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/module.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 /* No sane architecture will #define these to anything else */
50 #ifndef arch_add_exec_range
51 #define arch_add_exec_range(mm, limit) do { ; } while (0)
53 #ifndef arch_flush_exec_range
54 #define arch_flush_exec_range(mm) do { ; } while (0)
56 #ifndef arch_remove_exec_range
57 #define arch_remove_exec_range(mm, limit) do { ; } while (0)
61 static void unmap_region(struct mm_struct *mm,
62 struct vm_area_struct *vma, struct vm_area_struct *prev,
63 unsigned long start, unsigned long end);
66 * WARNING: the debugging will use recursive algorithms so never enable this
67 * unless you know what you are doing.
71 /* description of effects of mapping type and prot in current implementation.
72 * this is due to the limited x86 page protection hardware. The expected
73 * behavior is in parens:
76 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
77 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
78 * w: (no) no w: (no) no w: (yes) yes w: (no) no
79 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
81 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
82 * w: (no) no w: (no) no w: (copy) copy w: (no) no
83 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
86 pgprot_t protection_map[16] = {
87 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
88 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
91 pgprot_t vm_get_page_prot(unsigned long vm_flags)
93 return __pgprot(pgprot_val(protection_map[vm_flags &
94 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
95 pgprot_val(arch_vm_get_page_prot(vm_flags)));
97 EXPORT_SYMBOL(vm_get_page_prot);
99 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
100 int sysctl_overcommit_ratio = 50; /* default is 50% */
101 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
102 struct percpu_counter vm_committed_as;
105 * Check that a process has enough memory to allocate a new virtual
106 * mapping. 0 means there is enough memory for the allocation to
107 * succeed and -ENOMEM implies there is not.
109 * We currently support three overcommit policies, which are set via the
110 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
112 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
113 * Additional code 2002 Jul 20 by Robert Love.
115 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
117 * Note this is a helper function intended to be used by LSMs which
118 * wish to use this logic.
120 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
122 unsigned long free, allowed;
124 vm_acct_memory(pages);
127 * Sometimes we want to use more memory than we have
129 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
132 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
135 free = global_page_state(NR_FILE_PAGES);
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 the last 3% for root
156 * nr_free_pages() is very expensive on large systems,
157 * only call if we're about to fail.
162 * Leave reserved pages. The pages are not for anonymous pages.
164 if (n <= totalreserve_pages)
167 n -= totalreserve_pages;
170 * Leave the last 3% for root
182 allowed = (totalram_pages - hugetlb_total_pages())
183 * sysctl_overcommit_ratio / 100;
185 * Leave the last 3% for root
188 allowed -= allowed / 32;
189 allowed += total_swap_pages;
191 /* Don't let a single process grow too big:
192 leave 3% of the size of this process for other processes */
194 allowed -= mm->total_vm / 32;
196 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
199 vm_unacct_memory(pages);
205 * Requires inode->i_mapping->i_mmap_lock
207 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
208 struct file *file, struct address_space *mapping)
210 if (vma->vm_flags & VM_DENYWRITE)
211 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
212 if (vma->vm_flags & VM_SHARED)
213 mapping->i_mmap_writable--;
215 flush_dcache_mmap_lock(mapping);
216 if (unlikely(vma->vm_flags & VM_NONLINEAR))
217 list_del_init(&vma->shared.vm_set.list);
219 vma_prio_tree_remove(vma, &mapping->i_mmap);
220 flush_dcache_mmap_unlock(mapping);
224 * Unlink a file-based vm structure from its prio_tree, to hide
225 * vma from rmap and vmtruncate before freeing its page tables.
227 void unlink_file_vma(struct vm_area_struct *vma)
229 struct file *file = vma->vm_file;
232 struct address_space *mapping = file->f_mapping;
233 spin_lock(&mapping->i_mmap_lock);
234 __remove_shared_vm_struct(vma, file, mapping);
235 spin_unlock(&mapping->i_mmap_lock);
240 * Close a vm structure and free it, returning the next.
242 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
244 struct vm_area_struct *next = vma->vm_next;
247 if (vma->vm_ops && vma->vm_ops->close)
248 vma->vm_ops->close(vma);
251 if (vma->vm_flags & VM_EXECUTABLE)
252 removed_exe_file_vma(vma->vm_mm);
254 mpol_put(vma_policy(vma));
255 kmem_cache_free(vm_area_cachep, vma);
259 SYSCALL_DEFINE1(brk, unsigned long, brk)
261 unsigned long rlim, retval;
262 unsigned long newbrk, oldbrk;
263 struct mm_struct *mm = current->mm;
264 unsigned long min_brk;
266 down_write(&mm->mmap_sem);
268 #ifdef CONFIG_COMPAT_BRK
270 * CONFIG_COMPAT_BRK can still be overridden by setting
271 * randomize_va_space to 2, which will still cause mm->start_brk
272 * to be arbitrarily shifted
274 if (mm->start_brk > PAGE_ALIGN(mm->end_data))
275 min_brk = mm->start_brk;
277 min_brk = mm->end_data;
279 min_brk = mm->start_brk;
285 * Check against rlimit here. If this check is done later after the test
286 * of oldbrk with newbrk then it can escape the test and let the data
287 * segment grow beyond its set limit the in case where the limit is
288 * not page aligned -Ram Gupta
290 rlim = rlimit(RLIMIT_DATA);
291 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
292 (mm->end_data - mm->start_data) > rlim)
295 newbrk = PAGE_ALIGN(brk);
296 oldbrk = PAGE_ALIGN(mm->brk);
297 if (oldbrk == newbrk)
300 /* Always allow shrinking brk. */
301 if (brk <= mm->brk) {
302 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
307 /* Check against existing mmap mappings. */
308 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
311 /* Ok, looks good - let it rip. */
312 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
318 up_write(&mm->mmap_sem);
323 static int browse_rb(struct rb_root *root)
326 struct rb_node *nd, *pn = NULL;
327 unsigned long prev = 0, pend = 0;
329 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
330 struct vm_area_struct *vma;
331 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
332 if (vma->vm_start < prev)
333 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
334 if (vma->vm_start < pend)
335 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
336 if (vma->vm_start > vma->vm_end)
337 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
340 prev = vma->vm_start;
344 for (nd = pn; nd; nd = rb_prev(nd)) {
348 printk("backwards %d, forwards %d\n", j, i), i = 0;
352 void validate_mm(struct mm_struct *mm)
356 struct vm_area_struct *tmp = mm->mmap;
361 if (i != mm->map_count)
362 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
363 i = browse_rb(&mm->mm_rb);
364 if (i != mm->map_count)
365 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
369 #define validate_mm(mm) do { } while (0)
372 static struct vm_area_struct *
373 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
374 struct vm_area_struct **pprev, struct rb_node ***rb_link,
375 struct rb_node ** rb_parent)
377 struct vm_area_struct * vma;
378 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
380 __rb_link = &mm->mm_rb.rb_node;
381 rb_prev = __rb_parent = NULL;
385 struct vm_area_struct *vma_tmp;
387 __rb_parent = *__rb_link;
388 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
390 if (vma_tmp->vm_end > addr) {
392 if (vma_tmp->vm_start <= addr)
394 __rb_link = &__rb_parent->rb_left;
396 rb_prev = __rb_parent;
397 __rb_link = &__rb_parent->rb_right;
403 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
404 *rb_link = __rb_link;
405 *rb_parent = __rb_parent;
410 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
411 struct vm_area_struct *prev, struct rb_node *rb_parent)
413 struct vm_area_struct *next;
415 if (vma->vm_flags & VM_EXEC)
416 arch_add_exec_range(mm, vma->vm_end);
420 next = prev->vm_next;
425 next = rb_entry(rb_parent,
426 struct vm_area_struct, vm_rb);
435 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
436 struct rb_node **rb_link, struct rb_node *rb_parent)
438 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
439 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
442 static void __vma_link_file(struct vm_area_struct *vma)
448 struct address_space *mapping = file->f_mapping;
450 if (vma->vm_flags & VM_DENYWRITE)
451 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
452 if (vma->vm_flags & VM_SHARED)
453 mapping->i_mmap_writable++;
455 flush_dcache_mmap_lock(mapping);
456 if (unlikely(vma->vm_flags & VM_NONLINEAR))
457 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
459 vma_prio_tree_insert(vma, &mapping->i_mmap);
460 flush_dcache_mmap_unlock(mapping);
465 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
466 struct vm_area_struct *prev, struct rb_node **rb_link,
467 struct rb_node *rb_parent)
469 __vma_link_list(mm, vma, prev, rb_parent);
470 __vma_link_rb(mm, vma, rb_link, rb_parent);
473 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
474 struct vm_area_struct *prev, struct rb_node **rb_link,
475 struct rb_node *rb_parent)
477 struct address_space *mapping = NULL;
480 mapping = vma->vm_file->f_mapping;
483 spin_lock(&mapping->i_mmap_lock);
484 vma->vm_truncate_count = mapping->truncate_count;
487 __vma_link(mm, vma, prev, rb_link, rb_parent);
488 __vma_link_file(vma);
491 spin_unlock(&mapping->i_mmap_lock);
498 * Helper for vma_adjust in the split_vma insert case:
499 * insert vm structure into list and rbtree and anon_vma,
500 * but it has already been inserted into prio_tree earlier.
502 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
504 struct vm_area_struct *__vma, *prev;
505 struct rb_node **rb_link, *rb_parent;
507 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
508 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
509 __vma_link(mm, vma, prev, rb_link, rb_parent);
514 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
515 struct vm_area_struct *prev)
517 struct vm_area_struct *next = vma->vm_next;
519 prev->vm_next = next;
521 next->vm_prev = prev;
522 rb_erase(&vma->vm_rb, &mm->mm_rb);
523 if (mm->mmap_cache == vma)
524 mm->mmap_cache = prev;
525 if (vma->vm_flags & VM_EXEC)
526 arch_remove_exec_range(mm, vma->vm_end);
530 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
531 * is already present in an i_mmap tree without adjusting the tree.
532 * The following helper function should be used when such adjustments
533 * are necessary. The "insert" vma (if any) is to be inserted
534 * before we drop the necessary locks.
536 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
537 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
539 struct mm_struct *mm = vma->vm_mm;
540 struct vm_area_struct *next = vma->vm_next;
541 struct vm_area_struct *importer = NULL;
542 struct address_space *mapping = NULL;
543 struct prio_tree_root *root = NULL;
544 struct anon_vma *anon_vma = NULL;
545 struct file *file = vma->vm_file;
546 long adjust_next = 0;
549 if (next && !insert) {
550 struct vm_area_struct *exporter = NULL;
552 if (end >= next->vm_end) {
554 * vma expands, overlapping all the next, and
555 * perhaps the one after too (mprotect case 6).
557 again: remove_next = 1 + (end > next->vm_end);
561 } else if (end > next->vm_start) {
563 * vma expands, overlapping part of the next:
564 * mprotect case 5 shifting the boundary up.
566 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
569 } else if (end < vma->vm_end) {
571 * vma shrinks, and !insert tells it's not
572 * split_vma inserting another: so it must be
573 * mprotect case 4 shifting the boundary down.
575 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
581 * Easily overlooked: when mprotect shifts the boundary,
582 * make sure the expanding vma has anon_vma set if the
583 * shrinking vma had, to cover any anon pages imported.
585 if (exporter && exporter->anon_vma && !importer->anon_vma) {
586 if (anon_vma_clone(importer, exporter))
588 importer->anon_vma = exporter->anon_vma;
593 mapping = file->f_mapping;
594 if (!(vma->vm_flags & VM_NONLINEAR))
595 root = &mapping->i_mmap;
596 spin_lock(&mapping->i_mmap_lock);
598 vma->vm_truncate_count != next->vm_truncate_count) {
600 * unmap_mapping_range might be in progress:
601 * ensure that the expanding vma is rescanned.
603 importer->vm_truncate_count = 0;
606 insert->vm_truncate_count = vma->vm_truncate_count;
608 * Put into prio_tree now, so instantiated pages
609 * are visible to arm/parisc __flush_dcache_page
610 * throughout; but we cannot insert into address
611 * space until vma start or end is updated.
613 __vma_link_file(insert);
617 vma_adjust_trans_huge(vma, start, end, adjust_next);
620 * When changing only vma->vm_end, we don't really need anon_vma
621 * lock. This is a fairly rare case by itself, but the anon_vma
622 * lock may be shared between many sibling processes. Skipping
623 * the lock for brk adjustments makes a difference sometimes.
625 if (vma->anon_vma && (insert || importer || start != vma->vm_start)) {
626 anon_vma = vma->anon_vma;
627 anon_vma_lock(anon_vma);
631 flush_dcache_mmap_lock(mapping);
632 vma_prio_tree_remove(vma, root);
634 vma_prio_tree_remove(next, root);
637 vma->vm_start = start;
639 vma->vm_pgoff = pgoff;
641 next->vm_start += adjust_next << PAGE_SHIFT;
642 next->vm_pgoff += adjust_next;
647 vma_prio_tree_insert(next, root);
648 vma_prio_tree_insert(vma, root);
649 flush_dcache_mmap_unlock(mapping);
654 * vma_merge has merged next into vma, and needs
655 * us to remove next before dropping the locks.
657 __vma_unlink(mm, next, vma);
659 __remove_shared_vm_struct(next, file, mapping);
662 * split_vma has split insert from vma, and needs
663 * us to insert it before dropping the locks
664 * (it may either follow vma or precede it).
666 __insert_vm_struct(mm, insert);
670 anon_vma_unlock(anon_vma);
672 spin_unlock(&mapping->i_mmap_lock);
677 if (next->vm_flags & VM_EXECUTABLE)
678 removed_exe_file_vma(mm);
681 anon_vma_merge(vma, next);
683 mpol_put(vma_policy(next));
684 kmem_cache_free(vm_area_cachep, next);
686 * In mprotect's case 6 (see comments on vma_merge),
687 * we must remove another next too. It would clutter
688 * up the code too much to do both in one go.
690 if (remove_next == 2) {
702 * If the vma has a ->close operation then the driver probably needs to release
703 * per-vma resources, so we don't attempt to merge those.
705 static inline int is_mergeable_vma(struct vm_area_struct *vma,
706 struct file *file, unsigned long vm_flags)
708 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
709 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
711 if (vma->vm_file != file)
713 if (vma->vm_ops && vma->vm_ops->close)
718 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
719 struct anon_vma *anon_vma2)
721 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
725 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
726 * in front of (at a lower virtual address and file offset than) the vma.
728 * We cannot merge two vmas if they have differently assigned (non-NULL)
729 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
731 * We don't check here for the merged mmap wrapping around the end of pagecache
732 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
733 * wrap, nor mmaps which cover the final page at index -1UL.
736 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
737 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
739 if (is_mergeable_vma(vma, file, vm_flags) &&
740 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
741 if (vma->vm_pgoff == vm_pgoff)
748 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
749 * beyond (at a higher virtual address and file offset than) the vma.
751 * We cannot merge two vmas if they have differently assigned (non-NULL)
752 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
755 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
756 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
758 if (is_mergeable_vma(vma, file, vm_flags) &&
759 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
761 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
762 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
769 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
770 * whether that can be merged with its predecessor or its successor.
771 * Or both (it neatly fills a hole).
773 * In most cases - when called for mmap, brk or mremap - [addr,end) is
774 * certain not to be mapped by the time vma_merge is called; but when
775 * called for mprotect, it is certain to be already mapped (either at
776 * an offset within prev, or at the start of next), and the flags of
777 * this area are about to be changed to vm_flags - and the no-change
778 * case has already been eliminated.
780 * The following mprotect cases have to be considered, where AAAA is
781 * the area passed down from mprotect_fixup, never extending beyond one
782 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
784 * AAAA AAAA AAAA AAAA
785 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
786 * cannot merge might become might become might become
787 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
788 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
789 * mremap move: PPPPNNNNNNNN 8
791 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
792 * might become case 1 below case 2 below case 3 below
794 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
795 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
797 struct vm_area_struct *vma_merge(struct mm_struct *mm,
798 struct vm_area_struct *prev, unsigned long addr,
799 unsigned long end, unsigned long vm_flags,
800 struct anon_vma *anon_vma, struct file *file,
801 pgoff_t pgoff, struct mempolicy *policy)
803 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
804 struct vm_area_struct *area, *next;
808 * We later require that vma->vm_flags == vm_flags,
809 * so this tests vma->vm_flags & VM_SPECIAL, too.
811 if (vm_flags & VM_SPECIAL)
815 next = prev->vm_next;
819 if (next && next->vm_end == end) /* cases 6, 7, 8 */
820 next = next->vm_next;
823 * Can it merge with the predecessor?
825 if (prev && prev->vm_end == addr &&
826 mpol_equal(vma_policy(prev), policy) &&
827 can_vma_merge_after(prev, vm_flags,
828 anon_vma, file, pgoff)) {
830 * OK, it can. Can we now merge in the successor as well?
832 if (next && end == next->vm_start &&
833 mpol_equal(policy, vma_policy(next)) &&
834 can_vma_merge_before(next, vm_flags,
835 anon_vma, file, pgoff+pglen) &&
836 is_mergeable_anon_vma(prev->anon_vma,
839 err = vma_adjust(prev, prev->vm_start,
840 next->vm_end, prev->vm_pgoff, NULL);
841 } else /* cases 2, 5, 7 */
842 err = vma_adjust(prev, prev->vm_start,
843 end, prev->vm_pgoff, NULL);
844 if (prev->vm_flags & VM_EXEC)
845 arch_add_exec_range(mm, prev->vm_end);
848 khugepaged_enter_vma_merge(prev);
853 * Can this new request be merged in front of next?
855 if (next && end == next->vm_start &&
856 mpol_equal(policy, vma_policy(next)) &&
857 can_vma_merge_before(next, vm_flags,
858 anon_vma, file, pgoff+pglen)) {
859 if (prev && addr < prev->vm_end) /* case 4 */
860 err = vma_adjust(prev, prev->vm_start,
861 addr, prev->vm_pgoff, NULL);
862 else /* cases 3, 8 */
863 err = vma_adjust(area, addr, next->vm_end,
864 next->vm_pgoff - pglen, NULL);
867 khugepaged_enter_vma_merge(area);
875 * Rough compatbility check to quickly see if it's even worth looking
876 * at sharing an anon_vma.
878 * They need to have the same vm_file, and the flags can only differ
879 * in things that mprotect may change.
881 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
882 * we can merge the two vma's. For example, we refuse to merge a vma if
883 * there is a vm_ops->close() function, because that indicates that the
884 * driver is doing some kind of reference counting. But that doesn't
885 * really matter for the anon_vma sharing case.
887 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
889 return a->vm_end == b->vm_start &&
890 mpol_equal(vma_policy(a), vma_policy(b)) &&
891 a->vm_file == b->vm_file &&
892 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
893 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
897 * Do some basic sanity checking to see if we can re-use the anon_vma
898 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
899 * the same as 'old', the other will be the new one that is trying
900 * to share the anon_vma.
902 * NOTE! This runs with mm_sem held for reading, so it is possible that
903 * the anon_vma of 'old' is concurrently in the process of being set up
904 * by another page fault trying to merge _that_. But that's ok: if it
905 * is being set up, that automatically means that it will be a singleton
906 * acceptable for merging, so we can do all of this optimistically. But
907 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
909 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
910 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
911 * is to return an anon_vma that is "complex" due to having gone through
914 * We also make sure that the two vma's are compatible (adjacent,
915 * and with the same memory policies). That's all stable, even with just
916 * a read lock on the mm_sem.
918 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
920 if (anon_vma_compatible(a, b)) {
921 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
923 if (anon_vma && list_is_singular(&old->anon_vma_chain))
930 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
931 * neighbouring vmas for a suitable anon_vma, before it goes off
932 * to allocate a new anon_vma. It checks because a repetitive
933 * sequence of mprotects and faults may otherwise lead to distinct
934 * anon_vmas being allocated, preventing vma merge in subsequent
937 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
939 struct anon_vma *anon_vma;
940 struct vm_area_struct *near;
946 anon_vma = reusable_anon_vma(near, vma, near);
951 * It is potentially slow to have to call find_vma_prev here.
952 * But it's only on the first write fault on the vma, not
953 * every time, and we could devise a way to avoid it later
954 * (e.g. stash info in next's anon_vma_node when assigning
955 * an anon_vma, or when trying vma_merge). Another time.
957 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
961 anon_vma = reusable_anon_vma(near, near, vma);
966 * There's no absolute need to look only at touching neighbours:
967 * we could search further afield for "compatible" anon_vmas.
968 * But it would probably just be a waste of time searching,
969 * or lead to too many vmas hanging off the same anon_vma.
970 * We're trying to allow mprotect remerging later on,
971 * not trying to minimize memory used for anon_vmas.
976 #ifdef CONFIG_PROC_FS
977 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
978 struct file *file, long pages)
980 const unsigned long stack_flags
981 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
984 mm->shared_vm += pages;
985 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
986 mm->exec_vm += pages;
987 } else if (flags & stack_flags)
988 mm->stack_vm += pages;
989 if (flags & (VM_RESERVED|VM_IO))
990 mm->reserved_vm += pages;
992 #endif /* CONFIG_PROC_FS */
995 * The caller must hold down_write(¤t->mm->mmap_sem).
998 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
999 unsigned long len, unsigned long prot,
1000 unsigned long flags, unsigned long pgoff)
1002 struct mm_struct * mm = current->mm;
1003 struct inode *inode;
1004 unsigned int vm_flags;
1006 unsigned long reqprot = prot;
1009 * Does the application expect PROT_READ to imply PROT_EXEC?
1011 * (the exception is when the underlying filesystem is noexec
1012 * mounted, in which case we dont add PROT_EXEC.)
1014 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1015 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
1021 if (!(flags & MAP_FIXED))
1022 addr = round_hint_to_min(addr);
1024 /* Careful about overflows.. */
1025 len = PAGE_ALIGN(len);
1029 /* offset overflow? */
1030 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1033 /* Too many mappings? */
1034 if (mm->map_count > sysctl_max_map_count)
1037 /* Obtain the address to map to. we verify (or select) it and ensure
1038 * that it represents a valid section of the address space.
1040 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1041 if (addr & ~PAGE_MASK)
1044 /* Do simple checking here so the lower-level routines won't have
1045 * to. we assume access permissions have been handled by the open
1046 * of the memory object, so we don't do any here.
1048 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1049 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1051 if (flags & MAP_LOCKED)
1052 if (!can_do_mlock())
1055 /* mlock MCL_FUTURE? */
1056 if (vm_flags & VM_LOCKED) {
1057 unsigned long locked, lock_limit;
1058 locked = len >> PAGE_SHIFT;
1059 locked += mm->locked_vm;
1060 lock_limit = rlimit(RLIMIT_MEMLOCK);
1061 lock_limit >>= PAGE_SHIFT;
1062 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1066 inode = file ? file->f_path.dentry->d_inode : NULL;
1069 switch (flags & MAP_TYPE) {
1071 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1075 * Make sure we don't allow writing to an append-only
1078 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1082 * Make sure there are no mandatory locks on the file.
1084 if (locks_verify_locked(inode))
1087 vm_flags |= VM_SHARED | VM_MAYSHARE;
1088 if (!(file->f_mode & FMODE_WRITE))
1089 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1093 if (!(file->f_mode & FMODE_READ))
1095 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1096 if (vm_flags & VM_EXEC)
1098 vm_flags &= ~VM_MAYEXEC;
1101 if (!file->f_op || !file->f_op->mmap)
1109 switch (flags & MAP_TYPE) {
1115 vm_flags |= VM_SHARED | VM_MAYSHARE;
1119 * Set pgoff according to addr for anon_vma.
1121 pgoff = addr >> PAGE_SHIFT;
1128 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1132 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1134 EXPORT_SYMBOL(do_mmap_pgoff);
1136 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1137 unsigned long, prot, unsigned long, flags,
1138 unsigned long, fd, unsigned long, pgoff)
1140 struct file *file = NULL;
1141 unsigned long retval = -EBADF;
1143 if (!(flags & MAP_ANONYMOUS)) {
1144 audit_mmap_fd(fd, flags);
1145 if (unlikely(flags & MAP_HUGETLB))
1150 } else if (flags & MAP_HUGETLB) {
1151 struct user_struct *user = NULL;
1153 * VM_NORESERVE is used because the reservations will be
1154 * taken when vm_ops->mmap() is called
1155 * A dummy user value is used because we are not locking
1156 * memory so no accounting is necessary
1158 len = ALIGN(len, huge_page_size(&default_hstate));
1159 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1160 &user, HUGETLB_ANONHUGE_INODE);
1162 return PTR_ERR(file);
1165 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1167 down_write(¤t->mm->mmap_sem);
1168 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1169 up_write(¤t->mm->mmap_sem);
1177 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1178 struct mmap_arg_struct {
1182 unsigned long flags;
1184 unsigned long offset;
1187 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1189 struct mmap_arg_struct a;
1191 if (copy_from_user(&a, arg, sizeof(a)))
1193 if (a.offset & ~PAGE_MASK)
1196 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1197 a.offset >> PAGE_SHIFT);
1199 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1202 * Some shared mappigns will want the pages marked read-only
1203 * to track write events. If so, we'll downgrade vm_page_prot
1204 * to the private version (using protection_map[] without the
1207 int vma_wants_writenotify(struct vm_area_struct *vma)
1209 unsigned int vm_flags = vma->vm_flags;
1211 /* If it was private or non-writable, the write bit is already clear */
1212 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1215 /* The backer wishes to know when pages are first written to? */
1216 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1219 /* The open routine did something to the protections already? */
1220 if (pgprot_val(vma->vm_page_prot) !=
1221 pgprot_val(vm_get_page_prot(vm_flags)))
1224 /* Specialty mapping? */
1225 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1228 /* Can the mapping track the dirty pages? */
1229 return vma->vm_file && vma->vm_file->f_mapping &&
1230 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1234 * We account for memory if it's a private writeable mapping,
1235 * not hugepages and VM_NORESERVE wasn't set.
1237 static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
1240 * hugetlb has its own accounting separate from the core VM
1241 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1243 if (file && is_file_hugepages(file))
1246 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1249 unsigned long mmap_region(struct file *file, unsigned long addr,
1250 unsigned long len, unsigned long flags,
1251 unsigned int vm_flags, unsigned long pgoff)
1253 struct mm_struct *mm = current->mm;
1254 struct vm_area_struct *vma, *prev;
1255 int correct_wcount = 0;
1257 struct rb_node **rb_link, *rb_parent;
1258 unsigned long charged = 0;
1259 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1261 /* Clear old maps */
1264 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1265 if (vma && vma->vm_start < addr + len) {
1266 if (do_munmap(mm, addr, len))
1271 /* Check against address space limit. */
1272 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1276 * Set 'VM_NORESERVE' if we should not account for the
1277 * memory use of this mapping.
1279 if ((flags & MAP_NORESERVE)) {
1280 /* We honor MAP_NORESERVE if allowed to overcommit */
1281 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1282 vm_flags |= VM_NORESERVE;
1284 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1285 if (file && is_file_hugepages(file))
1286 vm_flags |= VM_NORESERVE;
1290 * Private writable mapping: check memory availability
1292 if (accountable_mapping(file, vm_flags)) {
1293 charged = len >> PAGE_SHIFT;
1294 if (security_vm_enough_memory(charged))
1296 vm_flags |= VM_ACCOUNT;
1300 * Can we just expand an old mapping?
1302 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1307 * Determine the object being mapped and call the appropriate
1308 * specific mapper. the address has already been validated, but
1309 * not unmapped, but the maps are removed from the list.
1311 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1318 vma->vm_start = addr;
1319 vma->vm_end = addr + len;
1320 vma->vm_flags = vm_flags;
1321 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1322 vma->vm_pgoff = pgoff;
1323 INIT_LIST_HEAD(&vma->anon_vma_chain);
1327 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1329 if (vm_flags & VM_DENYWRITE) {
1330 error = deny_write_access(file);
1335 vma->vm_file = file;
1337 error = file->f_op->mmap(file, vma);
1339 goto unmap_and_free_vma;
1340 if (vm_flags & VM_EXECUTABLE)
1341 added_exe_file_vma(mm);
1343 /* Can addr have changed??
1345 * Answer: Yes, several device drivers can do it in their
1346 * f_op->mmap method. -DaveM
1348 addr = vma->vm_start;
1349 pgoff = vma->vm_pgoff;
1350 vm_flags = vma->vm_flags;
1351 } else if (vm_flags & VM_SHARED) {
1352 error = shmem_zero_setup(vma);
1357 if (vma_wants_writenotify(vma)) {
1358 pgprot_t pprot = vma->vm_page_prot;
1360 /* Can vma->vm_page_prot have changed??
1362 * Answer: Yes, drivers may have changed it in their
1363 * f_op->mmap method.
1365 * Ensures that vmas marked as uncached stay that way.
1367 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1368 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1369 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1372 vma_link(mm, vma, prev, rb_link, rb_parent);
1373 file = vma->vm_file;
1375 /* Once vma denies write, undo our temporary denial count */
1377 atomic_inc(&inode->i_writecount);
1379 perf_event_mmap(vma);
1381 mm->total_vm += len >> PAGE_SHIFT;
1382 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1383 if (vm_flags & VM_LOCKED) {
1384 if (!mlock_vma_pages_range(vma, addr, addr + len))
1385 mm->locked_vm += (len >> PAGE_SHIFT);
1386 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1387 make_pages_present(addr, addr + len);
1392 atomic_inc(&inode->i_writecount);
1393 vma->vm_file = NULL;
1396 /* Undo any partial mapping done by a device driver. */
1397 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1400 kmem_cache_free(vm_area_cachep, vma);
1403 vm_unacct_memory(charged);
1407 /* Get an address range which is currently unmapped.
1408 * For shmat() with addr=0.
1410 * Ugly calling convention alert:
1411 * Return value with the low bits set means error value,
1413 * if (ret & ~PAGE_MASK)
1416 * This function "knows" that -ENOMEM has the bits set.
1418 #ifndef HAVE_ARCH_UNMAPPED_AREA
1420 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1421 unsigned long len, unsigned long pgoff, unsigned long flags)
1423 struct mm_struct *mm = current->mm;
1424 struct vm_area_struct *vma;
1425 unsigned long start_addr;
1427 if (len > TASK_SIZE)
1430 if (flags & MAP_FIXED)
1434 addr = PAGE_ALIGN(addr);
1435 vma = find_vma(mm, addr);
1436 if (TASK_SIZE - len >= addr &&
1437 (!vma || addr + len <= vma->vm_start))
1440 if (len > mm->cached_hole_size) {
1441 start_addr = addr = mm->free_area_cache;
1443 start_addr = addr = TASK_UNMAPPED_BASE;
1444 mm->cached_hole_size = 0;
1448 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1449 /* At this point: (!vma || addr < vma->vm_end). */
1450 if (TASK_SIZE - len < addr) {
1452 * Start a new search - just in case we missed
1455 if (start_addr != TASK_UNMAPPED_BASE) {
1456 addr = TASK_UNMAPPED_BASE;
1458 mm->cached_hole_size = 0;
1463 if (!vma || addr + len <= vma->vm_start) {
1465 * Remember the place where we stopped the search:
1467 mm->free_area_cache = addr + len;
1470 if (addr + mm->cached_hole_size < vma->vm_start)
1471 mm->cached_hole_size = vma->vm_start - addr;
1477 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1480 * Is this a new hole at the lowest possible address?
1482 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1483 mm->free_area_cache = addr;
1484 mm->cached_hole_size = ~0UL;
1489 * This mmap-allocator allocates new areas top-down from below the
1490 * stack's low limit (the base):
1492 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1494 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1495 const unsigned long len, const unsigned long pgoff,
1496 const unsigned long flags)
1498 struct vm_area_struct *vma;
1499 struct mm_struct *mm = current->mm;
1500 unsigned long addr = addr0;
1502 /* requested length too big for entire address space */
1503 if (len > TASK_SIZE)
1506 if (flags & MAP_FIXED)
1509 /* requesting a specific address */
1511 addr = PAGE_ALIGN(addr);
1512 vma = find_vma(mm, addr);
1513 if (TASK_SIZE - len >= addr &&
1514 (!vma || addr + len <= vma->vm_start))
1518 /* check if free_area_cache is useful for us */
1519 if (len <= mm->cached_hole_size) {
1520 mm->cached_hole_size = 0;
1521 mm->free_area_cache = mm->mmap_base;
1524 /* either no address requested or can't fit in requested address hole */
1525 addr = mm->free_area_cache;
1527 /* make sure it can fit in the remaining address space */
1529 vma = find_vma(mm, addr-len);
1530 if (!vma || addr <= vma->vm_start)
1531 /* remember the address as a hint for next time */
1532 return (mm->free_area_cache = addr-len);
1535 if (mm->mmap_base < len)
1538 addr = mm->mmap_base-len;
1542 * Lookup failure means no vma is above this address,
1543 * else if new region fits below vma->vm_start,
1544 * return with success:
1546 vma = find_vma(mm, addr);
1547 if (!vma || addr+len <= vma->vm_start)
1548 /* remember the address as a hint for next time */
1549 return (mm->free_area_cache = addr);
1551 /* remember the largest hole we saw so far */
1552 if (addr + mm->cached_hole_size < vma->vm_start)
1553 mm->cached_hole_size = vma->vm_start - addr;
1555 /* try just below the current vma->vm_start */
1556 addr = vma->vm_start-len;
1557 } while (len < vma->vm_start);
1561 * A failed mmap() very likely causes application failure,
1562 * so fall back to the bottom-up function here. This scenario
1563 * can happen with large stack limits and large mmap()
1566 mm->cached_hole_size = ~0UL;
1567 mm->free_area_cache = TASK_UNMAPPED_BASE;
1568 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1570 * Restore the topdown base:
1572 mm->free_area_cache = mm->mmap_base;
1573 mm->cached_hole_size = ~0UL;
1579 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1582 * Is this a new hole at the highest possible address?
1584 if (addr > mm->free_area_cache)
1585 mm->free_area_cache = addr;
1587 /* dont allow allocations above current base */
1588 if (mm->free_area_cache > mm->mmap_base)
1589 mm->free_area_cache = mm->mmap_base;
1593 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1594 unsigned long pgoff, unsigned long flags)
1596 unsigned long (*get_area)(struct file *, unsigned long,
1597 unsigned long, unsigned long, unsigned long);
1599 unsigned long error = arch_mmap_check(addr, len, flags);
1603 /* Careful about overflows.. */
1604 if (len > TASK_SIZE)
1607 get_area = current->mm->get_unmapped_area;
1608 if (file && file->f_op && file->f_op->get_unmapped_area)
1609 get_area = file->f_op->get_unmapped_area;
1610 addr = get_area(file, addr, len, pgoff, flags);
1611 if (IS_ERR_VALUE(addr))
1614 if (addr > TASK_SIZE - len)
1616 if (addr & ~PAGE_MASK)
1619 return arch_rebalance_pgtables(addr, len);
1622 EXPORT_SYMBOL(get_unmapped_area);
1624 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1625 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1627 struct vm_area_struct *vma = NULL;
1630 /* Check the cache first. */
1631 /* (Cache hit rate is typically around 35%.) */
1632 vma = mm->mmap_cache;
1633 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1634 struct rb_node * rb_node;
1636 rb_node = mm->mm_rb.rb_node;
1640 struct vm_area_struct * vma_tmp;
1642 vma_tmp = rb_entry(rb_node,
1643 struct vm_area_struct, vm_rb);
1645 if (vma_tmp->vm_end > addr) {
1647 if (vma_tmp->vm_start <= addr)
1649 rb_node = rb_node->rb_left;
1651 rb_node = rb_node->rb_right;
1654 mm->mmap_cache = vma;
1660 EXPORT_SYMBOL(find_vma);
1662 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1663 struct vm_area_struct *
1664 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1665 struct vm_area_struct **pprev)
1667 struct vm_area_struct *vma = NULL, *prev = NULL;
1668 struct rb_node *rb_node;
1672 /* Guard against addr being lower than the first VMA */
1675 /* Go through the RB tree quickly. */
1676 rb_node = mm->mm_rb.rb_node;
1679 struct vm_area_struct *vma_tmp;
1680 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1682 if (addr < vma_tmp->vm_end) {
1683 rb_node = rb_node->rb_left;
1686 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1688 rb_node = rb_node->rb_right;
1694 return prev ? prev->vm_next : vma;
1698 * Verify that the stack growth is acceptable and
1699 * update accounting. This is shared with both the
1700 * grow-up and grow-down cases.
1702 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1704 struct mm_struct *mm = vma->vm_mm;
1705 struct rlimit *rlim = current->signal->rlim;
1706 unsigned long new_start;
1708 /* address space limit tests */
1709 if (!may_expand_vm(mm, grow))
1712 /* Stack limit test */
1713 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1716 /* mlock limit tests */
1717 if (vma->vm_flags & VM_LOCKED) {
1718 unsigned long locked;
1719 unsigned long limit;
1720 locked = mm->locked_vm + grow;
1721 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1722 limit >>= PAGE_SHIFT;
1723 if (locked > limit && !capable(CAP_IPC_LOCK))
1727 /* Check to ensure the stack will not grow into a hugetlb-only region */
1728 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1730 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1734 * Overcommit.. This must be the final test, as it will
1735 * update security statistics.
1737 if (security_vm_enough_memory_mm(mm, grow))
1740 /* Ok, everything looks good - let it rip */
1741 mm->total_vm += grow;
1742 if (vma->vm_flags & VM_LOCKED)
1743 mm->locked_vm += grow;
1744 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1748 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1750 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1751 * vma is the last one with address > vma->vm_end. Have to extend vma.
1753 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1757 if (!(vma->vm_flags & VM_GROWSUP))
1761 * We must make sure the anon_vma is allocated
1762 * so that the anon_vma locking is not a noop.
1764 if (unlikely(anon_vma_prepare(vma)))
1766 vma_lock_anon_vma(vma);
1769 * vma->vm_start/vm_end cannot change under us because the caller
1770 * is required to hold the mmap_sem in read mode. We need the
1771 * anon_vma lock to serialize against concurrent expand_stacks.
1772 * Also guard against wrapping around to address 0.
1774 if (address < PAGE_ALIGN(address+4))
1775 address = PAGE_ALIGN(address+4);
1777 vma_unlock_anon_vma(vma);
1782 /* Somebody else might have raced and expanded it already */
1783 if (address > vma->vm_end) {
1784 unsigned long size, grow;
1786 size = address - vma->vm_start;
1787 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1789 error = acct_stack_growth(vma, size, grow);
1791 vma->vm_end = address;
1792 perf_event_mmap(vma);
1795 vma_unlock_anon_vma(vma);
1796 khugepaged_enter_vma_merge(vma);
1799 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1802 * vma is the first one with address < vma->vm_start. Have to extend vma.
1804 static int expand_downwards(struct vm_area_struct *vma,
1805 unsigned long address)
1810 * We must make sure the anon_vma is allocated
1811 * so that the anon_vma locking is not a noop.
1813 if (unlikely(anon_vma_prepare(vma)))
1816 address &= PAGE_MASK;
1817 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1821 vma_lock_anon_vma(vma);
1824 * vma->vm_start/vm_end cannot change under us because the caller
1825 * is required to hold the mmap_sem in read mode. We need the
1826 * anon_vma lock to serialize against concurrent expand_stacks.
1829 /* Somebody else might have raced and expanded it already */
1830 if (address < vma->vm_start) {
1831 unsigned long size, grow;
1833 size = vma->vm_end - address;
1834 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1836 error = acct_stack_growth(vma, size, grow);
1838 vma->vm_start = address;
1839 vma->vm_pgoff -= grow;
1840 perf_event_mmap(vma);
1843 vma_unlock_anon_vma(vma);
1844 khugepaged_enter_vma_merge(vma);
1848 int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1850 return expand_downwards(vma, address);
1853 #ifdef CONFIG_STACK_GROWSUP
1854 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1856 return expand_upwards(vma, address);
1859 struct vm_area_struct *
1860 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1862 struct vm_area_struct *vma, *prev;
1865 vma = find_vma_prev(mm, addr, &prev);
1866 if (vma && (vma->vm_start <= addr))
1868 if (!prev || expand_stack(prev, addr))
1870 if (prev->vm_flags & VM_LOCKED) {
1871 mlock_vma_pages_range(prev, addr, prev->vm_end);
1876 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1878 return expand_downwards(vma, address);
1881 struct vm_area_struct *
1882 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1884 struct vm_area_struct * vma;
1885 unsigned long start;
1888 vma = find_vma(mm,addr);
1891 if (vma->vm_start <= addr)
1893 if (!(vma->vm_flags & VM_GROWSDOWN))
1895 start = vma->vm_start;
1896 if (expand_stack(vma, addr))
1898 if (vma->vm_flags & VM_LOCKED) {
1899 mlock_vma_pages_range(vma, addr, start);
1906 * Ok - we have the memory areas we should free on the vma list,
1907 * so release them, and do the vma updates.
1909 * Called with the mm semaphore held.
1911 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1913 /* Update high watermark before we lower total_vm */
1914 update_hiwater_vm(mm);
1916 long nrpages = vma_pages(vma);
1918 mm->total_vm -= nrpages;
1919 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1920 vma = remove_vma(vma);
1926 * Get rid of page table information in the indicated region.
1928 * Called with the mm semaphore held.
1930 static void unmap_region(struct mm_struct *mm,
1931 struct vm_area_struct *vma, struct vm_area_struct *prev,
1932 unsigned long start, unsigned long end)
1934 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1935 struct mmu_gather *tlb;
1936 unsigned long nr_accounted = 0;
1939 tlb = tlb_gather_mmu(mm, 0);
1940 update_hiwater_rss(mm);
1941 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1942 vm_unacct_memory(nr_accounted);
1943 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1944 next? next->vm_start: 0);
1945 tlb_finish_mmu(tlb, start, end);
1949 * Create a list of vma's touched by the unmap, removing them from the mm's
1950 * vma list as we go..
1953 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1954 struct vm_area_struct *prev, unsigned long end)
1956 struct vm_area_struct **insertion_point;
1957 struct vm_area_struct *tail_vma = NULL;
1960 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1961 vma->vm_prev = NULL;
1963 rb_erase(&vma->vm_rb, &mm->mm_rb);
1967 } while (vma && vma->vm_start < end);
1968 *insertion_point = vma;
1970 vma->vm_prev = prev;
1971 tail_vma->vm_next = NULL;
1972 if (mm->unmap_area == arch_unmap_area)
1973 addr = prev ? prev->vm_end : mm->mmap_base;
1975 addr = vma ? vma->vm_start : mm->mmap_base;
1976 mm->unmap_area(mm, addr);
1977 mm->mmap_cache = NULL; /* Kill the cache. */
1981 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1982 * munmap path where it doesn't make sense to fail.
1984 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1985 unsigned long addr, int new_below)
1987 struct mempolicy *pol;
1988 struct vm_area_struct *new;
1991 if (is_vm_hugetlb_page(vma) && (addr &
1992 ~(huge_page_mask(hstate_vma(vma)))))
1995 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1999 /* most fields are the same, copy all, and then fixup */
2002 INIT_LIST_HEAD(&new->anon_vma_chain);
2007 new->vm_start = addr;
2008 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2011 pol = mpol_dup(vma_policy(vma));
2016 vma_set_policy(new, pol);
2018 if (anon_vma_clone(new, vma))
2022 get_file(new->vm_file);
2023 if (vma->vm_flags & VM_EXECUTABLE)
2024 added_exe_file_vma(mm);
2027 if (new->vm_ops && new->vm_ops->open)
2028 new->vm_ops->open(new);
2031 unsigned long old_end = vma->vm_end;
2033 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2034 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2035 if (vma->vm_flags & VM_EXEC)
2036 arch_remove_exec_range(mm, old_end);
2038 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2044 /* Clean everything up if vma_adjust failed. */
2045 if (new->vm_ops && new->vm_ops->close)
2046 new->vm_ops->close(new);
2048 if (vma->vm_flags & VM_EXECUTABLE)
2049 removed_exe_file_vma(mm);
2052 unlink_anon_vmas(new);
2056 kmem_cache_free(vm_area_cachep, new);
2062 * Split a vma into two pieces at address 'addr', a new vma is allocated
2063 * either for the first part or the tail.
2065 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2066 unsigned long addr, int new_below)
2068 if (mm->map_count >= sysctl_max_map_count)
2071 return __split_vma(mm, vma, addr, new_below);
2074 /* Munmap is split into 2 main parts -- this part which finds
2075 * what needs doing, and the areas themselves, which do the
2076 * work. This now handles partial unmappings.
2077 * Jeremy Fitzhardinge <jeremy@goop.org>
2079 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2082 struct vm_area_struct *vma, *prev, *last;
2084 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2087 if ((len = PAGE_ALIGN(len)) == 0)
2090 /* Find the first overlapping VMA */
2091 vma = find_vma_prev(mm, start, &prev);
2094 /* we have start < vma->vm_end */
2096 /* if it doesn't overlap, we have nothing.. */
2098 if (vma->vm_start >= end)
2102 * If we need to split any vma, do it now to save pain later.
2104 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2105 * unmapped vm_area_struct will remain in use: so lower split_vma
2106 * places tmp vma above, and higher split_vma places tmp vma below.
2108 if (start > vma->vm_start) {
2112 * Make sure that map_count on return from munmap() will
2113 * not exceed its limit; but let map_count go just above
2114 * its limit temporarily, to help free resources as expected.
2116 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2119 error = __split_vma(mm, vma, start, 0);
2125 /* Does it split the last one? */
2126 last = find_vma(mm, end);
2127 if (last && end > last->vm_start) {
2128 int error = __split_vma(mm, last, end, 1);
2132 vma = prev? prev->vm_next: mm->mmap;
2135 * unlock any mlock()ed ranges before detaching vmas
2137 if (mm->locked_vm) {
2138 struct vm_area_struct *tmp = vma;
2139 while (tmp && tmp->vm_start < end) {
2140 if (tmp->vm_flags & VM_LOCKED) {
2141 mm->locked_vm -= vma_pages(tmp);
2142 munlock_vma_pages_all(tmp);
2149 * Remove the vma's, and unmap the actual pages
2151 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2152 unmap_region(mm, vma, prev, start, end);
2154 /* Fix up all other VM information */
2155 remove_vma_list(mm, vma);
2160 EXPORT_SYMBOL(do_munmap);
2162 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2165 struct mm_struct *mm = current->mm;
2167 profile_munmap(addr);
2169 down_write(&mm->mmap_sem);
2170 ret = do_munmap(mm, addr, len);
2171 up_write(&mm->mmap_sem);
2175 static inline void verify_mm_writelocked(struct mm_struct *mm)
2177 #ifdef CONFIG_DEBUG_VM
2178 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2180 up_read(&mm->mmap_sem);
2186 * this is really a simplified "do_mmap". it only handles
2187 * anonymous maps. eventually we may be able to do some
2188 * brk-specific accounting here.
2190 unsigned long do_brk(unsigned long addr, unsigned long len)
2192 struct mm_struct * mm = current->mm;
2193 struct vm_area_struct * vma, * prev;
2194 unsigned long flags;
2195 struct rb_node ** rb_link, * rb_parent;
2196 pgoff_t pgoff = addr >> PAGE_SHIFT;
2199 len = PAGE_ALIGN(len);
2203 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2207 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2209 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2210 if (error & ~PAGE_MASK)
2216 if (mm->def_flags & VM_LOCKED) {
2217 unsigned long locked, lock_limit;
2218 locked = len >> PAGE_SHIFT;
2219 locked += mm->locked_vm;
2220 lock_limit = rlimit(RLIMIT_MEMLOCK);
2221 lock_limit >>= PAGE_SHIFT;
2222 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2227 * mm->mmap_sem is required to protect against another thread
2228 * changing the mappings in case we sleep.
2230 verify_mm_writelocked(mm);
2233 * Clear old maps. this also does some error checking for us
2236 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2237 if (vma && vma->vm_start < addr + len) {
2238 if (do_munmap(mm, addr, len))
2243 /* Check against address space limits *after* clearing old maps... */
2244 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2247 if (mm->map_count > sysctl_max_map_count)
2250 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2253 /* Can we just expand an old private anonymous mapping? */
2254 vma = vma_merge(mm, prev, addr, addr + len, flags,
2255 NULL, NULL, pgoff, NULL);
2260 * create a vma struct for an anonymous mapping
2262 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2264 vm_unacct_memory(len >> PAGE_SHIFT);
2268 INIT_LIST_HEAD(&vma->anon_vma_chain);
2270 vma->vm_start = addr;
2271 vma->vm_end = addr + len;
2272 vma->vm_pgoff = pgoff;
2273 vma->vm_flags = flags;
2274 vma->vm_page_prot = vm_get_page_prot(flags);
2275 vma_link(mm, vma, prev, rb_link, rb_parent);
2277 perf_event_mmap(vma);
2278 mm->total_vm += len >> PAGE_SHIFT;
2279 if (flags & VM_LOCKED) {
2280 if (!mlock_vma_pages_range(vma, addr, addr + len))
2281 mm->locked_vm += (len >> PAGE_SHIFT);
2286 EXPORT_SYMBOL(do_brk);
2288 /* Release all mmaps. */
2289 void exit_mmap(struct mm_struct *mm)
2291 struct mmu_gather *tlb;
2292 struct vm_area_struct *vma;
2293 unsigned long nr_accounted = 0;
2296 /* mm's last user has gone, and its about to be pulled down */
2297 mmu_notifier_release(mm);
2299 if (mm->locked_vm) {
2302 if (vma->vm_flags & VM_LOCKED)
2303 munlock_vma_pages_all(vma);
2311 if (!vma) /* Can happen if dup_mmap() received an OOM */
2316 tlb = tlb_gather_mmu(mm, 1);
2317 /* update_hiwater_rss(mm) here? but nobody should be looking */
2318 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2319 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2320 vm_unacct_memory(nr_accounted);
2322 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
2323 tlb_finish_mmu(tlb, 0, end);
2324 arch_flush_exec_range(mm);
2327 * Walk the list again, actually closing and freeing it,
2328 * with preemption enabled, without holding any MM locks.
2331 vma = remove_vma(vma);
2333 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2336 /* Insert vm structure into process list sorted by address
2337 * and into the inode's i_mmap tree. If vm_file is non-NULL
2338 * then i_mmap_lock is taken here.
2340 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2342 struct vm_area_struct * __vma, * prev;
2343 struct rb_node ** rb_link, * rb_parent;
2346 * The vm_pgoff of a purely anonymous vma should be irrelevant
2347 * until its first write fault, when page's anon_vma and index
2348 * are set. But now set the vm_pgoff it will almost certainly
2349 * end up with (unless mremap moves it elsewhere before that
2350 * first wfault), so /proc/pid/maps tells a consistent story.
2352 * By setting it to reflect the virtual start address of the
2353 * vma, merges and splits can happen in a seamless way, just
2354 * using the existing file pgoff checks and manipulations.
2355 * Similarly in do_mmap_pgoff and in do_brk.
2357 if (!vma->vm_file) {
2358 BUG_ON(vma->anon_vma);
2359 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2361 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2362 if (__vma && __vma->vm_start < vma->vm_end)
2364 if ((vma->vm_flags & VM_ACCOUNT) &&
2365 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2367 vma_link(mm, vma, prev, rb_link, rb_parent);
2372 * Copy the vma structure to a new location in the same mm,
2373 * prior to moving page table entries, to effect an mremap move.
2375 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2376 unsigned long addr, unsigned long len, pgoff_t pgoff)
2378 struct vm_area_struct *vma = *vmap;
2379 unsigned long vma_start = vma->vm_start;
2380 struct mm_struct *mm = vma->vm_mm;
2381 struct vm_area_struct *new_vma, *prev;
2382 struct rb_node **rb_link, *rb_parent;
2383 struct mempolicy *pol;
2386 * If anonymous vma has not yet been faulted, update new pgoff
2387 * to match new location, to increase its chance of merging.
2389 if (!vma->vm_file && !vma->anon_vma)
2390 pgoff = addr >> PAGE_SHIFT;
2392 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2393 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2394 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2397 * Source vma may have been merged into new_vma
2399 if (vma_start >= new_vma->vm_start &&
2400 vma_start < new_vma->vm_end)
2403 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2406 pol = mpol_dup(vma_policy(vma));
2409 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2410 if (anon_vma_clone(new_vma, vma))
2411 goto out_free_mempol;
2412 vma_set_policy(new_vma, pol);
2413 new_vma->vm_start = addr;
2414 new_vma->vm_end = addr + len;
2415 new_vma->vm_pgoff = pgoff;
2416 if (new_vma->vm_file) {
2417 get_file(new_vma->vm_file);
2418 if (vma->vm_flags & VM_EXECUTABLE)
2419 added_exe_file_vma(mm);
2421 if (new_vma->vm_ops && new_vma->vm_ops->open)
2422 new_vma->vm_ops->open(new_vma);
2423 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2431 kmem_cache_free(vm_area_cachep, new_vma);
2436 * Return true if the calling process may expand its vm space by the passed
2439 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2441 unsigned long cur = mm->total_vm; /* pages */
2444 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2446 if (cur + npages > lim)
2452 static int special_mapping_fault(struct vm_area_struct *vma,
2453 struct vm_fault *vmf)
2456 struct page **pages;
2459 * special mappings have no vm_file, and in that case, the mm
2460 * uses vm_pgoff internally. So we have to subtract it from here.
2461 * We are allowed to do this because we are the mm; do not copy
2462 * this code into drivers!
2464 pgoff = vmf->pgoff - vma->vm_pgoff;
2466 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2470 struct page *page = *pages;
2476 return VM_FAULT_SIGBUS;
2480 * Having a close hook prevents vma merging regardless of flags.
2482 static void special_mapping_close(struct vm_area_struct *vma)
2486 static const struct vm_operations_struct special_mapping_vmops = {
2487 .close = special_mapping_close,
2488 .fault = special_mapping_fault,
2492 * Called with mm->mmap_sem held for writing.
2493 * Insert a new vma covering the given region, with the given flags.
2494 * Its pages are supplied by the given array of struct page *.
2495 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2496 * The region past the last page supplied will always produce SIGBUS.
2497 * The array pointer and the pages it points to are assumed to stay alive
2498 * for as long as this mapping might exist.
2500 int install_special_mapping(struct mm_struct *mm,
2501 unsigned long addr, unsigned long len,
2502 unsigned long vm_flags, struct page **pages)
2505 struct vm_area_struct *vma;
2507 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2508 if (unlikely(vma == NULL))
2511 INIT_LIST_HEAD(&vma->anon_vma_chain);
2513 vma->vm_start = addr;
2514 vma->vm_end = addr + len;
2516 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2517 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2519 vma->vm_ops = &special_mapping_vmops;
2520 vma->vm_private_data = pages;
2522 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2526 ret = insert_vm_struct(mm, vma);
2530 mm->total_vm += len >> PAGE_SHIFT;
2532 perf_event_mmap(vma);
2537 kmem_cache_free(vm_area_cachep, vma);
2541 static DEFINE_MUTEX(mm_all_locks_mutex);
2543 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2545 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2547 * The LSB of head.next can't change from under us
2548 * because we hold the mm_all_locks_mutex.
2550 spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem);
2552 * We can safely modify head.next after taking the
2553 * anon_vma->root->lock. If some other vma in this mm shares
2554 * the same anon_vma we won't take it again.
2556 * No need of atomic instructions here, head.next
2557 * can't change from under us thanks to the
2558 * anon_vma->root->lock.
2560 if (__test_and_set_bit(0, (unsigned long *)
2561 &anon_vma->root->head.next))
2566 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2568 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2570 * AS_MM_ALL_LOCKS can't change from under us because
2571 * we hold the mm_all_locks_mutex.
2573 * Operations on ->flags have to be atomic because
2574 * even if AS_MM_ALL_LOCKS is stable thanks to the
2575 * mm_all_locks_mutex, there may be other cpus
2576 * changing other bitflags in parallel to us.
2578 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2580 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
2585 * This operation locks against the VM for all pte/vma/mm related
2586 * operations that could ever happen on a certain mm. This includes
2587 * vmtruncate, try_to_unmap, and all page faults.
2589 * The caller must take the mmap_sem in write mode before calling
2590 * mm_take_all_locks(). The caller isn't allowed to release the
2591 * mmap_sem until mm_drop_all_locks() returns.
2593 * mmap_sem in write mode is required in order to block all operations
2594 * that could modify pagetables and free pages without need of
2595 * altering the vma layout (for example populate_range() with
2596 * nonlinear vmas). It's also needed in write mode to avoid new
2597 * anon_vmas to be associated with existing vmas.
2599 * A single task can't take more than one mm_take_all_locks() in a row
2600 * or it would deadlock.
2602 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2603 * mapping->flags avoid to take the same lock twice, if more than one
2604 * vma in this mm is backed by the same anon_vma or address_space.
2606 * We can take all the locks in random order because the VM code
2607 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2608 * takes more than one of them in a row. Secondly we're protected
2609 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2611 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2612 * that may have to take thousand of locks.
2614 * mm_take_all_locks() can fail if it's interrupted by signals.
2616 int mm_take_all_locks(struct mm_struct *mm)
2618 struct vm_area_struct *vma;
2619 struct anon_vma_chain *avc;
2622 BUG_ON(down_read_trylock(&mm->mmap_sem));
2624 mutex_lock(&mm_all_locks_mutex);
2626 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2627 if (signal_pending(current))
2629 if (vma->vm_file && vma->vm_file->f_mapping)
2630 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2633 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2634 if (signal_pending(current))
2637 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2638 vm_lock_anon_vma(mm, avc->anon_vma);
2645 mm_drop_all_locks(mm);
2650 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2652 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2654 * The LSB of head.next can't change to 0 from under
2655 * us because we hold the mm_all_locks_mutex.
2657 * We must however clear the bitflag before unlocking
2658 * the vma so the users using the anon_vma->head will
2659 * never see our bitflag.
2661 * No need of atomic instructions here, head.next
2662 * can't change from under us until we release the
2663 * anon_vma->root->lock.
2665 if (!__test_and_clear_bit(0, (unsigned long *)
2666 &anon_vma->root->head.next))
2668 anon_vma_unlock(anon_vma);
2672 static void vm_unlock_mapping(struct address_space *mapping)
2674 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2676 * AS_MM_ALL_LOCKS can't change to 0 from under us
2677 * because we hold the mm_all_locks_mutex.
2679 spin_unlock(&mapping->i_mmap_lock);
2680 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2687 * The mmap_sem cannot be released by the caller until
2688 * mm_drop_all_locks() returns.
2690 void mm_drop_all_locks(struct mm_struct *mm)
2692 struct vm_area_struct *vma;
2693 struct anon_vma_chain *avc;
2695 BUG_ON(down_read_trylock(&mm->mmap_sem));
2696 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2698 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2700 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2701 vm_unlock_anon_vma(avc->anon_vma);
2702 if (vma->vm_file && vma->vm_file->f_mapping)
2703 vm_unlock_mapping(vma->vm_file->f_mapping);
2706 mutex_unlock(&mm_all_locks_mutex);
2710 * initialise the VMA slab
2712 void __init mmap_init(void)
2716 ret = percpu_counter_init(&vm_committed_as, 0);