6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 #include <linux/slab.h>
10 #include <linux/backing-dev.h>
12 #include <linux/shm.h>
13 #include <linux/mman.h>
14 #include <linux/pagemap.h>
15 #include <linux/swap.h>
16 #include <linux/syscalls.h>
17 #include <linux/capability.h>
18 #include <linux/init.h>
19 #include <linux/file.h>
21 #include <linux/personality.h>
22 #include <linux/security.h>
23 #include <linux/hugetlb.h>
24 #include <linux/profile.h>
25 #include <linux/export.h>
26 #include <linux/mount.h>
27 #include <linux/mempolicy.h>
28 #include <linux/rmap.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/perf_event.h>
31 #include <linux/audit.h>
32 #include <linux/khugepaged.h>
34 #include <asm/uaccess.h>
35 #include <asm/cacheflush.h>
37 #include <asm/mmu_context.h>
41 #ifndef arch_mmap_check
42 #define arch_mmap_check(addr, len, flags) (0)
45 #ifndef arch_rebalance_pgtables
46 #define arch_rebalance_pgtables(addr, len) (addr)
49 static void unmap_region(struct mm_struct *mm,
50 struct vm_area_struct *vma, struct vm_area_struct *prev,
51 unsigned long start, unsigned long end);
54 * WARNING: the debugging will use recursive algorithms so never enable this
55 * unless you know what you are doing.
59 /* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
74 pgprot_t protection_map[16] = {
75 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
76 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
79 pgprot_t vm_get_page_prot(unsigned long vm_flags)
81 return __pgprot(pgprot_val(protection_map[vm_flags &
82 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
83 pgprot_val(arch_vm_get_page_prot(vm_flags)));
85 EXPORT_SYMBOL(vm_get_page_prot);
87 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
88 int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
89 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
91 * Make sure vm_committed_as in one cacheline and not cacheline shared with
92 * other variables. It can be updated by several CPUs frequently.
94 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
97 * Check that a process has enough memory to allocate a new virtual
98 * mapping. 0 means there is enough memory for the allocation to
99 * succeed and -ENOMEM implies there is not.
101 * We currently support three overcommit policies, which are set via the
102 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
104 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
105 * Additional code 2002 Jul 20 by Robert Love.
107 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
109 * Note this is a helper function intended to be used by LSMs which
110 * wish to use this logic.
112 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
116 vm_acct_memory(pages);
119 * Sometimes we want to use more memory than we have
121 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
124 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
125 free = global_page_state(NR_FREE_PAGES);
126 free += global_page_state(NR_FILE_PAGES);
129 * shmem pages shouldn't be counted as free in this
130 * case, they can't be purged, only swapped out, and
131 * that won't affect the overall amount of available
132 * memory in the system.
134 free -= global_page_state(NR_SHMEM);
136 free += nr_swap_pages;
139 * Any slabs which are created with the
140 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
141 * which are reclaimable, under pressure. The dentry
142 * cache and most inode caches should fall into this
144 free += global_page_state(NR_SLAB_RECLAIMABLE);
147 * Leave reserved pages. The pages are not for anonymous pages.
149 if (free <= totalreserve_pages)
152 free -= totalreserve_pages;
155 * Leave the last 3% for root
166 allowed = (totalram_pages - hugetlb_total_pages())
167 * sysctl_overcommit_ratio / 100;
169 * Leave the last 3% for root
172 allowed -= allowed / 32;
173 allowed += total_swap_pages;
175 /* Don't let a single process grow too big:
176 leave 3% of the size of this process for other processes */
178 allowed -= mm->total_vm / 32;
180 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
183 vm_unacct_memory(pages);
189 * Requires inode->i_mapping->i_mmap_mutex
191 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
192 struct file *file, struct address_space *mapping)
194 if (vma->vm_flags & VM_DENYWRITE)
195 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
196 if (vma->vm_flags & VM_SHARED)
197 mapping->i_mmap_writable--;
199 flush_dcache_mmap_lock(mapping);
200 if (unlikely(vma->vm_flags & VM_NONLINEAR))
201 list_del_init(&vma->shared.vm_set.list);
203 vma_prio_tree_remove(vma, &mapping->i_mmap);
204 flush_dcache_mmap_unlock(mapping);
208 * Unlink a file-based vm structure from its prio_tree, to hide
209 * vma from rmap and vmtruncate before freeing its page tables.
211 void unlink_file_vma(struct vm_area_struct *vma)
213 struct file *file = vma->vm_file;
216 struct address_space *mapping = file->f_mapping;
217 mutex_lock(&mapping->i_mmap_mutex);
218 __remove_shared_vm_struct(vma, file, mapping);
219 mutex_unlock(&mapping->i_mmap_mutex);
224 * Close a vm structure and free it, returning the next.
226 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
228 struct vm_area_struct *next = vma->vm_next;
231 if (vma->vm_ops && vma->vm_ops->close)
232 vma->vm_ops->close(vma);
235 if (vma->vm_flags & VM_EXECUTABLE)
236 removed_exe_file_vma(vma->vm_mm);
238 mpol_put(vma_policy(vma));
239 kmem_cache_free(vm_area_cachep, vma);
243 SYSCALL_DEFINE1(brk, unsigned long, brk)
245 unsigned long rlim, retval;
246 unsigned long newbrk, oldbrk;
247 struct mm_struct *mm = current->mm;
248 struct vm_area_struct *next;
249 unsigned long min_brk;
251 down_write(&mm->mmap_sem);
253 #ifdef CONFIG_COMPAT_BRK
255 * CONFIG_COMPAT_BRK can still be overridden by setting
256 * randomize_va_space to 2, which will still cause mm->start_brk
257 * to be arbitrarily shifted
259 if (current->brk_randomized)
260 min_brk = mm->start_brk;
262 min_brk = mm->end_data;
264 min_brk = mm->start_brk;
270 * Check against rlimit here. If this check is done later after the test
271 * of oldbrk with newbrk then it can escape the test and let the data
272 * segment grow beyond its set limit the in case where the limit is
273 * not page aligned -Ram Gupta
275 rlim = rlimit(RLIMIT_DATA);
276 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
277 (mm->end_data - mm->start_data) > rlim)
280 newbrk = PAGE_ALIGN(brk);
281 oldbrk = PAGE_ALIGN(mm->brk);
282 if (oldbrk == newbrk)
285 /* Always allow shrinking brk. */
286 if (brk <= mm->brk) {
287 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
292 /* Check against existing mmap mappings. */
293 next = find_vma(mm, oldbrk);
294 if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
297 /* Ok, looks good - let it rip. */
298 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
304 up_write(&mm->mmap_sem);
309 static int browse_rb(struct rb_root *root)
312 struct rb_node *nd, *pn = NULL;
313 unsigned long prev = 0, pend = 0;
315 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
316 struct vm_area_struct *vma;
317 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
318 if (vma->vm_start < prev)
319 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
320 if (vma->vm_start < pend)
321 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
322 if (vma->vm_start > vma->vm_end)
323 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
326 prev = vma->vm_start;
330 for (nd = pn; nd; nd = rb_prev(nd)) {
334 printk("backwards %d, forwards %d\n", j, i), i = 0;
338 void validate_mm(struct mm_struct *mm)
342 struct vm_area_struct *tmp = mm->mmap;
347 if (i != mm->map_count)
348 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
349 i = browse_rb(&mm->mm_rb);
350 if (i != mm->map_count)
351 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
355 #define validate_mm(mm) do { } while (0)
358 static struct vm_area_struct *
359 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
360 struct vm_area_struct **pprev, struct rb_node ***rb_link,
361 struct rb_node ** rb_parent)
363 struct vm_area_struct * vma;
364 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
366 __rb_link = &mm->mm_rb.rb_node;
367 rb_prev = __rb_parent = NULL;
371 struct vm_area_struct *vma_tmp;
373 __rb_parent = *__rb_link;
374 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
376 if (vma_tmp->vm_end > addr) {
378 if (vma_tmp->vm_start <= addr)
380 __rb_link = &__rb_parent->rb_left;
382 rb_prev = __rb_parent;
383 __rb_link = &__rb_parent->rb_right;
389 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
390 *rb_link = __rb_link;
391 *rb_parent = __rb_parent;
395 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
396 struct rb_node **rb_link, struct rb_node *rb_parent)
398 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
399 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
402 static void __vma_link_file(struct vm_area_struct *vma)
408 struct address_space *mapping = file->f_mapping;
410 if (vma->vm_flags & VM_DENYWRITE)
411 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
412 if (vma->vm_flags & VM_SHARED)
413 mapping->i_mmap_writable++;
415 flush_dcache_mmap_lock(mapping);
416 if (unlikely(vma->vm_flags & VM_NONLINEAR))
417 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
419 vma_prio_tree_insert(vma, &mapping->i_mmap);
420 flush_dcache_mmap_unlock(mapping);
425 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
426 struct vm_area_struct *prev, struct rb_node **rb_link,
427 struct rb_node *rb_parent)
429 __vma_link_list(mm, vma, prev, rb_parent);
430 __vma_link_rb(mm, vma, rb_link, rb_parent);
433 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
434 struct vm_area_struct *prev, struct rb_node **rb_link,
435 struct rb_node *rb_parent)
437 struct address_space *mapping = NULL;
440 mapping = vma->vm_file->f_mapping;
443 mutex_lock(&mapping->i_mmap_mutex);
445 __vma_link(mm, vma, prev, rb_link, rb_parent);
446 __vma_link_file(vma);
449 mutex_unlock(&mapping->i_mmap_mutex);
456 * Helper for vma_adjust in the split_vma insert case:
457 * insert vm structure into list and rbtree and anon_vma,
458 * but it has already been inserted into prio_tree earlier.
460 static void __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 struct vm_area_struct *next = vma->vm_next;
477 prev->vm_next = next;
479 next->vm_prev = prev;
480 rb_erase(&vma->vm_rb, &mm->mm_rb);
481 if (mm->mmap_cache == vma)
482 mm->mmap_cache = prev;
486 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
487 * is already present in an i_mmap tree without adjusting the tree.
488 * The following helper function should be used when such adjustments
489 * are necessary. The "insert" vma (if any) is to be inserted
490 * before we drop the necessary locks.
492 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
493 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
495 struct mm_struct *mm = vma->vm_mm;
496 struct vm_area_struct *next = vma->vm_next;
497 struct vm_area_struct *importer = NULL;
498 struct address_space *mapping = NULL;
499 struct prio_tree_root *root = NULL;
500 struct anon_vma *anon_vma = NULL;
501 struct file *file = vma->vm_file;
502 long adjust_next = 0;
505 if (next && !insert) {
506 struct vm_area_struct *exporter = NULL;
508 if (end >= next->vm_end) {
510 * vma expands, overlapping all the next, and
511 * perhaps the one after too (mprotect case 6).
513 again: remove_next = 1 + (end > next->vm_end);
517 } else if (end > next->vm_start) {
519 * vma expands, overlapping part of the next:
520 * mprotect case 5 shifting the boundary up.
522 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
525 } else if (end < vma->vm_end) {
527 * vma shrinks, and !insert tells it's not
528 * split_vma inserting another: so it must be
529 * mprotect case 4 shifting the boundary down.
531 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
537 * Easily overlooked: when mprotect shifts the boundary,
538 * make sure the expanding vma has anon_vma set if the
539 * shrinking vma had, to cover any anon pages imported.
541 if (exporter && exporter->anon_vma && !importer->anon_vma) {
544 importer->anon_vma = exporter->anon_vma;
545 error = anon_vma_clone(importer, exporter);
552 mapping = file->f_mapping;
553 if (!(vma->vm_flags & VM_NONLINEAR))
554 root = &mapping->i_mmap;
555 mutex_lock(&mapping->i_mmap_mutex);
558 * Put into prio_tree now, so instantiated pages
559 * are visible to arm/parisc __flush_dcache_page
560 * throughout; but we cannot insert into address
561 * space until vma start or end is updated.
563 __vma_link_file(insert);
567 vma_adjust_trans_huge(vma, start, end, adjust_next);
570 * When changing only vma->vm_end, we don't really need anon_vma
571 * lock. This is a fairly rare case by itself, but the anon_vma
572 * lock may be shared between many sibling processes. Skipping
573 * the lock for brk adjustments makes a difference sometimes.
575 if (vma->anon_vma && (importer || start != vma->vm_start)) {
576 anon_vma = vma->anon_vma;
577 anon_vma_lock(anon_vma);
581 flush_dcache_mmap_lock(mapping);
582 vma_prio_tree_remove(vma, root);
584 vma_prio_tree_remove(next, root);
587 vma->vm_start = start;
589 vma->vm_pgoff = pgoff;
591 next->vm_start += adjust_next << PAGE_SHIFT;
592 next->vm_pgoff += adjust_next;
597 vma_prio_tree_insert(next, root);
598 vma_prio_tree_insert(vma, root);
599 flush_dcache_mmap_unlock(mapping);
604 * vma_merge has merged next into vma, and needs
605 * us to remove next before dropping the locks.
607 __vma_unlink(mm, next, vma);
609 __remove_shared_vm_struct(next, file, mapping);
612 * split_vma has split insert from vma, and needs
613 * us to insert it before dropping the locks
614 * (it may either follow vma or precede it).
616 __insert_vm_struct(mm, insert);
620 anon_vma_unlock(anon_vma);
622 mutex_unlock(&mapping->i_mmap_mutex);
627 if (next->vm_flags & VM_EXECUTABLE)
628 removed_exe_file_vma(mm);
631 anon_vma_merge(vma, next);
633 mpol_put(vma_policy(next));
634 kmem_cache_free(vm_area_cachep, next);
636 * In mprotect's case 6 (see comments on vma_merge),
637 * we must remove another next too. It would clutter
638 * up the code too much to do both in one go.
640 if (remove_next == 2) {
652 * If the vma has a ->close operation then the driver probably needs to release
653 * per-vma resources, so we don't attempt to merge those.
655 static inline int is_mergeable_vma(struct vm_area_struct *vma,
656 struct file *file, unsigned long vm_flags)
658 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
659 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
661 if (vma->vm_file != file)
663 if (vma->vm_ops && vma->vm_ops->close)
668 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
669 struct anon_vma *anon_vma2,
670 struct vm_area_struct *vma)
673 * The list_is_singular() test is to avoid merging VMA cloned from
674 * parents. This can improve scalability caused by anon_vma lock.
676 if ((!anon_vma1 || !anon_vma2) && (!vma ||
677 list_is_singular(&vma->anon_vma_chain)))
679 return anon_vma1 == anon_vma2;
683 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
684 * in front of (at a lower virtual address and file offset than) the vma.
686 * We cannot merge two vmas if they have differently assigned (non-NULL)
687 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
689 * We don't check here for the merged mmap wrapping around the end of pagecache
690 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
691 * wrap, nor mmaps which cover the final page at index -1UL.
694 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
695 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
697 if (is_mergeable_vma(vma, file, vm_flags) &&
698 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
699 if (vma->vm_pgoff == vm_pgoff)
706 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
707 * beyond (at a higher virtual address and file offset than) the vma.
709 * We cannot merge two vmas if they have differently assigned (non-NULL)
710 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
713 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
714 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
716 if (is_mergeable_vma(vma, file, vm_flags) &&
717 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
719 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
720 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
727 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
728 * whether that can be merged with its predecessor or its successor.
729 * Or both (it neatly fills a hole).
731 * In most cases - when called for mmap, brk or mremap - [addr,end) is
732 * certain not to be mapped by the time vma_merge is called; but when
733 * called for mprotect, it is certain to be already mapped (either at
734 * an offset within prev, or at the start of next), and the flags of
735 * this area are about to be changed to vm_flags - and the no-change
736 * case has already been eliminated.
738 * The following mprotect cases have to be considered, where AAAA is
739 * the area passed down from mprotect_fixup, never extending beyond one
740 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
742 * AAAA AAAA AAAA AAAA
743 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
744 * cannot merge might become might become might become
745 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
746 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
747 * mremap move: PPPPNNNNNNNN 8
749 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
750 * might become case 1 below case 2 below case 3 below
752 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
753 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
755 struct vm_area_struct *vma_merge(struct mm_struct *mm,
756 struct vm_area_struct *prev, unsigned long addr,
757 unsigned long end, unsigned long vm_flags,
758 struct anon_vma *anon_vma, struct file *file,
759 pgoff_t pgoff, struct mempolicy *policy)
761 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
762 struct vm_area_struct *area, *next;
766 * We later require that vma->vm_flags == vm_flags,
767 * so this tests vma->vm_flags & VM_SPECIAL, too.
769 if (vm_flags & VM_SPECIAL)
773 next = prev->vm_next;
777 if (next && next->vm_end == end) /* cases 6, 7, 8 */
778 next = next->vm_next;
781 * Can it merge with the predecessor?
783 if (prev && prev->vm_end == addr &&
784 mpol_equal(vma_policy(prev), policy) &&
785 can_vma_merge_after(prev, vm_flags,
786 anon_vma, file, pgoff)) {
788 * OK, it can. Can we now merge in the successor as well?
790 if (next && end == next->vm_start &&
791 mpol_equal(policy, vma_policy(next)) &&
792 can_vma_merge_before(next, vm_flags,
793 anon_vma, file, pgoff+pglen) &&
794 is_mergeable_anon_vma(prev->anon_vma,
795 next->anon_vma, NULL)) {
797 err = vma_adjust(prev, prev->vm_start,
798 next->vm_end, prev->vm_pgoff, NULL);
799 } else /* cases 2, 5, 7 */
800 err = vma_adjust(prev, prev->vm_start,
801 end, prev->vm_pgoff, NULL);
804 khugepaged_enter_vma_merge(prev, vm_flags);
809 * Can this new request be merged in front of next?
811 if (next && end == next->vm_start &&
812 mpol_equal(policy, vma_policy(next)) &&
813 can_vma_merge_before(next, vm_flags,
814 anon_vma, file, pgoff+pglen)) {
815 if (prev && addr < prev->vm_end) /* case 4 */
816 err = vma_adjust(prev, prev->vm_start,
817 addr, prev->vm_pgoff, NULL);
818 else /* cases 3, 8 */
819 err = vma_adjust(area, addr, next->vm_end,
820 next->vm_pgoff - pglen, NULL);
823 khugepaged_enter_vma_merge(area, vm_flags);
831 * Rough compatbility check to quickly see if it's even worth looking
832 * at sharing an anon_vma.
834 * They need to have the same vm_file, and the flags can only differ
835 * in things that mprotect may change.
837 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
838 * we can merge the two vma's. For example, we refuse to merge a vma if
839 * there is a vm_ops->close() function, because that indicates that the
840 * driver is doing some kind of reference counting. But that doesn't
841 * really matter for the anon_vma sharing case.
843 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
845 return a->vm_end == b->vm_start &&
846 mpol_equal(vma_policy(a), vma_policy(b)) &&
847 a->vm_file == b->vm_file &&
848 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
849 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
853 * Do some basic sanity checking to see if we can re-use the anon_vma
854 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
855 * the same as 'old', the other will be the new one that is trying
856 * to share the anon_vma.
858 * NOTE! This runs with mm_sem held for reading, so it is possible that
859 * the anon_vma of 'old' is concurrently in the process of being set up
860 * by another page fault trying to merge _that_. But that's ok: if it
861 * is being set up, that automatically means that it will be a singleton
862 * acceptable for merging, so we can do all of this optimistically. But
863 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
865 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
866 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
867 * is to return an anon_vma that is "complex" due to having gone through
870 * We also make sure that the two vma's are compatible (adjacent,
871 * and with the same memory policies). That's all stable, even with just
872 * a read lock on the mm_sem.
874 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
876 if (anon_vma_compatible(a, b)) {
877 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
879 if (anon_vma && list_is_singular(&old->anon_vma_chain))
886 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
887 * neighbouring vmas for a suitable anon_vma, before it goes off
888 * to allocate a new anon_vma. It checks because a repetitive
889 * sequence of mprotects and faults may otherwise lead to distinct
890 * anon_vmas being allocated, preventing vma merge in subsequent
893 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
895 struct anon_vma *anon_vma;
896 struct vm_area_struct *near;
902 anon_vma = reusable_anon_vma(near, vma, near);
910 anon_vma = reusable_anon_vma(near, near, vma);
915 * There's no absolute need to look only at touching neighbours:
916 * we could search further afield for "compatible" anon_vmas.
917 * But it would probably just be a waste of time searching,
918 * or lead to too many vmas hanging off the same anon_vma.
919 * We're trying to allow mprotect remerging later on,
920 * not trying to minimize memory used for anon_vmas.
925 #ifdef CONFIG_PROC_FS
926 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
927 struct file *file, long pages)
929 const unsigned long stack_flags
930 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
933 mm->shared_vm += pages;
934 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
935 mm->exec_vm += pages;
936 } else if (flags & stack_flags)
937 mm->stack_vm += pages;
938 if (flags & (VM_RESERVED|VM_IO))
939 mm->reserved_vm += pages;
941 #endif /* CONFIG_PROC_FS */
944 * The caller must hold down_write(¤t->mm->mmap_sem).
947 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
948 unsigned long len, unsigned long prot,
949 unsigned long flags, unsigned long pgoff)
951 struct mm_struct * mm = current->mm;
955 unsigned long reqprot = prot;
958 * Does the application expect PROT_READ to imply PROT_EXEC?
960 * (the exception is when the underlying filesystem is noexec
961 * mounted, in which case we dont add PROT_EXEC.)
963 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
964 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
970 if (!(flags & MAP_FIXED))
971 addr = round_hint_to_min(addr);
973 /* Careful about overflows.. */
974 len = PAGE_ALIGN(len);
978 /* offset overflow? */
979 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
982 /* Too many mappings? */
983 if (mm->map_count > sysctl_max_map_count)
986 /* Obtain the address to map to. we verify (or select) it and ensure
987 * that it represents a valid section of the address space.
989 addr = get_unmapped_area(file, addr, len, pgoff, flags);
990 if (addr & ~PAGE_MASK)
993 /* Do simple checking here so the lower-level routines won't have
994 * to. we assume access permissions have been handled by the open
995 * of the memory object, so we don't do any here.
997 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
998 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1000 if (flags & MAP_LOCKED)
1001 if (!can_do_mlock())
1004 /* mlock MCL_FUTURE? */
1005 if (vm_flags & VM_LOCKED) {
1006 unsigned long locked, lock_limit;
1007 locked = len >> PAGE_SHIFT;
1008 locked += mm->locked_vm;
1009 lock_limit = rlimit(RLIMIT_MEMLOCK);
1010 lock_limit >>= PAGE_SHIFT;
1011 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1015 inode = file ? file->f_path.dentry->d_inode : NULL;
1018 switch (flags & MAP_TYPE) {
1020 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1024 * Make sure we don't allow writing to an append-only
1027 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1031 * Make sure there are no mandatory locks on the file.
1033 if (locks_verify_locked(inode))
1036 vm_flags |= VM_SHARED | VM_MAYSHARE;
1037 if (!(file->f_mode & FMODE_WRITE))
1038 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1042 if (!(file->f_mode & FMODE_READ))
1044 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1045 if (vm_flags & VM_EXEC)
1047 vm_flags &= ~VM_MAYEXEC;
1050 if (!file->f_op || !file->f_op->mmap)
1058 switch (flags & MAP_TYPE) {
1064 vm_flags |= VM_SHARED | VM_MAYSHARE;
1068 * Set pgoff according to addr for anon_vma.
1070 pgoff = addr >> PAGE_SHIFT;
1077 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1081 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1083 EXPORT_SYMBOL(do_mmap_pgoff);
1085 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1086 unsigned long, prot, unsigned long, flags,
1087 unsigned long, fd, unsigned long, pgoff)
1089 struct file *file = NULL;
1090 unsigned long retval = -EBADF;
1092 if (!(flags & MAP_ANONYMOUS)) {
1093 audit_mmap_fd(fd, flags);
1094 if (unlikely(flags & MAP_HUGETLB))
1099 } else if (flags & MAP_HUGETLB) {
1100 struct user_struct *user = NULL;
1102 * VM_NORESERVE is used because the reservations will be
1103 * taken when vm_ops->mmap() is called
1104 * A dummy user value is used because we are not locking
1105 * memory so no accounting is necessary
1107 len = ALIGN(len, huge_page_size(&default_hstate));
1108 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1109 &user, HUGETLB_ANONHUGE_INODE);
1111 return PTR_ERR(file);
1114 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1116 down_write(¤t->mm->mmap_sem);
1117 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1118 up_write(¤t->mm->mmap_sem);
1126 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1127 struct mmap_arg_struct {
1131 unsigned long flags;
1133 unsigned long offset;
1136 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1138 struct mmap_arg_struct a;
1140 if (copy_from_user(&a, arg, sizeof(a)))
1142 if (a.offset & ~PAGE_MASK)
1145 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1146 a.offset >> PAGE_SHIFT);
1148 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1151 * Some shared mappigns will want the pages marked read-only
1152 * to track write events. If so, we'll downgrade vm_page_prot
1153 * to the private version (using protection_map[] without the
1156 int vma_wants_writenotify(struct vm_area_struct *vma)
1158 vm_flags_t vm_flags = vma->vm_flags;
1160 /* If it was private or non-writable, the write bit is already clear */
1161 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1164 /* The backer wishes to know when pages are first written to? */
1165 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1168 /* The open routine did something to the protections already? */
1169 if (pgprot_val(vma->vm_page_prot) !=
1170 pgprot_val(vm_get_page_prot(vm_flags)))
1173 /* Specialty mapping? */
1174 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1177 /* Can the mapping track the dirty pages? */
1178 return vma->vm_file && vma->vm_file->f_mapping &&
1179 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1183 * We account for memory if it's a private writeable mapping,
1184 * not hugepages and VM_NORESERVE wasn't set.
1186 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1189 * hugetlb has its own accounting separate from the core VM
1190 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1192 if (file && is_file_hugepages(file))
1195 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1198 unsigned long mmap_region(struct file *file, unsigned long addr,
1199 unsigned long len, unsigned long flags,
1200 vm_flags_t vm_flags, unsigned long pgoff)
1202 struct mm_struct *mm = current->mm;
1203 struct vm_area_struct *vma, *prev;
1204 int correct_wcount = 0;
1206 struct rb_node **rb_link, *rb_parent;
1207 unsigned long charged = 0;
1208 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1210 /* Clear old maps */
1213 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1214 if (vma && vma->vm_start < addr + len) {
1215 if (do_munmap(mm, addr, len))
1220 /* Check against address space limit. */
1221 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1225 * Set 'VM_NORESERVE' if we should not account for the
1226 * memory use of this mapping.
1228 if ((flags & MAP_NORESERVE)) {
1229 /* We honor MAP_NORESERVE if allowed to overcommit */
1230 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1231 vm_flags |= VM_NORESERVE;
1233 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1234 if (file && is_file_hugepages(file))
1235 vm_flags |= VM_NORESERVE;
1239 * Private writable mapping: check memory availability
1241 if (accountable_mapping(file, vm_flags)) {
1242 charged = len >> PAGE_SHIFT;
1243 if (security_vm_enough_memory(charged))
1245 vm_flags |= VM_ACCOUNT;
1249 * Can we just expand an old mapping?
1251 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1256 * Determine the object being mapped and call the appropriate
1257 * specific mapper. the address has already been validated, but
1258 * not unmapped, but the maps are removed from the list.
1260 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1267 vma->vm_start = addr;
1268 vma->vm_end = addr + len;
1269 vma->vm_flags = vm_flags;
1270 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1271 vma->vm_pgoff = pgoff;
1272 INIT_LIST_HEAD(&vma->anon_vma_chain);
1276 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1278 if (vm_flags & VM_DENYWRITE) {
1279 error = deny_write_access(file);
1284 vma->vm_file = file;
1286 error = file->f_op->mmap(file, vma);
1288 goto unmap_and_free_vma;
1289 if (vm_flags & VM_EXECUTABLE)
1290 added_exe_file_vma(mm);
1292 /* Can addr have changed??
1294 * Answer: Yes, several device drivers can do it in their
1295 * f_op->mmap method. -DaveM
1297 addr = vma->vm_start;
1298 pgoff = vma->vm_pgoff;
1299 vm_flags = vma->vm_flags;
1300 } else if (vm_flags & VM_SHARED) {
1301 error = shmem_zero_setup(vma);
1306 if (vma_wants_writenotify(vma)) {
1307 pgprot_t pprot = vma->vm_page_prot;
1309 /* Can vma->vm_page_prot have changed??
1311 * Answer: Yes, drivers may have changed it in their
1312 * f_op->mmap method.
1314 * Ensures that vmas marked as uncached stay that way.
1316 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1317 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1318 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1321 vma_link(mm, vma, prev, rb_link, rb_parent);
1322 file = vma->vm_file;
1324 /* Once vma denies write, undo our temporary denial count */
1326 atomic_inc(&inode->i_writecount);
1328 perf_event_mmap(vma);
1330 mm->total_vm += len >> PAGE_SHIFT;
1331 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1332 if (vm_flags & VM_LOCKED) {
1333 if (!mlock_vma_pages_range(vma, addr, addr + len))
1334 mm->locked_vm += (len >> PAGE_SHIFT);
1335 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1336 make_pages_present(addr, addr + len);
1341 atomic_inc(&inode->i_writecount);
1342 vma->vm_file = NULL;
1345 /* Undo any partial mapping done by a device driver. */
1346 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1349 kmem_cache_free(vm_area_cachep, vma);
1352 vm_unacct_memory(charged);
1356 /* Get an address range which is currently unmapped.
1357 * For shmat() with addr=0.
1359 * Ugly calling convention alert:
1360 * Return value with the low bits set means error value,
1362 * if (ret & ~PAGE_MASK)
1365 * This function "knows" that -ENOMEM has the bits set.
1367 #ifndef HAVE_ARCH_UNMAPPED_AREA
1369 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1370 unsigned long len, unsigned long pgoff, unsigned long flags)
1372 struct mm_struct *mm = current->mm;
1373 struct vm_area_struct *vma, *prev;
1374 unsigned long start_addr, vm_start, prev_end;
1376 if (len > TASK_SIZE - mmap_min_addr)
1379 if (flags & MAP_FIXED)
1383 addr = PAGE_ALIGN(addr);
1384 vma = find_vma_prev(mm, addr, &prev);
1385 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1386 (!vma || addr + len <= vm_start_gap(vma)) &&
1387 (!prev || addr >= vm_end_gap(prev)))
1390 if (len > mm->cached_hole_size) {
1391 start_addr = addr = mm->free_area_cache;
1393 start_addr = addr = TASK_UNMAPPED_BASE;
1394 mm->cached_hole_size = 0;
1398 for (vma = find_vma_prev(mm, addr, &prev); ; prev = vma,
1399 vma = vma->vm_next) {
1401 prev_end = vm_end_gap(prev);
1402 if (addr < prev_end) {
1404 /* If vma already violates gap, forget it */
1405 if (vma && addr > vma->vm_start)
1406 addr = vma->vm_start;
1409 /* At this point: (!vma || addr < vma->vm_end). */
1410 if (TASK_SIZE - len < addr) {
1412 * Start a new search - just in case we missed
1415 if (start_addr != TASK_UNMAPPED_BASE) {
1416 addr = TASK_UNMAPPED_BASE;
1418 mm->cached_hole_size = 0;
1423 vm_start = vma ? vm_start_gap(vma) : TASK_SIZE;
1424 if (addr + len <= vm_start) {
1426 * Remember the place where we stopped the search:
1428 mm->free_area_cache = addr + len;
1431 if (addr + mm->cached_hole_size < vm_start)
1432 mm->cached_hole_size = vm_start - addr;
1437 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1440 * Is this a new hole at the lowest possible address?
1442 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1443 mm->free_area_cache = addr;
1444 mm->cached_hole_size = ~0UL;
1449 * This mmap-allocator allocates new areas top-down from below the
1450 * stack's low limit (the base):
1452 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1454 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1455 const unsigned long len, const unsigned long pgoff,
1456 const unsigned long flags)
1458 struct vm_area_struct *vma, *prev;
1459 struct mm_struct *mm = current->mm;
1460 unsigned long addr = addr0;
1461 unsigned long vm_start, prev_end;
1462 unsigned long low_limit = max(PAGE_SIZE, mmap_min_addr);
1464 /* requested length too big for entire address space */
1465 if (len > TASK_SIZE - mmap_min_addr)
1468 if (flags & MAP_FIXED)
1471 /* requesting a specific address */
1473 addr = PAGE_ALIGN(addr);
1474 vma = find_vma_prev(mm, addr, &prev);
1475 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1476 (!vma || addr + len <= vm_start_gap(vma)) &&
1477 (!prev || addr >= vm_end_gap(prev)))
1481 /* check if free_area_cache is useful for us */
1482 if (len <= mm->cached_hole_size) {
1483 mm->cached_hole_size = 0;
1484 mm->free_area_cache = mm->mmap_base;
1487 /* either no address requested or can't fit in requested address hole */
1488 addr = mm->free_area_cache;
1490 /* make sure it can fit in the remaining address space */
1491 if (addr >= low_limit + len) {
1492 vma = find_vma_prev(mm, addr-len, &prev);
1493 if ((!vma || addr <= vm_start_gap(vma)) &&
1494 (!prev || addr-len >= vm_end_gap(prev)))
1495 /* remember the address as a hint for next time */
1496 return (mm->free_area_cache = addr-len);
1499 if (mm->mmap_base < low_limit + len)
1502 addr = mm->mmap_base-len;
1506 * Lookup failure means no vma is above this address,
1507 * else if new region fits below vma->vm_start,
1508 * return with success:
1510 vma = find_vma_prev(mm, addr, &prev);
1511 vm_start = vma ? vm_start_gap(vma) : mm->mmap_base;
1512 prev_end = prev ? vm_end_gap(prev) : low_limit;
1514 if (addr + len <= vm_start && addr >= prev_end)
1515 /* remember the address as a hint for next time */
1516 return (mm->free_area_cache = addr);
1518 /* remember the largest hole we saw so far */
1519 if (addr + mm->cached_hole_size < vm_start)
1520 mm->cached_hole_size = vm_start - addr;
1522 /* try just below the current vma->vm_start */
1523 addr = vm_start - len;
1524 } while (vm_start >= low_limit + len);
1528 * A failed mmap() very likely causes application failure,
1529 * so fall back to the bottom-up function here. This scenario
1530 * can happen with large stack limits and large mmap()
1533 mm->cached_hole_size = ~0UL;
1534 mm->free_area_cache = TASK_UNMAPPED_BASE;
1535 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1537 * Restore the topdown base:
1539 mm->free_area_cache = mm->mmap_base;
1540 mm->cached_hole_size = ~0UL;
1546 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1549 * Is this a new hole at the highest possible address?
1551 if (addr > mm->free_area_cache)
1552 mm->free_area_cache = addr;
1554 /* dont allow allocations above current base */
1555 if (mm->free_area_cache > mm->mmap_base)
1556 mm->free_area_cache = mm->mmap_base;
1560 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1561 unsigned long pgoff, unsigned long flags)
1563 unsigned long (*get_area)(struct file *, unsigned long,
1564 unsigned long, unsigned long, unsigned long);
1566 unsigned long error = arch_mmap_check(addr, len, flags);
1570 /* Careful about overflows.. */
1571 if (len > TASK_SIZE)
1574 get_area = current->mm->get_unmapped_area;
1575 if (file && file->f_op && file->f_op->get_unmapped_area)
1576 get_area = file->f_op->get_unmapped_area;
1577 addr = get_area(file, addr, len, pgoff, flags);
1578 if (IS_ERR_VALUE(addr))
1581 if (addr > TASK_SIZE - len)
1583 if (addr & ~PAGE_MASK)
1586 return arch_rebalance_pgtables(addr, len);
1589 EXPORT_SYMBOL(get_unmapped_area);
1591 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1592 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1594 struct vm_area_struct *vma = NULL;
1597 /* Check the cache first. */
1598 /* (Cache hit rate is typically around 35%.) */
1599 vma = ACCESS_ONCE(mm->mmap_cache);
1600 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1601 struct rb_node * rb_node;
1603 rb_node = mm->mm_rb.rb_node;
1607 struct vm_area_struct * vma_tmp;
1609 vma_tmp = rb_entry(rb_node,
1610 struct vm_area_struct, vm_rb);
1612 if (vma_tmp->vm_end > addr) {
1614 if (vma_tmp->vm_start <= addr)
1616 rb_node = rb_node->rb_left;
1618 rb_node = rb_node->rb_right;
1621 mm->mmap_cache = vma;
1627 EXPORT_SYMBOL(find_vma);
1630 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1632 struct vm_area_struct *
1633 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1634 struct vm_area_struct **pprev)
1636 struct vm_area_struct *vma;
1638 vma = find_vma(mm, addr);
1640 *pprev = vma->vm_prev;
1642 struct rb_node *rb_node = mm->mm_rb.rb_node;
1645 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1646 rb_node = rb_node->rb_right;
1653 * Verify that the stack growth is acceptable and
1654 * update accounting. This is shared with both the
1655 * grow-up and grow-down cases.
1657 static int acct_stack_growth(struct vm_area_struct *vma,
1658 unsigned long size, unsigned long grow)
1660 struct mm_struct *mm = vma->vm_mm;
1661 struct rlimit *rlim = current->signal->rlim;
1662 unsigned long new_start;
1664 /* address space limit tests */
1665 if (!may_expand_vm(mm, grow))
1668 /* Stack limit test */
1669 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1672 /* mlock limit tests */
1673 if (vma->vm_flags & VM_LOCKED) {
1674 unsigned long locked;
1675 unsigned long limit;
1676 locked = mm->locked_vm + grow;
1677 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1678 limit >>= PAGE_SHIFT;
1679 if (locked > limit && !capable(CAP_IPC_LOCK))
1683 /* Check to ensure the stack will not grow into a hugetlb-only region */
1684 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1686 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1690 * Overcommit.. This must be the final test, as it will
1691 * update security statistics.
1693 if (security_vm_enough_memory_mm(mm, grow))
1696 /* Ok, everything looks good - let it rip */
1697 mm->total_vm += grow;
1698 if (vma->vm_flags & VM_LOCKED)
1699 mm->locked_vm += grow;
1700 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1704 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1706 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1707 * vma is the last one with address > vma->vm_end. Have to extend vma.
1709 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1711 struct vm_area_struct *next;
1712 unsigned long gap_addr;
1715 if (!(vma->vm_flags & VM_GROWSUP))
1718 /* Guard against exceeding limits of the address space. */
1719 address &= PAGE_MASK;
1720 if (address >= (TASK_SIZE & PAGE_MASK))
1722 address += PAGE_SIZE;
1724 /* Enforce stack_guard_gap */
1725 gap_addr = address + stack_guard_gap;
1727 /* Guard against overflow */
1728 if (gap_addr < address || gap_addr > TASK_SIZE)
1729 gap_addr = TASK_SIZE;
1731 next = vma->vm_next;
1732 if (next && next->vm_start < gap_addr &&
1733 (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
1734 if (!(next->vm_flags & VM_GROWSUP))
1736 /* Check that both stack segments have the same anon_vma? */
1739 /* We must make sure the anon_vma is allocated. */
1740 if (unlikely(anon_vma_prepare(vma)))
1744 * vma->vm_start/vm_end cannot change under us because the caller
1745 * is required to hold the mmap_sem in read mode. We need the
1746 * anon_vma lock to serialize against concurrent expand_stacks.
1748 vma_lock_anon_vma(vma);
1750 /* Somebody else might have raced and expanded it already */
1751 if (address > vma->vm_end) {
1752 unsigned long size, grow;
1754 size = address - vma->vm_start;
1755 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1758 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1759 error = acct_stack_growth(vma, size, grow);
1761 vma->vm_end = address;
1762 perf_event_mmap(vma);
1766 vma_unlock_anon_vma(vma);
1767 khugepaged_enter_vma_merge(vma, vma->vm_flags);
1770 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1773 * vma is the first one with address < vma->vm_start. Have to extend vma.
1775 int expand_downwards(struct vm_area_struct *vma,
1776 unsigned long address)
1778 struct vm_area_struct *prev;
1779 unsigned long gap_addr;
1782 address &= PAGE_MASK;
1783 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1787 /* Enforce stack_guard_gap */
1788 gap_addr = address - stack_guard_gap;
1789 if (gap_addr > address)
1791 prev = vma->vm_prev;
1792 if (prev && prev->vm_end > gap_addr &&
1793 (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
1794 if (!(prev->vm_flags & VM_GROWSDOWN))
1796 /* Check that both stack segments have the same anon_vma? */
1799 /* We must make sure the anon_vma is allocated. */
1800 if (unlikely(anon_vma_prepare(vma)))
1804 * vma->vm_start/vm_end cannot change under us because the caller
1805 * is required to hold the mmap_sem in read mode. We need the
1806 * anon_vma lock to serialize against concurrent expand_stacks.
1808 vma_lock_anon_vma(vma);
1810 /* Somebody else might have raced and expanded it already */
1811 if (address < vma->vm_start) {
1812 unsigned long size, grow;
1814 size = vma->vm_end - address;
1815 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1818 if (grow <= vma->vm_pgoff) {
1819 error = acct_stack_growth(vma, size, grow);
1821 vma->vm_start = address;
1822 vma->vm_pgoff -= grow;
1823 perf_event_mmap(vma);
1827 vma_unlock_anon_vma(vma);
1828 khugepaged_enter_vma_merge(vma, vma->vm_flags);
1832 /* enforced gap between the expanding stack and other mappings. */
1833 unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
1835 static int __init cmdline_parse_stack_guard_gap(char *p)
1840 val = simple_strtoul(p, &endptr, 10);
1842 stack_guard_gap = val << PAGE_SHIFT;
1846 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
1848 #ifdef CONFIG_STACK_GROWSUP
1849 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1851 return expand_upwards(vma, address);
1854 struct vm_area_struct *
1855 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1857 struct vm_area_struct *vma, *prev;
1860 vma = find_vma_prev(mm, addr, &prev);
1861 if (vma && (vma->vm_start <= addr))
1863 if (!prev || expand_stack(prev, addr))
1865 if (prev->vm_flags & VM_LOCKED) {
1866 mlock_vma_pages_range(prev, addr, prev->vm_end);
1871 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1873 return expand_downwards(vma, address);
1876 struct vm_area_struct *
1877 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1879 struct vm_area_struct * vma;
1880 unsigned long start;
1883 vma = find_vma(mm,addr);
1886 if (vma->vm_start <= addr)
1888 if (!(vma->vm_flags & VM_GROWSDOWN))
1890 start = vma->vm_start;
1891 if (expand_stack(vma, addr))
1893 if (vma->vm_flags & VM_LOCKED) {
1894 mlock_vma_pages_range(vma, addr, start);
1901 * Ok - we have the memory areas we should free on the vma list,
1902 * so release them, and do the vma updates.
1904 * Called with the mm semaphore held.
1906 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1908 /* Update high watermark before we lower total_vm */
1909 update_hiwater_vm(mm);
1911 long nrpages = vma_pages(vma);
1913 mm->total_vm -= nrpages;
1914 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1915 vma = remove_vma(vma);
1921 * Get rid of page table information in the indicated region.
1923 * Called with the mm semaphore held.
1925 static void unmap_region(struct mm_struct *mm,
1926 struct vm_area_struct *vma, struct vm_area_struct *prev,
1927 unsigned long start, unsigned long end)
1929 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1930 struct mmu_gather tlb;
1931 unsigned long nr_accounted = 0;
1934 tlb_gather_mmu(&tlb, mm, 0);
1935 update_hiwater_rss(mm);
1936 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1937 vm_unacct_memory(nr_accounted);
1938 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1939 next ? next->vm_start : 0);
1940 tlb_finish_mmu(&tlb, start, end);
1944 * Create a list of vma's touched by the unmap, removing them from the mm's
1945 * vma list as we go..
1948 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1949 struct vm_area_struct *prev, unsigned long end)
1951 struct vm_area_struct **insertion_point;
1952 struct vm_area_struct *tail_vma = NULL;
1955 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1956 vma->vm_prev = NULL;
1958 rb_erase(&vma->vm_rb, &mm->mm_rb);
1962 } while (vma && vma->vm_start < end);
1963 *insertion_point = vma;
1965 vma->vm_prev = prev;
1966 tail_vma->vm_next = NULL;
1967 if (mm->unmap_area == arch_unmap_area)
1968 addr = prev ? prev->vm_end : mm->mmap_base;
1970 addr = vma ? vma->vm_start : mm->mmap_base;
1971 mm->unmap_area(mm, addr);
1972 mm->mmap_cache = NULL; /* Kill the cache. */
1976 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1977 * munmap path where it doesn't make sense to fail.
1979 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1980 unsigned long addr, int new_below)
1982 struct mempolicy *pol;
1983 struct vm_area_struct *new;
1986 if (is_vm_hugetlb_page(vma) && (addr &
1987 ~(huge_page_mask(hstate_vma(vma)))))
1990 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1994 /* most fields are the same, copy all, and then fixup */
1997 INIT_LIST_HEAD(&new->anon_vma_chain);
2002 new->vm_start = addr;
2003 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2006 pol = mpol_dup(vma_policy(vma));
2011 vma_set_policy(new, pol);
2013 if (anon_vma_clone(new, vma))
2017 get_file(new->vm_file);
2018 if (vma->vm_flags & VM_EXECUTABLE)
2019 added_exe_file_vma(mm);
2022 if (new->vm_ops && new->vm_ops->open)
2023 new->vm_ops->open(new);
2026 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2027 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2029 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2035 /* Clean everything up if vma_adjust failed. */
2036 if (new->vm_ops && new->vm_ops->close)
2037 new->vm_ops->close(new);
2039 if (vma->vm_flags & VM_EXECUTABLE)
2040 removed_exe_file_vma(mm);
2043 unlink_anon_vmas(new);
2047 kmem_cache_free(vm_area_cachep, new);
2053 * Split a vma into two pieces at address 'addr', a new vma is allocated
2054 * either for the first part or the tail.
2056 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2057 unsigned long addr, int new_below)
2059 if (mm->map_count >= sysctl_max_map_count)
2062 return __split_vma(mm, vma, addr, new_below);
2065 /* Munmap is split into 2 main parts -- this part which finds
2066 * what needs doing, and the areas themselves, which do the
2067 * work. This now handles partial unmappings.
2068 * Jeremy Fitzhardinge <jeremy@goop.org>
2070 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2073 struct vm_area_struct *vma, *prev, *last;
2075 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2078 if ((len = PAGE_ALIGN(len)) == 0)
2081 /* Find the first overlapping VMA */
2082 vma = find_vma(mm, start);
2085 prev = vma->vm_prev;
2086 /* we have start < vma->vm_end */
2088 /* if it doesn't overlap, we have nothing.. */
2090 if (vma->vm_start >= end)
2094 * If we need to split any vma, do it now to save pain later.
2096 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2097 * unmapped vm_area_struct will remain in use: so lower split_vma
2098 * places tmp vma above, and higher split_vma places tmp vma below.
2100 if (start > vma->vm_start) {
2104 * Make sure that map_count on return from munmap() will
2105 * not exceed its limit; but let map_count go just above
2106 * its limit temporarily, to help free resources as expected.
2108 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2111 error = __split_vma(mm, vma, start, 0);
2117 /* Does it split the last one? */
2118 last = find_vma(mm, end);
2119 if (last && end > last->vm_start) {
2120 int error = __split_vma(mm, last, end, 1);
2124 vma = prev? prev->vm_next: mm->mmap;
2127 * unlock any mlock()ed ranges before detaching vmas
2129 if (mm->locked_vm) {
2130 struct vm_area_struct *tmp = vma;
2131 while (tmp && tmp->vm_start < end) {
2132 if (tmp->vm_flags & VM_LOCKED) {
2133 mm->locked_vm -= vma_pages(tmp);
2134 munlock_vma_pages_all(tmp);
2141 * Remove the vma's, and unmap the actual pages
2143 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2144 unmap_region(mm, vma, prev, start, end);
2146 /* Fix up all other VM information */
2147 remove_vma_list(mm, vma);
2152 EXPORT_SYMBOL(do_munmap);
2154 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2157 struct mm_struct *mm = current->mm;
2159 profile_munmap(addr);
2161 down_write(&mm->mmap_sem);
2162 ret = do_munmap(mm, addr, len);
2163 up_write(&mm->mmap_sem);
2167 static inline void verify_mm_writelocked(struct mm_struct *mm)
2169 #ifdef CONFIG_DEBUG_VM
2170 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2172 up_read(&mm->mmap_sem);
2178 * this is really a simplified "do_mmap". it only handles
2179 * anonymous maps. eventually we may be able to do some
2180 * brk-specific accounting here.
2182 unsigned long do_brk(unsigned long addr, unsigned long len)
2184 struct mm_struct * mm = current->mm;
2185 struct vm_area_struct * vma, * prev;
2186 unsigned long flags;
2187 struct rb_node ** rb_link, * rb_parent;
2188 pgoff_t pgoff = addr >> PAGE_SHIFT;
2191 len = PAGE_ALIGN(len);
2195 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2199 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2201 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2202 if (error & ~PAGE_MASK)
2208 if (mm->def_flags & VM_LOCKED) {
2209 unsigned long locked, lock_limit;
2210 locked = len >> PAGE_SHIFT;
2211 locked += mm->locked_vm;
2212 lock_limit = rlimit(RLIMIT_MEMLOCK);
2213 lock_limit >>= PAGE_SHIFT;
2214 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2219 * mm->mmap_sem is required to protect against another thread
2220 * changing the mappings in case we sleep.
2222 verify_mm_writelocked(mm);
2225 * Clear old maps. this also does some error checking for us
2228 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2229 if (vma && vma->vm_start < addr + len) {
2230 if (do_munmap(mm, addr, len))
2235 /* Check against address space limits *after* clearing old maps... */
2236 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2239 if (mm->map_count > sysctl_max_map_count)
2242 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2245 /* Can we just expand an old private anonymous mapping? */
2246 vma = vma_merge(mm, prev, addr, addr + len, flags,
2247 NULL, NULL, pgoff, NULL);
2252 * create a vma struct for an anonymous mapping
2254 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2256 vm_unacct_memory(len >> PAGE_SHIFT);
2260 INIT_LIST_HEAD(&vma->anon_vma_chain);
2262 vma->vm_start = addr;
2263 vma->vm_end = addr + len;
2264 vma->vm_pgoff = pgoff;
2265 vma->vm_flags = flags;
2266 vma->vm_page_prot = vm_get_page_prot(flags);
2267 vma_link(mm, vma, prev, rb_link, rb_parent);
2269 perf_event_mmap(vma);
2270 mm->total_vm += len >> PAGE_SHIFT;
2271 if (flags & VM_LOCKED) {
2272 if (!mlock_vma_pages_range(vma, addr, addr + len))
2273 mm->locked_vm += (len >> PAGE_SHIFT);
2278 EXPORT_SYMBOL(do_brk);
2280 /* Release all mmaps. */
2281 void exit_mmap(struct mm_struct *mm)
2283 struct mmu_gather tlb;
2284 struct vm_area_struct *vma;
2285 unsigned long nr_accounted = 0;
2288 /* mm's last user has gone, and its about to be pulled down */
2289 mmu_notifier_release(mm);
2291 if (mm->locked_vm) {
2294 if (vma->vm_flags & VM_LOCKED)
2295 munlock_vma_pages_all(vma);
2303 if (!vma) /* Can happen if dup_mmap() received an OOM */
2308 tlb_gather_mmu(&tlb, mm, 1);
2309 /* update_hiwater_rss(mm) here? but nobody should be looking */
2310 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2311 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2312 vm_unacct_memory(nr_accounted);
2314 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2315 tlb_finish_mmu(&tlb, 0, end);
2318 * Walk the list again, actually closing and freeing it,
2319 * with preemption enabled, without holding any MM locks.
2322 vma = remove_vma(vma);
2324 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2327 /* Insert vm structure into process list sorted by address
2328 * and into the inode's i_mmap tree. If vm_file is non-NULL
2329 * then i_mmap_mutex is taken here.
2331 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2333 struct vm_area_struct * __vma, * prev;
2334 struct rb_node ** rb_link, * rb_parent;
2337 * The vm_pgoff of a purely anonymous vma should be irrelevant
2338 * until its first write fault, when page's anon_vma and index
2339 * are set. But now set the vm_pgoff it will almost certainly
2340 * end up with (unless mremap moves it elsewhere before that
2341 * first wfault), so /proc/pid/maps tells a consistent story.
2343 * By setting it to reflect the virtual start address of the
2344 * vma, merges and splits can happen in a seamless way, just
2345 * using the existing file pgoff checks and manipulations.
2346 * Similarly in do_mmap_pgoff and in do_brk.
2348 if (!vma->vm_file) {
2349 BUG_ON(vma->anon_vma);
2350 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2352 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2353 if (__vma && __vma->vm_start < vma->vm_end)
2355 if ((vma->vm_flags & VM_ACCOUNT) &&
2356 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2358 vma_link(mm, vma, prev, rb_link, rb_parent);
2363 * Copy the vma structure to a new location in the same mm,
2364 * prior to moving page table entries, to effect an mremap move.
2366 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2367 unsigned long addr, unsigned long len, pgoff_t pgoff)
2369 struct vm_area_struct *vma = *vmap;
2370 unsigned long vma_start = vma->vm_start;
2371 struct mm_struct *mm = vma->vm_mm;
2372 struct vm_area_struct *new_vma, *prev;
2373 struct rb_node **rb_link, *rb_parent;
2374 struct mempolicy *pol;
2377 * If anonymous vma has not yet been faulted, update new pgoff
2378 * to match new location, to increase its chance of merging.
2380 if (!vma->vm_file && !vma->anon_vma)
2381 pgoff = addr >> PAGE_SHIFT;
2383 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2384 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2385 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2388 * Source vma may have been merged into new_vma
2390 if (vma_start >= new_vma->vm_start &&
2391 vma_start < new_vma->vm_end)
2394 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2397 pol = mpol_dup(vma_policy(vma));
2400 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2401 if (anon_vma_clone(new_vma, vma))
2402 goto out_free_mempol;
2403 vma_set_policy(new_vma, pol);
2404 new_vma->vm_start = addr;
2405 new_vma->vm_end = addr + len;
2406 new_vma->vm_pgoff = pgoff;
2407 if (new_vma->vm_file) {
2408 get_file(new_vma->vm_file);
2409 if (vma->vm_flags & VM_EXECUTABLE)
2410 added_exe_file_vma(mm);
2412 if (new_vma->vm_ops && new_vma->vm_ops->open)
2413 new_vma->vm_ops->open(new_vma);
2414 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2422 kmem_cache_free(vm_area_cachep, new_vma);
2427 * Return true if the calling process may expand its vm space by the passed
2430 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2432 unsigned long cur = mm->total_vm; /* pages */
2435 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2437 if (cur + npages > lim)
2443 static int special_mapping_fault(struct vm_area_struct *vma,
2444 struct vm_fault *vmf)
2447 struct page **pages;
2450 * special mappings have no vm_file, and in that case, the mm
2451 * uses vm_pgoff internally. So we have to subtract it from here.
2452 * We are allowed to do this because we are the mm; do not copy
2453 * this code into drivers!
2455 pgoff = vmf->pgoff - vma->vm_pgoff;
2457 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2461 struct page *page = *pages;
2467 return VM_FAULT_SIGBUS;
2471 * Having a close hook prevents vma merging regardless of flags.
2473 static void special_mapping_close(struct vm_area_struct *vma)
2477 static const struct vm_operations_struct special_mapping_vmops = {
2478 .close = special_mapping_close,
2479 .fault = special_mapping_fault,
2483 * Called with mm->mmap_sem held for writing.
2484 * Insert a new vma covering the given region, with the given flags.
2485 * Its pages are supplied by the given array of struct page *.
2486 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2487 * The region past the last page supplied will always produce SIGBUS.
2488 * The array pointer and the pages it points to are assumed to stay alive
2489 * for as long as this mapping might exist.
2491 int install_special_mapping(struct mm_struct *mm,
2492 unsigned long addr, unsigned long len,
2493 unsigned long vm_flags, struct page **pages)
2496 struct vm_area_struct *vma;
2498 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2499 if (unlikely(vma == NULL))
2502 INIT_LIST_HEAD(&vma->anon_vma_chain);
2504 vma->vm_start = addr;
2505 vma->vm_end = addr + len;
2507 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2508 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2510 vma->vm_ops = &special_mapping_vmops;
2511 vma->vm_private_data = pages;
2513 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2517 ret = insert_vm_struct(mm, vma);
2521 mm->total_vm += len >> PAGE_SHIFT;
2523 perf_event_mmap(vma);
2528 kmem_cache_free(vm_area_cachep, vma);
2532 static DEFINE_MUTEX(mm_all_locks_mutex);
2534 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2536 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2538 * The LSB of head.next can't change from under us
2539 * because we hold the mm_all_locks_mutex.
2541 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2543 * We can safely modify head.next after taking the
2544 * anon_vma->root->mutex. If some other vma in this mm shares
2545 * the same anon_vma we won't take it again.
2547 * No need of atomic instructions here, head.next
2548 * can't change from under us thanks to the
2549 * anon_vma->root->mutex.
2551 if (__test_and_set_bit(0, (unsigned long *)
2552 &anon_vma->root->head.next))
2557 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2559 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2561 * AS_MM_ALL_LOCKS can't change from under us because
2562 * we hold the mm_all_locks_mutex.
2564 * Operations on ->flags have to be atomic because
2565 * even if AS_MM_ALL_LOCKS is stable thanks to the
2566 * mm_all_locks_mutex, there may be other cpus
2567 * changing other bitflags in parallel to us.
2569 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2571 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2576 * This operation locks against the VM for all pte/vma/mm related
2577 * operations that could ever happen on a certain mm. This includes
2578 * vmtruncate, try_to_unmap, and all page faults.
2580 * The caller must take the mmap_sem in write mode before calling
2581 * mm_take_all_locks(). The caller isn't allowed to release the
2582 * mmap_sem until mm_drop_all_locks() returns.
2584 * mmap_sem in write mode is required in order to block all operations
2585 * that could modify pagetables and free pages without need of
2586 * altering the vma layout (for example populate_range() with
2587 * nonlinear vmas). It's also needed in write mode to avoid new
2588 * anon_vmas to be associated with existing vmas.
2590 * A single task can't take more than one mm_take_all_locks() in a row
2591 * or it would deadlock.
2593 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2594 * mapping->flags avoid to take the same lock twice, if more than one
2595 * vma in this mm is backed by the same anon_vma or address_space.
2597 * We can take all the locks in random order because the VM code
2598 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2599 * takes more than one of them in a row. Secondly we're protected
2600 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2602 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2603 * that may have to take thousand of locks.
2605 * mm_take_all_locks() can fail if it's interrupted by signals.
2607 int mm_take_all_locks(struct mm_struct *mm)
2609 struct vm_area_struct *vma;
2610 struct anon_vma_chain *avc;
2612 BUG_ON(down_read_trylock(&mm->mmap_sem));
2614 mutex_lock(&mm_all_locks_mutex);
2616 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2617 if (signal_pending(current))
2619 if (vma->vm_file && vma->vm_file->f_mapping)
2620 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2623 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2624 if (signal_pending(current))
2627 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2628 vm_lock_anon_vma(mm, avc->anon_vma);
2634 mm_drop_all_locks(mm);
2638 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2640 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2642 * The LSB of head.next can't change to 0 from under
2643 * us because we hold the mm_all_locks_mutex.
2645 * We must however clear the bitflag before unlocking
2646 * the vma so the users using the anon_vma->head will
2647 * never see our bitflag.
2649 * No need of atomic instructions here, head.next
2650 * can't change from under us until we release the
2651 * anon_vma->root->mutex.
2653 if (!__test_and_clear_bit(0, (unsigned long *)
2654 &anon_vma->root->head.next))
2656 anon_vma_unlock(anon_vma);
2660 static void vm_unlock_mapping(struct address_space *mapping)
2662 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2664 * AS_MM_ALL_LOCKS can't change to 0 from under us
2665 * because we hold the mm_all_locks_mutex.
2667 mutex_unlock(&mapping->i_mmap_mutex);
2668 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2675 * The mmap_sem cannot be released by the caller until
2676 * mm_drop_all_locks() returns.
2678 void mm_drop_all_locks(struct mm_struct *mm)
2680 struct vm_area_struct *vma;
2681 struct anon_vma_chain *avc;
2683 BUG_ON(down_read_trylock(&mm->mmap_sem));
2684 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2686 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2688 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2689 vm_unlock_anon_vma(avc->anon_vma);
2690 if (vma->vm_file && vma->vm_file->f_mapping)
2691 vm_unlock_mapping(vma->vm_file->f_mapping);
2694 mutex_unlock(&mm_all_locks_mutex);
2698 * initialise the VMA slab
2700 void __init mmap_init(void)
2704 ret = percpu_counter_init(&vm_committed_as, 0);