (over_stack_limit(size))
1574 /* mlock limit tests */
1575 if (vma->vm_flags & VM_LOCKED) {
1576 unsigned long locked;
1577 unsigned long limit;
1578 locked = mm->locked_vm + grow;
1579 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1580 if (locked > limit && !capable(CAP_IPC_LOCK))
1585 * Overcommit.. This must be the final test, as it will
1586 * update security statistics.
1588 if (security_vm_enough_memory(grow))
1591 /* Ok, everything looks good - let it rip */
1592 vx_vmpages_add(mm, grow);
1593 if (vma->vm_flags & VM_LOCKED)
1594 vx_vmlocked_add(mm, grow);
1595 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1599 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1601 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1602 * vma is the last one with address > vma->vm_end. Have to extend vma.
1607 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1611 if (!(vma->vm_flags & VM_GROWSUP))
1615 * We must make sure the anon_vma is allocated
1616 * so that the anon_vma locking is not a noop.
1618 if (unlikely(anon_vma_prepare(vma)))
1623 * vma->vm_start/vm_end cannot change under us because the caller
1624 * is required to hold the mmap_sem in read mode. We need the
1625 * anon_vma lock to serialize against concurrent expand_stacks.
1627 address += 4 + PAGE_SIZE - 1;
1628 address &= PAGE_MASK;
1631 /* Somebody else might have raced and expanded it already */
1632 if (address > vma->vm_end) {
1633 unsigned long size, grow;
1635 size = address - vma->vm_start;
1636 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1638 error = acct_stack_growth(vma, size, grow);
1640 vma->vm_end = address;
1642 anon_vma_unlock(vma);
1645 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1647 #ifdef CONFIG_STACK_GROWSUP
1648 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1650 return expand_upwards(vma, address);
1653 struct vm_area_struct *
1654 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1656 struct vm_area_struct *vma, *prev;
1659 vma = find_vma_prev(mm, addr, &prev);
1660 if (vma && (vma->vm_start <= addr))
1662 if (!prev || expand_stack(prev, addr))
1664 if (prev->vm_flags & VM_LOCKED) {
1665 make_pages_present(addr, prev->vm_end);
1671 * vma is the first one with address < vma->vm_start. Have to extend vma.
1673 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1678 * We must make sure the anon_vma is allocated
1679 * so that the anon_vma locking is not a noop.
1681 if (unlikely(anon_vma_prepare(vma)))
1686 * vma->vm_start/vm_end cannot change under us because the caller
1687 * is required to hold the mmap_sem in read mode. We need the
1688 * anon_vma lock to serialize against concurrent expand_stacks.
1690 address &= PAGE_MASK;
1693 /* Somebody else might have raced and expanded it already */
1694 if (address < vma->vm_start) {
1695 unsigned long size, grow;
1697 size = vma->vm_end - address;
1698 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1700 error = acct_stack_growth(vma, size, grow);
1702 vma->vm_start = address;
1703 vma->vm_pgoff -= grow;
1706 anon_vma_unlock(vma);
1710 struct vm_area_struct *
1711 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1713 struct vm_area_struct * vma;
1714 unsigned long start;
1717 vma = find_vma(mm,addr);
1720 if (vma->vm_start <= addr)
1722 if (!(vma->vm_flags & VM_GROWSDOWN))
1724 start = vma->vm_start;
1725 if (expand_stack(vma, addr))
1727 if (vma->vm_flags & VM_LOCKED) {
1728 make_pages_present(addr, start);
1735 * Ok - we have the memory areas we should free on the vma list,
1736 * so release them, and do the vma updates.
1738 * Called with the mm semaphore held.
1740 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1742 /* Update high watermark before we lower total_vm */
1743 update_hiwater_vm(mm);
1745 long nrpages = vma_pages(vma);
1747 vx_vmpages_sub(mm, nrpages);
1748 if (vma->vm_flags & VM_LOCKED)
1749 vx_vmlocked_sub(mm, nrpages);
1750 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1751 vma = remove_vma(vma);
1757 * Get rid of page table information in the indicated region.
1759 * Called with the mm semaphore held.
1761 static void unmap_region(struct mm_struct *mm,
1762 struct vm_area_struct *vma, struct vm_area_struct *prev,
1763 unsigned long start, unsigned long end)
1765 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1766 struct mmu_gather *tlb;
1767 unsigned long nr_accounted = 0;
1770 tlb = tlb_gather_mmu(mm, 0);
1771 update_hiwater_rss(mm);
1772 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1773 vm_unacct_memory(nr_accounted);
1774 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1775 next? next->vm_start: 0);
1776 tlb_finish_mmu(tlb, start, end);
1780 * Create a list of vma's touched by the unmap, removing them from the mm's
1781 * vma list as we go..
1784 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1785 struct vm_area_struct *prev, unsigned long end)
1787 struct vm_area_struct **insertion_point;
1788 struct vm_area_struct *tail_vma = NULL;
1791 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1793 rb_erase(&vma->vm_rb, &mm->mm_rb);
1797 } while (vma && vma->vm_start < end);
1798 *insertion_point = vma;
1799 tail_vma->vm_next = NULL;
1800 if (mm->unmap_area == arch_unmap_area)
1801 addr = prev ? prev->vm_end : mm->mmap_base;
1803 addr = vma ? vma->vm_start : mm->mmap_base;
1804 mm->unmap_area(mm, addr);
1805 mm->mmap_cache = NULL; /* Kill the cache. */
1809 * Split a vma into two pieces at address 'addr', a new vma is allocated
1810 * either for the first part or the the tail.
1812 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1813 unsigned long addr, int new_below)
1815 struct mempolicy *pol;
1816 struct vm_area_struct *new;
1818 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1821 if (mm->map_count >= sysctl_max_map_count)
1824 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1828 /* most fields are the same, copy all, and then fixup */
1834 new->vm_start = addr;
1835 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1838 pol = mpol_copy(vma_policy(vma));
1840 kmem_cache_free(vm_area_cachep, new);
1841 return PTR_ERR(pol);
1843 vma_set_policy(new, pol);
1846 get_file(new->vm_file);
1848 if (new->vm_ops && new->vm_ops->open)
1849 new->vm_ops->open(new);
1852 unsigned long old_end = vma->vm_end;
1854 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1855 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1856 if (vma->vm_flags & VM_EXEC)
1857 arch_remove_exec_range(mm, old_end);
1859 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1864 /* Munmap is split into 2 main parts -- this part which finds
1865 * what needs doing, and the areas themselves, which do the
1866 * work. This now handles partial unmappings.
1867 * Jeremy Fitzhardinge <jeremy@goop.org>
1869 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1872 struct vm_area_struct *vma, *prev, *last;
1874 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1877 if ((len = PAGE_ALIGN(len)) == 0)
1880 /* Find the first overlapping VMA */
1881 vma = find_vma_prev(mm, start, &prev);
1884 /* we have start < vma->vm_end */
1886 /* if it doesn't overlap, we have nothing.. */
1888 if (vma->vm_start >= end)
1892 * If we need to split any vma, do it now to save pain later.
1894 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1895 * unmapped vm_area_struct will remain in use: so lower split_vma
1896 * places tmp vma above, and higher split_vma places tmp vma below.
1898 if (start > vma->vm_start) {
1899 int error = split_vma(mm, vma, start, 0);
1905 /* Does it split the last one? */
1906 last = find_vma(mm, end);
1907 if (last && end > last->vm_start) {
1908 int error = split_vma(mm, last, end, 1);
1912 vma = prev? prev->vm_next: mm->mmap;
1915 * Remove the vma's, and unmap the actual pages
1917 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1918 unmap_region(mm, vma, prev, start, end);
1920 /* Fix up all other VM information */
1921 remove_vma_list(mm, vma);
1926 EXPORT_SYMBOL(do_munmap);
1928 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1931 struct mm_struct *mm = current->mm;
1933 profile_munmap(addr);
1935 down_write(&mm->mmap_sem);
1936 ret = do_munmap(mm, addr, len);
1937 up_write(&mm->mmap_sem);
1941 static inline void verify_mm_writelocked(struct mm_struct *mm)
1943 #ifdef CONFIG_DEBUG_VM
1944 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1946 up_read(&mm->mmap_sem);
1952 * this is really a simplified "do_mmap". it only handles
1953 * anonymous maps. eventually we may be able to do some
1954 * brk-specific accounting here.
1956 unsigned long do_brk(unsigned long addr, unsigned long len)
1958 struct mm_struct * mm = current->mm;
1959 struct vm_area_struct * vma, * prev;
1960 unsigned long flags;
1961 struct rb_node ** rb_link, * rb_parent;
1962 pgoff_t pgoff = addr >> PAGE_SHIFT;
1965 len = PAGE_ALIGN(len);
1969 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1972 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1974 error = arch_mmap_check(addr, len, flags);
1981 if (mm->def_flags & VM_LOCKED) {
1982 unsigned long locked, lock_limit;
1983 locked = len >> PAGE_SHIFT;
1984 locked += mm->locked_vm;
1985 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1986 lock_limit >>= PAGE_SHIFT;
1987 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1989 if (!vx_vmlocked_avail(mm, len >> PAGE_SHIFT))
1994 * mm->mmap_sem is required to protect against another thread
1995 * changing the mappings in case we sleep.
1997 verify_mm_writelocked(mm);
2000 * Clear old maps. this also does some error checking for us
2003 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2004 if (vma && vma->vm_start < addr + len) {
2005 if (do_munmap(mm, addr, len))
2010 /* Check against address space limits *after* clearing old maps... */
2011 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2014 if (mm->map_count > sysctl_max_map_count)
2017 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
2018 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
2021 /* Can we just expand an old private anonymous mapping? */
2022 if (vma_merge(mm, prev, addr, addr + len, flags,
2023 NULL, NULL, pgoff, NULL))
2027 * create a vma struct for an anonymous mapping
2029 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2031 vm_unacct_memory(len >> PAGE_SHIFT);
2036 vma->vm_start = addr;
2037 vma->vm_end = addr + len;
2038 vma->vm_pgoff = pgoff;
2039 vma->vm_flags = flags;
2040 vma->vm_page_prot = protection_map[flags &
2041 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
2042 vma_link(mm, vma, prev, rb_link, rb_parent);
2044 vx_vmpages_add(mm, len >> PAGE_SHIFT);
2045 if (flags & VM_LOCKED) {
2046 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
2047 make_pages_present(addr, addr + len);
2052 EXPORT_SYMBOL(do_brk);
2054 /* Release all mmaps. */
2055 void exit_mmap(struct mm_struct *mm)
2057 struct mmu_gather *tlb;
2058 struct vm_area_struct *vma = mm->mmap;
2059 unsigned long nr_accounted = 0;
2062 #ifdef arch_exit_mmap
2068 tlb = tlb_gather_mmu(mm, 1);
2069 /* Don't update_hiwater_rss(mm) here, do_exit already did */
2070 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2071 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2072 vm_unacct_memory(nr_accounted);
2073 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2074 tlb_finish_mmu(tlb, 0, end);
2075 arch_flush_exec_range(mm);
2077 set_mm_counter(mm, file_rss, 0);
2078 set_mm_counter(mm, anon_rss, 0);
2079 vx_vmpages_sub(mm, mm->total_vm);
2080 vx_vmlocked_sub(mm, mm->locked_vm);
2083 * Walk the list again, actually closing and freeing it,
2084 * with preemption enabled, without holding any MM locks.
2087 vma = remove_vma(vma);
2089 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2092 /* Insert vm structure into process list sorted by address
2093 * and into the inode's i_mmap tree. If vm_file is non-NULL
2094 * then i_mmap_lock is taken here.
2096 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2098 struct vm_area_struct * __vma, * prev;
2099 struct rb_node ** rb_link, * rb_parent;
2102 * The vm_pgoff of a purely anonymous vma should be irrelevant
2103 * until its first write fault, when page's anon_vma and index
2104 * are set. But now set the vm_pgoff it will almost certainly
2105 * end up with (unless mremap moves it elsewhere before that
2106 * first wfault), so /proc/pid/maps tells a consistent story.
2108 * By setting it to reflect the virtual start address of the
2109 * vma, merges and splits can happen in a seamless way, just
2110 * using the existing file pgoff checks and manipulations.
2111 * Similarly in do_mmap_pgoff and in do_brk.
2113 if (!vma->vm_file) {
2114 BUG_ON(vma->anon_vma);
2115 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2117 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2118 if (__vma && __vma->vm_start < vma->vm_end)
2120 if ((vma->vm_flags & VM_ACCOUNT) &&
2121 (security_vm_enough_memory(vma_pages(vma)) ||
2122 !vx_vmpages_avail(mm, vma_pages(vma))))
2124 vma_link(mm, vma, prev, rb_link, rb_parent);
2129 * Copy the vma structure to a new location in the same mm,
2130 * prior to moving page table entries, to effect an mremap move.
2132 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2133 unsigned long addr, unsigned long len, pgoff_t pgoff)
2135 struct vm_area_struct *vma = *vmap;
2136 unsigned long vma_start = vma->vm_start;
2137 struct mm_struct *mm = vma->vm_mm;
2138 struct vm_area_struct *new_vma, *prev;
2139 struct rb_node **rb_link, *rb_parent;
2140 struct mempolicy *pol;
2143 * If anonymous vma has not yet been faulted, update new pgoff
2144 * to match new location, to increase its chance of merging.
2146 if (!vma->vm_file && !vma->anon_vma)
2147 pgoff = addr >> PAGE_SHIFT;
2149 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2150 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2151 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2154 * Source vma may have been merged into new_vma
2156 if (vma_start >= new_vma->vm_start &&
2157 vma_start < new_vma->vm_end)
2160 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2163 pol = mpol_copy(vma_policy(vma));
2165 kmem_cache_free(vm_area_cachep, new_vma);
2168 vma_set_policy(new_vma, pol);
2169 new_vma->vm_start = addr;
2170 new_vma->vm_end = addr + len;
2171 new_vma->vm_pgoff = pgoff;
2172 if (new_vma->vm_file)
2173 get_file(new_vma->vm_file);
2174 if (new_vma->vm_ops && new_vma->vm_ops->open)
2175 new_vma->vm_ops->open(new_vma);
2176 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2183 * Return true if the calling process may expand its vm space by the passed
2186 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2188 unsigned long cur = mm->total_vm; /* pages */
2191 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2193 if (cur + npages > lim)
2195 if (!vx_vmpages_avail(mm, npages))
2201 static struct page *
2202 special_mapping_nopage(struct vm_area_struct *vma,
2203 unsigned long address, int *type)
2205 struct page **pages;
2207 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
2209 address -= vma->vm_start;
2210 for (pages = vma->vm_private_data; address > 0 && *pages; ++pages)
2211 address -= PAGE_SIZE;
2218 return NOPAGE_SIGBUS;
2221 static struct vm_operations_struct special_mapping_vmops = {
2222 .nopage = special_mapping_nopage,
2225 unsigned int vdso_populate = 1;
2228 * Insert a new vma covering the given region, with the given flags and
2229 * protections. Its pages are supplied by the given null-terminated array.
2230 * The region past the last page supplied will always produce SIGBUS.
2231 * The array pointer and the pages it points to are assumed to stay alive
2232 * for as long as this mapping might exist.
2234 int install_special_mapping(struct mm_struct *mm,
2235 unsigned long addr, unsigned long len,
2236 unsigned long vm_flags, pgprot_t pgprot,
2237 struct page **pages)
2239 struct vm_area_struct *vma;
2242 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2243 if (unlikely(vma == NULL))
2245 memset(vma, 0, sizeof(*vma));
2248 vma->vm_start = addr;
2249 vma->vm_end = addr + len;
2251 vma->vm_flags = vm_flags;
2252 vma->vm_page_prot = pgprot;
2254 vma->vm_ops = &special_mapping_vmops;
2255 vma->vm_private_data = pages;
2257 insert_vm_struct(mm, vma);
2258 vx_vmpages_add(mm, len >> PAGE_SHIFT);
2265 struct page *page = *pages++;
2267 err = install_page(mm, vma, addr, page, vma->vm_page_prot);