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;
90 int sysctl_mmap_noexec_taint __read_mostly = CONFIG_MMAP_NOEXEC_TAINT;
92 * Make sure vm_committed_as in one cacheline and not cacheline shared with
93 * other variables. It can be updated by several CPUs frequently.
95 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
98 * Check that a process has enough memory to allocate a new virtual
99 * mapping. 0 means there is enough memory for the allocation to
100 * succeed and -ENOMEM implies there is not.
102 * We currently support three overcommit policies, which are set via the
103 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
105 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
106 * Additional code 2002 Jul 20 by Robert Love.
108 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
110 * Note this is a helper function intended to be used by LSMs which
111 * wish to use this logic.
113 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
115 unsigned long free, allowed;
117 vm_acct_memory(pages);
120 * Sometimes we want to use more memory than we have
122 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
125 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
126 free = global_page_state(NR_FREE_PAGES);
127 free += global_page_state(NR_FILE_PAGES);
130 * shmem pages shouldn't be counted as free in this
131 * case, they can't be purged, only swapped out, and
132 * that won't affect the overall amount of available
133 * memory in the system.
135 free -= global_page_state(NR_SHMEM);
137 free += nr_swap_pages;
140 * Any slabs which are created with the
141 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
142 * which are reclaimable, under pressure. The dentry
143 * cache and most inode caches should fall into this
145 free += global_page_state(NR_SLAB_RECLAIMABLE);
148 * Leave reserved pages. The pages are not for anonymous pages.
150 if (free <= totalreserve_pages)
153 free -= totalreserve_pages;
156 * Leave the last 3% for root
167 allowed = (totalram_pages - hugetlb_total_pages())
168 * sysctl_overcommit_ratio / 100;
170 * Leave the last 3% for root
173 allowed -= allowed / 32;
174 allowed += total_swap_pages;
176 /* Don't let a single process grow too big:
177 leave 3% of the size of this process for other processes */
179 allowed -= mm->total_vm / 32;
181 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
184 vm_unacct_memory(pages);
190 * Requires inode->i_mapping->i_mmap_mutex
192 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
193 struct file *file, struct address_space *mapping)
195 if (vma->vm_flags & VM_DENYWRITE)
196 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
197 if (vma->vm_flags & VM_SHARED)
198 mapping->i_mmap_writable--;
200 flush_dcache_mmap_lock(mapping);
201 if (unlikely(vma->vm_flags & VM_NONLINEAR))
202 list_del_init(&vma->shared.vm_set.list);
204 vma_prio_tree_remove(vma, &mapping->i_mmap);
205 flush_dcache_mmap_unlock(mapping);
209 * Unlink a file-based vm structure from its prio_tree, to hide
210 * vma from rmap and vmtruncate before freeing its page tables.
212 void unlink_file_vma(struct vm_area_struct *vma)
214 struct file *file = vma->vm_file;
217 struct address_space *mapping = file->f_mapping;
218 mutex_lock(&mapping->i_mmap_mutex);
219 __remove_shared_vm_struct(vma, file, mapping);
220 mutex_unlock(&mapping->i_mmap_mutex);
225 * Close a vm structure and free it, returning the next.
227 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
229 struct vm_area_struct *next = vma->vm_next;
232 if (vma->vm_ops && vma->vm_ops->close)
233 vma->vm_ops->close(vma);
236 if (vma->vm_flags & VM_EXECUTABLE)
237 removed_exe_file_vma(vma->vm_mm);
239 mpol_put(vma_policy(vma));
240 kmem_cache_free(vm_area_cachep, vma);
244 static unsigned long do_brk(unsigned long addr, unsigned long len);
246 SYSCALL_DEFINE1(brk, unsigned long, brk)
248 unsigned long rlim, retval;
249 unsigned long newbrk, oldbrk;
250 struct mm_struct *mm = current->mm;
251 unsigned long min_brk;
253 down_write(&mm->mmap_sem);
255 #ifdef CONFIG_COMPAT_BRK
257 * CONFIG_COMPAT_BRK can still be overridden by setting
258 * randomize_va_space to 2, which will still cause mm->start_brk
259 * to be arbitrarily shifted
261 if (current->brk_randomized)
262 min_brk = mm->start_brk;
264 min_brk = mm->end_data;
266 min_brk = mm->start_brk;
272 * Check against rlimit here. If this check is done later after the test
273 * of oldbrk with newbrk then it can escape the test and let the data
274 * segment grow beyond its set limit the in case where the limit is
275 * not page aligned -Ram Gupta
277 rlim = rlimit(RLIMIT_DATA);
278 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
279 (mm->end_data - mm->start_data) > rlim)
282 newbrk = PAGE_ALIGN(brk);
283 oldbrk = PAGE_ALIGN(mm->brk);
284 if (oldbrk == newbrk)
287 /* Always allow shrinking brk. */
288 if (brk <= mm->brk) {
289 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
294 /* Check against existing mmap mappings. */
295 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
298 /* Ok, looks good - let it rip. */
299 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
305 up_write(&mm->mmap_sem);
310 static int browse_rb(struct rb_root *root)
313 struct rb_node *nd, *pn = NULL;
314 unsigned long prev = 0, pend = 0;
316 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
317 struct vm_area_struct *vma;
318 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
319 if (vma->vm_start < prev)
320 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
321 if (vma->vm_start < pend)
322 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
323 if (vma->vm_start > vma->vm_end)
324 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
327 prev = vma->vm_start;
331 for (nd = pn; nd; nd = rb_prev(nd)) {
335 printk("backwards %d, forwards %d\n", j, i), i = 0;
339 void validate_mm(struct mm_struct *mm)
343 struct vm_area_struct *tmp = mm->mmap;
348 if (i != mm->map_count)
349 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
350 i = browse_rb(&mm->mm_rb);
351 if (i != mm->map_count)
352 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
356 #define validate_mm(mm) do { } while (0)
359 static struct vm_area_struct *
360 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
361 struct vm_area_struct **pprev, struct rb_node ***rb_link,
362 struct rb_node ** rb_parent)
364 struct vm_area_struct * vma;
365 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
367 __rb_link = &mm->mm_rb.rb_node;
368 rb_prev = __rb_parent = NULL;
372 struct vm_area_struct *vma_tmp;
374 __rb_parent = *__rb_link;
375 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
377 if (vma_tmp->vm_end > addr) {
379 if (vma_tmp->vm_start <= addr)
381 __rb_link = &__rb_parent->rb_left;
383 rb_prev = __rb_parent;
384 __rb_link = &__rb_parent->rb_right;
390 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
391 *rb_link = __rb_link;
392 *rb_parent = __rb_parent;
396 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
397 struct rb_node **rb_link, struct rb_node *rb_parent)
399 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
400 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
403 static void __vma_link_file(struct vm_area_struct *vma)
409 struct address_space *mapping = file->f_mapping;
411 if (vma->vm_flags & VM_DENYWRITE)
412 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
413 if (vma->vm_flags & VM_SHARED)
414 mapping->i_mmap_writable++;
416 flush_dcache_mmap_lock(mapping);
417 if (unlikely(vma->vm_flags & VM_NONLINEAR))
418 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
420 vma_prio_tree_insert(vma, &mapping->i_mmap);
421 flush_dcache_mmap_unlock(mapping);
426 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
427 struct vm_area_struct *prev, struct rb_node **rb_link,
428 struct rb_node *rb_parent)
430 __vma_link_list(mm, vma, prev, rb_parent);
431 __vma_link_rb(mm, vma, rb_link, rb_parent);
434 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
435 struct vm_area_struct *prev, struct rb_node **rb_link,
436 struct rb_node *rb_parent)
438 struct address_space *mapping = NULL;
441 mapping = vma->vm_file->f_mapping;
444 mutex_lock(&mapping->i_mmap_mutex);
446 __vma_link(mm, vma, prev, rb_link, rb_parent);
447 __vma_link_file(vma);
450 mutex_unlock(&mapping->i_mmap_mutex);
457 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
458 * mm's list and rbtree. It has already been inserted into the prio_tree.
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) {
542 if (anon_vma_clone(importer, exporter))
544 importer->anon_vma = exporter->anon_vma;
549 mapping = file->f_mapping;
550 if (!(vma->vm_flags & VM_NONLINEAR))
551 root = &mapping->i_mmap;
552 mutex_lock(&mapping->i_mmap_mutex);
555 * Put into prio_tree now, so instantiated pages
556 * are visible to arm/parisc __flush_dcache_page
557 * throughout; but we cannot insert into address
558 * space until vma start or end is updated.
560 __vma_link_file(insert);
564 vma_adjust_trans_huge(vma, start, end, adjust_next);
567 * When changing only vma->vm_end, we don't really need anon_vma
568 * lock. This is a fairly rare case by itself, but the anon_vma
569 * lock may be shared between many sibling processes. Skipping
570 * the lock for brk adjustments makes a difference sometimes.
572 if (vma->anon_vma && (importer || start != vma->vm_start)) {
573 anon_vma = vma->anon_vma;
574 anon_vma_lock(anon_vma);
578 flush_dcache_mmap_lock(mapping);
579 vma_prio_tree_remove(vma, root);
581 vma_prio_tree_remove(next, root);
584 vma->vm_start = start;
586 vma->vm_pgoff = pgoff;
588 next->vm_start += adjust_next << PAGE_SHIFT;
589 next->vm_pgoff += adjust_next;
594 vma_prio_tree_insert(next, root);
595 vma_prio_tree_insert(vma, root);
596 flush_dcache_mmap_unlock(mapping);
601 * vma_merge has merged next into vma, and needs
602 * us to remove next before dropping the locks.
604 __vma_unlink(mm, next, vma);
606 __remove_shared_vm_struct(next, file, mapping);
609 * split_vma has split insert from vma, and needs
610 * us to insert it before dropping the locks
611 * (it may either follow vma or precede it).
613 __insert_vm_struct(mm, insert);
617 anon_vma_unlock(anon_vma);
619 mutex_unlock(&mapping->i_mmap_mutex);
624 if (next->vm_flags & VM_EXECUTABLE)
625 removed_exe_file_vma(mm);
628 anon_vma_merge(vma, next);
630 mpol_put(vma_policy(next));
631 kmem_cache_free(vm_area_cachep, next);
633 * In mprotect's case 6 (see comments on vma_merge),
634 * we must remove another next too. It would clutter
635 * up the code too much to do both in one go.
637 if (remove_next == 2) {
649 * If the vma has a ->close operation then the driver probably needs to release
650 * per-vma resources, so we don't attempt to merge those.
652 static inline int is_mergeable_vma(struct vm_area_struct *vma,
653 struct file *file, unsigned long vm_flags)
655 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
656 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
658 if (vma->vm_file != file)
660 if (vma->vm_ops && vma->vm_ops->close)
665 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
666 struct anon_vma *anon_vma2,
667 struct vm_area_struct *vma)
670 * The list_is_singular() test is to avoid merging VMA cloned from
671 * parents. This can improve scalability caused by anon_vma lock.
673 if ((!anon_vma1 || !anon_vma2) && (!vma ||
674 list_is_singular(&vma->anon_vma_chain)))
676 return anon_vma1 == anon_vma2;
680 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
681 * in front of (at a lower virtual address and file offset than) the vma.
683 * We cannot merge two vmas if they have differently assigned (non-NULL)
684 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
686 * We don't check here for the merged mmap wrapping around the end of pagecache
687 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
688 * wrap, nor mmaps which cover the final page at index -1UL.
691 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
692 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
694 if (is_mergeable_vma(vma, file, vm_flags) &&
695 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
696 if (vma->vm_pgoff == vm_pgoff)
703 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
704 * beyond (at a higher virtual address and file offset than) the vma.
706 * We cannot merge two vmas if they have differently assigned (non-NULL)
707 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
710 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
711 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
713 if (is_mergeable_vma(vma, file, vm_flags) &&
714 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
716 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
717 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
724 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
725 * whether that can be merged with its predecessor or its successor.
726 * Or both (it neatly fills a hole).
728 * In most cases - when called for mmap, brk or mremap - [addr,end) is
729 * certain not to be mapped by the time vma_merge is called; but when
730 * called for mprotect, it is certain to be already mapped (either at
731 * an offset within prev, or at the start of next), and the flags of
732 * this area are about to be changed to vm_flags - and the no-change
733 * case has already been eliminated.
735 * The following mprotect cases have to be considered, where AAAA is
736 * the area passed down from mprotect_fixup, never extending beyond one
737 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
739 * AAAA AAAA AAAA AAAA
740 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
741 * cannot merge might become might become might become
742 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
743 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
744 * mremap move: PPPPNNNNNNNN 8
746 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
747 * might become case 1 below case 2 below case 3 below
749 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
750 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
752 struct vm_area_struct *vma_merge(struct mm_struct *mm,
753 struct vm_area_struct *prev, unsigned long addr,
754 unsigned long end, unsigned long vm_flags,
755 struct anon_vma *anon_vma, struct file *file,
756 pgoff_t pgoff, struct mempolicy *policy)
758 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
759 struct vm_area_struct *area, *next;
763 * We later require that vma->vm_flags == vm_flags,
764 * so this tests vma->vm_flags & VM_SPECIAL, too.
766 if (vm_flags & VM_SPECIAL)
770 next = prev->vm_next;
774 if (next && next->vm_end == end) /* cases 6, 7, 8 */
775 next = next->vm_next;
778 * Can it merge with the predecessor?
780 if (prev && prev->vm_end == addr &&
781 mpol_equal(vma_policy(prev), policy) &&
782 can_vma_merge_after(prev, vm_flags,
783 anon_vma, file, pgoff)) {
785 * OK, it can. Can we now merge in the successor as well?
787 if (next && end == next->vm_start &&
788 mpol_equal(policy, vma_policy(next)) &&
789 can_vma_merge_before(next, vm_flags,
790 anon_vma, file, pgoff+pglen) &&
791 is_mergeable_anon_vma(prev->anon_vma,
792 next->anon_vma, NULL)) {
794 err = vma_adjust(prev, prev->vm_start,
795 next->vm_end, prev->vm_pgoff, NULL);
796 } else /* cases 2, 5, 7 */
797 err = vma_adjust(prev, prev->vm_start,
798 end, prev->vm_pgoff, NULL);
801 khugepaged_enter_vma_merge(prev);
806 * Can this new request be merged in front of next?
808 if (next && end == next->vm_start &&
809 mpol_equal(policy, vma_policy(next)) &&
810 can_vma_merge_before(next, vm_flags,
811 anon_vma, file, pgoff+pglen)) {
812 if (prev && addr < prev->vm_end) /* case 4 */
813 err = vma_adjust(prev, prev->vm_start,
814 addr, prev->vm_pgoff, NULL);
815 else /* cases 3, 8 */
816 err = vma_adjust(area, addr, next->vm_end,
817 next->vm_pgoff - pglen, NULL);
820 khugepaged_enter_vma_merge(area);
828 * Rough compatbility check to quickly see if it's even worth looking
829 * at sharing an anon_vma.
831 * They need to have the same vm_file, and the flags can only differ
832 * in things that mprotect may change.
834 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
835 * we can merge the two vma's. For example, we refuse to merge a vma if
836 * there is a vm_ops->close() function, because that indicates that the
837 * driver is doing some kind of reference counting. But that doesn't
838 * really matter for the anon_vma sharing case.
840 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
842 return a->vm_end == b->vm_start &&
843 mpol_equal(vma_policy(a), vma_policy(b)) &&
844 a->vm_file == b->vm_file &&
845 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
846 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
850 * Do some basic sanity checking to see if we can re-use the anon_vma
851 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
852 * the same as 'old', the other will be the new one that is trying
853 * to share the anon_vma.
855 * NOTE! This runs with mm_sem held for reading, so it is possible that
856 * the anon_vma of 'old' is concurrently in the process of being set up
857 * by another page fault trying to merge _that_. But that's ok: if it
858 * is being set up, that automatically means that it will be a singleton
859 * acceptable for merging, so we can do all of this optimistically. But
860 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
862 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
863 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
864 * is to return an anon_vma that is "complex" due to having gone through
867 * We also make sure that the two vma's are compatible (adjacent,
868 * and with the same memory policies). That's all stable, even with just
869 * a read lock on the mm_sem.
871 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
873 if (anon_vma_compatible(a, b)) {
874 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
876 if (anon_vma && list_is_singular(&old->anon_vma_chain))
883 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
884 * neighbouring vmas for a suitable anon_vma, before it goes off
885 * to allocate a new anon_vma. It checks because a repetitive
886 * sequence of mprotects and faults may otherwise lead to distinct
887 * anon_vmas being allocated, preventing vma merge in subsequent
890 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
892 struct anon_vma *anon_vma;
893 struct vm_area_struct *near;
899 anon_vma = reusable_anon_vma(near, vma, near);
907 anon_vma = reusable_anon_vma(near, near, vma);
912 * There's no absolute need to look only at touching neighbours:
913 * we could search further afield for "compatible" anon_vmas.
914 * But it would probably just be a waste of time searching,
915 * or lead to too many vmas hanging off the same anon_vma.
916 * We're trying to allow mprotect remerging later on,
917 * not trying to minimize memory used for anon_vmas.
922 #ifdef CONFIG_PROC_FS
923 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
924 struct file *file, long pages)
926 const unsigned long stack_flags
927 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
930 mm->shared_vm += pages;
931 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
932 mm->exec_vm += pages;
933 } else if (flags & stack_flags)
934 mm->stack_vm += pages;
935 if (flags & (VM_RESERVED|VM_IO))
936 mm->reserved_vm += pages;
938 #endif /* CONFIG_PROC_FS */
941 * If a hint addr is less than mmap_min_addr change hint to be as
942 * low as possible but still greater than mmap_min_addr
944 static inline unsigned long round_hint_to_min(unsigned long hint)
947 if (((void *)hint != NULL) &&
948 (hint < mmap_min_addr))
949 return PAGE_ALIGN(mmap_min_addr);
954 * The caller must hold down_write(¤t->mm->mmap_sem).
957 static unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
958 unsigned long len, unsigned long prot,
959 unsigned long flags, unsigned long pgoff)
961 struct mm_struct * mm = current->mm;
965 unsigned long reqprot = prot;
968 * Does the application expect PROT_READ to imply PROT_EXEC?
970 * (the exception is when the underlying filesystem is noexec
971 * mounted, in which case we dont add PROT_EXEC.)
973 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
974 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
980 if (!(flags & MAP_FIXED))
981 addr = round_hint_to_min(addr);
983 /* Careful about overflows.. */
984 len = PAGE_ALIGN(len);
988 /* offset overflow? */
989 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
992 /* Too many mappings? */
993 if (mm->map_count > sysctl_max_map_count)
996 /* Obtain the address to map to. we verify (or select) it and ensure
997 * that it represents a valid section of the address space.
999 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1000 if (addr & ~PAGE_MASK)
1003 /* Do simple checking here so the lower-level routines won't have
1004 * to. we assume access permissions have been handled by the open
1005 * of the memory object, so we don't do any here.
1007 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1008 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1010 if (flags & MAP_LOCKED)
1011 if (!can_do_mlock())
1014 /* mlock MCL_FUTURE? */
1015 if (vm_flags & VM_LOCKED) {
1016 unsigned long locked, lock_limit;
1017 locked = len >> PAGE_SHIFT;
1018 locked += mm->locked_vm;
1019 lock_limit = rlimit(RLIMIT_MEMLOCK);
1020 lock_limit >>= PAGE_SHIFT;
1021 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1025 inode = file ? file->f_path.dentry->d_inode : NULL;
1028 switch (flags & MAP_TYPE) {
1030 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1034 * Make sure we don't allow writing to an append-only
1037 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1041 * Make sure there are no mandatory locks on the file.
1043 if (locks_verify_locked(inode))
1046 vm_flags |= VM_SHARED | VM_MAYSHARE;
1047 if (!(file->f_mode & FMODE_WRITE))
1048 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1052 if (!(file->f_mode & FMODE_READ))
1054 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1055 if (vm_flags & VM_EXEC)
1057 if (sysctl_mmap_noexec_taint)
1058 vm_flags &= ~VM_MAYEXEC;
1061 if (!file->f_op || !file->f_op->mmap)
1069 switch (flags & MAP_TYPE) {
1075 vm_flags |= VM_SHARED | VM_MAYSHARE;
1079 * Set pgoff according to addr for anon_vma.
1081 pgoff = addr >> PAGE_SHIFT;
1088 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1092 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1095 unsigned long do_mmap(struct file *file, unsigned long addr,
1096 unsigned long len, unsigned long prot,
1097 unsigned long flag, unsigned long offset)
1099 if (unlikely(offset + PAGE_ALIGN(len) < offset))
1101 if (unlikely(offset & ~PAGE_MASK))
1103 return do_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
1105 EXPORT_SYMBOL(do_mmap);
1107 unsigned long vm_mmap(struct file *file, unsigned long addr,
1108 unsigned long len, unsigned long prot,
1109 unsigned long flag, unsigned long offset)
1112 struct mm_struct *mm = current->mm;
1114 down_write(&mm->mmap_sem);
1115 ret = do_mmap(file, addr, len, prot, flag, offset);
1116 up_write(&mm->mmap_sem);
1119 EXPORT_SYMBOL(vm_mmap);
1121 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1122 unsigned long, prot, unsigned long, flags,
1123 unsigned long, fd, unsigned long, pgoff)
1125 struct file *file = NULL;
1126 unsigned long retval = -EBADF;
1128 if (!(flags & MAP_ANONYMOUS)) {
1129 audit_mmap_fd(fd, flags);
1130 if (unlikely(flags & MAP_HUGETLB))
1135 } else if (flags & MAP_HUGETLB) {
1136 struct user_struct *user = NULL;
1138 * VM_NORESERVE is used because the reservations will be
1139 * taken when vm_ops->mmap() is called
1140 * A dummy user value is used because we are not locking
1141 * memory so no accounting is necessary
1143 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1144 VM_NORESERVE, &user,
1145 HUGETLB_ANONHUGE_INODE);
1147 return PTR_ERR(file);
1150 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1152 down_write(¤t->mm->mmap_sem);
1153 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1154 up_write(¤t->mm->mmap_sem);
1162 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1163 struct mmap_arg_struct {
1167 unsigned long flags;
1169 unsigned long offset;
1172 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1174 struct mmap_arg_struct a;
1176 if (copy_from_user(&a, arg, sizeof(a)))
1178 if (a.offset & ~PAGE_MASK)
1181 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1182 a.offset >> PAGE_SHIFT);
1184 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1187 * Some shared mappigns will want the pages marked read-only
1188 * to track write events. If so, we'll downgrade vm_page_prot
1189 * to the private version (using protection_map[] without the
1192 int vma_wants_writenotify(struct vm_area_struct *vma)
1194 vm_flags_t vm_flags = vma->vm_flags;
1196 /* If it was private or non-writable, the write bit is already clear */
1197 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1200 /* The backer wishes to know when pages are first written to? */
1201 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1204 /* The open routine did something to the protections already? */
1205 if (pgprot_val(vma->vm_page_prot) !=
1206 pgprot_val(vm_get_page_prot(vm_flags)))
1209 /* Specialty mapping? */
1210 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1213 /* Can the mapping track the dirty pages? */
1214 return vma->vm_file && vma->vm_file->f_mapping &&
1215 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1219 * We account for memory if it's a private writeable mapping,
1220 * not hugepages and VM_NORESERVE wasn't set.
1222 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1225 * hugetlb has its own accounting separate from the core VM
1226 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1228 if (file && is_file_hugepages(file))
1231 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1234 unsigned long mmap_region(struct file *file, unsigned long addr,
1235 unsigned long len, unsigned long flags,
1236 vm_flags_t vm_flags, unsigned long pgoff)
1238 struct mm_struct *mm = current->mm;
1239 struct vm_area_struct *vma, *prev;
1240 int correct_wcount = 0;
1242 struct rb_node **rb_link, *rb_parent;
1243 unsigned long charged = 0;
1244 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1246 /* Clear old maps */
1249 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1250 if (vma && vma->vm_start < addr + len) {
1251 if (do_munmap(mm, addr, len))
1256 /* Check against address space limit. */
1257 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1261 * Set 'VM_NORESERVE' if we should not account for the
1262 * memory use of this mapping.
1264 if ((flags & MAP_NORESERVE)) {
1265 /* We honor MAP_NORESERVE if allowed to overcommit */
1266 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1267 vm_flags |= VM_NORESERVE;
1269 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1270 if (file && is_file_hugepages(file))
1271 vm_flags |= VM_NORESERVE;
1275 * Private writable mapping: check memory availability
1277 if (accountable_mapping(file, vm_flags)) {
1278 charged = len >> PAGE_SHIFT;
1279 if (security_vm_enough_memory_mm(mm, charged))
1281 vm_flags |= VM_ACCOUNT;
1285 * Can we just expand an old mapping?
1287 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1292 * Determine the object being mapped and call the appropriate
1293 * specific mapper. the address has already been validated, but
1294 * not unmapped, but the maps are removed from the list.
1296 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1303 vma->vm_start = addr;
1304 vma->vm_end = addr + len;
1305 vma->vm_flags = vm_flags;
1306 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1307 vma->vm_pgoff = pgoff;
1308 INIT_LIST_HEAD(&vma->anon_vma_chain);
1310 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1313 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1315 if (vm_flags & VM_DENYWRITE) {
1316 error = deny_write_access(file);
1321 vma->vm_file = file;
1323 error = file->f_op->mmap(file, vma);
1325 goto unmap_and_free_vma;
1326 if (vm_flags & VM_EXECUTABLE)
1327 added_exe_file_vma(mm);
1329 /* Can addr have changed??
1331 * Answer: Yes, several device drivers can do it in their
1332 * f_op->mmap method. -DaveM
1334 addr = vma->vm_start;
1335 pgoff = vma->vm_pgoff;
1336 vm_flags = vma->vm_flags;
1337 } else if (vm_flags & VM_SHARED) {
1338 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1340 error = shmem_zero_setup(vma);
1345 if (vma_wants_writenotify(vma)) {
1346 pgprot_t pprot = vma->vm_page_prot;
1348 /* Can vma->vm_page_prot have changed??
1350 * Answer: Yes, drivers may have changed it in their
1351 * f_op->mmap method.
1353 * Ensures that vmas marked as uncached stay that way.
1355 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1356 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1357 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1360 vma_link(mm, vma, prev, rb_link, rb_parent);
1361 file = vma->vm_file;
1363 /* Once vma denies write, undo our temporary denial count */
1365 atomic_inc(&inode->i_writecount);
1367 perf_event_mmap(vma);
1369 mm->total_vm += len >> PAGE_SHIFT;
1370 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1371 if (vm_flags & VM_LOCKED) {
1372 if (!mlock_vma_pages_range(vma, addr, addr + len))
1373 mm->locked_vm += (len >> PAGE_SHIFT);
1374 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1375 make_pages_present(addr, addr + len);
1380 atomic_inc(&inode->i_writecount);
1381 vma->vm_file = NULL;
1384 /* Undo any partial mapping done by a device driver. */
1385 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1388 kmem_cache_free(vm_area_cachep, vma);
1391 vm_unacct_memory(charged);
1395 /* Get an address range which is currently unmapped.
1396 * For shmat() with addr=0.
1398 * Ugly calling convention alert:
1399 * Return value with the low bits set means error value,
1401 * if (ret & ~PAGE_MASK)
1404 * This function "knows" that -ENOMEM has the bits set.
1406 #ifndef HAVE_ARCH_UNMAPPED_AREA
1408 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1409 unsigned long len, unsigned long pgoff, unsigned long flags)
1411 struct mm_struct *mm = current->mm;
1412 struct vm_area_struct *vma;
1413 unsigned long start_addr;
1415 if (len > TASK_SIZE)
1418 if (flags & MAP_FIXED)
1422 addr = PAGE_ALIGN(addr);
1423 vma = find_vma(mm, addr);
1424 if (TASK_SIZE - len >= addr &&
1425 (!vma || addr + len <= vma->vm_start))
1428 if (len > mm->cached_hole_size) {
1429 start_addr = addr = mm->free_area_cache;
1431 start_addr = addr = TASK_UNMAPPED_BASE;
1432 mm->cached_hole_size = 0;
1436 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1437 /* At this point: (!vma || addr < vma->vm_end). */
1438 if (TASK_SIZE - len < addr) {
1440 * Start a new search - just in case we missed
1443 if (start_addr != TASK_UNMAPPED_BASE) {
1444 addr = TASK_UNMAPPED_BASE;
1446 mm->cached_hole_size = 0;
1451 if (!vma || addr + len <= vma->vm_start) {
1453 * Remember the place where we stopped the search:
1455 mm->free_area_cache = addr + len;
1458 if (addr + mm->cached_hole_size < vma->vm_start)
1459 mm->cached_hole_size = vma->vm_start - addr;
1465 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1468 * Is this a new hole at the lowest possible address?
1470 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
1471 mm->free_area_cache = addr;
1475 * This mmap-allocator allocates new areas top-down from below the
1476 * stack's low limit (the base):
1478 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1480 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1481 const unsigned long len, const unsigned long pgoff,
1482 const unsigned long flags)
1484 struct vm_area_struct *vma;
1485 struct mm_struct *mm = current->mm;
1486 unsigned long addr = addr0, start_addr;
1488 /* requested length too big for entire address space */
1489 if (len > TASK_SIZE)
1492 if (flags & MAP_FIXED)
1495 /* requesting a specific address */
1497 addr = PAGE_ALIGN(addr);
1498 vma = find_vma(mm, addr);
1499 if (TASK_SIZE - len >= addr &&
1500 (!vma || addr + len <= vma->vm_start))
1504 /* check if free_area_cache is useful for us */
1505 if (len <= mm->cached_hole_size) {
1506 mm->cached_hole_size = 0;
1507 mm->free_area_cache = mm->mmap_base;
1511 /* either no address requested or can't fit in requested address hole */
1512 start_addr = addr = mm->free_area_cache;
1520 * Lookup failure means no vma is above this address,
1521 * else if new region fits below vma->vm_start,
1522 * return with success:
1524 vma = find_vma(mm, addr);
1525 if (!vma || addr+len <= vma->vm_start)
1526 /* remember the address as a hint for next time */
1527 return (mm->free_area_cache = addr);
1529 /* remember the largest hole we saw so far */
1530 if (addr + mm->cached_hole_size < vma->vm_start)
1531 mm->cached_hole_size = vma->vm_start - addr;
1533 /* try just below the current vma->vm_start */
1534 addr = vma->vm_start-len;
1535 } while (len < vma->vm_start);
1539 * if hint left us with no space for the requested
1540 * mapping then try again:
1542 * Note: this is different with the case of bottomup
1543 * which does the fully line-search, but we use find_vma
1544 * here that causes some holes skipped.
1546 if (start_addr != mm->mmap_base) {
1547 mm->free_area_cache = mm->mmap_base;
1548 mm->cached_hole_size = 0;
1553 * A failed mmap() very likely causes application failure,
1554 * so fall back to the bottom-up function here. This scenario
1555 * can happen with large stack limits and large mmap()
1558 mm->cached_hole_size = ~0UL;
1559 mm->free_area_cache = TASK_UNMAPPED_BASE;
1560 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1562 * Restore the topdown base:
1564 mm->free_area_cache = mm->mmap_base;
1565 mm->cached_hole_size = ~0UL;
1571 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1574 * Is this a new hole at the highest possible address?
1576 if (addr > mm->free_area_cache)
1577 mm->free_area_cache = addr;
1579 /* dont allow allocations above current base */
1580 if (mm->free_area_cache > mm->mmap_base)
1581 mm->free_area_cache = mm->mmap_base;
1585 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1586 unsigned long pgoff, unsigned long flags)
1588 unsigned long (*get_area)(struct file *, unsigned long,
1589 unsigned long, unsigned long, unsigned long);
1591 unsigned long error = arch_mmap_check(addr, len, flags);
1595 /* Careful about overflows.. */
1596 if (len > TASK_SIZE)
1599 get_area = current->mm->get_unmapped_area;
1600 if (file && file->f_op && file->f_op->get_unmapped_area)
1601 get_area = file->f_op->get_unmapped_area;
1602 addr = get_area(file, addr, len, pgoff, flags);
1603 if (IS_ERR_VALUE(addr))
1606 if (addr > TASK_SIZE - len)
1608 if (addr & ~PAGE_MASK)
1611 return arch_rebalance_pgtables(addr, len);
1614 EXPORT_SYMBOL(get_unmapped_area);
1616 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1617 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1619 struct vm_area_struct *vma = NULL;
1622 /* Check the cache first. */
1623 /* (Cache hit rate is typically around 35%.) */
1624 vma = mm->mmap_cache;
1625 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1626 struct rb_node * rb_node;
1628 rb_node = mm->mm_rb.rb_node;
1632 struct vm_area_struct * vma_tmp;
1634 vma_tmp = rb_entry(rb_node,
1635 struct vm_area_struct, vm_rb);
1637 if (vma_tmp->vm_end > addr) {
1639 if (vma_tmp->vm_start <= addr)
1641 rb_node = rb_node->rb_left;
1643 rb_node = rb_node->rb_right;
1646 mm->mmap_cache = vma;
1652 EXPORT_SYMBOL(find_vma);
1655 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1657 struct vm_area_struct *
1658 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1659 struct vm_area_struct **pprev)
1661 struct vm_area_struct *vma;
1663 vma = find_vma(mm, addr);
1665 *pprev = vma->vm_prev;
1667 struct rb_node *rb_node = mm->mm_rb.rb_node;
1670 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1671 rb_node = rb_node->rb_right;
1678 * Verify that the stack growth is acceptable and
1679 * update accounting. This is shared with both the
1680 * grow-up and grow-down cases.
1682 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1684 struct mm_struct *mm = vma->vm_mm;
1685 struct rlimit *rlim = current->signal->rlim;
1686 unsigned long new_start;
1688 /* address space limit tests */
1689 if (!may_expand_vm(mm, grow))
1692 /* Stack limit test */
1693 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1696 /* mlock limit tests */
1697 if (vma->vm_flags & VM_LOCKED) {
1698 unsigned long locked;
1699 unsigned long limit;
1700 locked = mm->locked_vm + grow;
1701 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1702 limit >>= PAGE_SHIFT;
1703 if (locked > limit && !capable(CAP_IPC_LOCK))
1707 /* Check to ensure the stack will not grow into a hugetlb-only region */
1708 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1710 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1714 * Overcommit.. This must be the final test, as it will
1715 * update security statistics.
1717 if (security_vm_enough_memory_mm(mm, grow))
1720 /* Ok, everything looks good - let it rip */
1721 mm->total_vm += grow;
1722 if (vma->vm_flags & VM_LOCKED)
1723 mm->locked_vm += grow;
1724 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1728 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1730 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1731 * vma is the last one with address > vma->vm_end. Have to extend vma.
1733 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1737 if (!(vma->vm_flags & VM_GROWSUP))
1741 * We must make sure the anon_vma is allocated
1742 * so that the anon_vma locking is not a noop.
1744 if (unlikely(anon_vma_prepare(vma)))
1746 vma_lock_anon_vma(vma);
1749 * vma->vm_start/vm_end cannot change under us because the caller
1750 * is required to hold the mmap_sem in read mode. We need the
1751 * anon_vma lock to serialize against concurrent expand_stacks.
1752 * Also guard against wrapping around to address 0.
1754 if (address < PAGE_ALIGN(address+4))
1755 address = PAGE_ALIGN(address+4);
1757 vma_unlock_anon_vma(vma);
1762 /* Somebody else might have raced and expanded it already */
1763 if (address > vma->vm_end) {
1764 unsigned long size, grow;
1766 size = address - vma->vm_start;
1767 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1770 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1771 error = acct_stack_growth(vma, size, grow);
1773 vma->vm_end = address;
1774 perf_event_mmap(vma);
1778 vma_unlock_anon_vma(vma);
1779 khugepaged_enter_vma_merge(vma);
1782 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1785 * vma is the first one with address < vma->vm_start. Have to extend vma.
1787 int expand_downwards(struct vm_area_struct *vma,
1788 unsigned long address)
1793 * We must make sure the anon_vma is allocated
1794 * so that the anon_vma locking is not a noop.
1796 if (unlikely(anon_vma_prepare(vma)))
1799 address &= PAGE_MASK;
1800 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1804 vma_lock_anon_vma(vma);
1807 * vma->vm_start/vm_end cannot change under us because the caller
1808 * is required to hold the mmap_sem in read mode. We need the
1809 * anon_vma lock to serialize against concurrent expand_stacks.
1812 /* Somebody else might have raced and expanded it already */
1813 if (address < vma->vm_start) {
1814 unsigned long size, grow;
1816 size = vma->vm_end - address;
1817 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1820 if (grow <= vma->vm_pgoff) {
1821 error = acct_stack_growth(vma, size, grow);
1823 vma->vm_start = address;
1824 vma->vm_pgoff -= grow;
1825 perf_event_mmap(vma);
1829 vma_unlock_anon_vma(vma);
1830 khugepaged_enter_vma_merge(vma);
1834 #ifdef CONFIG_STACK_GROWSUP
1835 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1837 return expand_upwards(vma, address);
1840 struct vm_area_struct *
1841 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1843 struct vm_area_struct *vma, *prev;
1846 vma = find_vma_prev(mm, addr, &prev);
1847 if (vma && (vma->vm_start <= addr))
1849 if (!prev || expand_stack(prev, addr))
1851 if (prev->vm_flags & VM_LOCKED) {
1852 mlock_vma_pages_range(prev, addr, prev->vm_end);
1857 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1859 return expand_downwards(vma, address);
1862 struct vm_area_struct *
1863 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1865 struct vm_area_struct * vma;
1866 unsigned long start;
1869 vma = find_vma(mm,addr);
1872 if (vma->vm_start <= addr)
1874 if (!(vma->vm_flags & VM_GROWSDOWN))
1876 start = vma->vm_start;
1877 if (expand_stack(vma, addr))
1879 if (vma->vm_flags & VM_LOCKED) {
1880 mlock_vma_pages_range(vma, addr, start);
1887 * Ok - we have the memory areas we should free on the vma list,
1888 * so release them, and do the vma updates.
1890 * Called with the mm semaphore held.
1892 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1894 /* Update high watermark before we lower total_vm */
1895 update_hiwater_vm(mm);
1897 long nrpages = vma_pages(vma);
1899 mm->total_vm -= nrpages;
1900 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1901 vma = remove_vma(vma);
1907 * Get rid of page table information in the indicated region.
1909 * Called with the mm semaphore held.
1911 static void unmap_region(struct mm_struct *mm,
1912 struct vm_area_struct *vma, struct vm_area_struct *prev,
1913 unsigned long start, unsigned long end)
1915 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1916 struct mmu_gather tlb;
1917 unsigned long nr_accounted = 0;
1920 tlb_gather_mmu(&tlb, mm, 0);
1921 update_hiwater_rss(mm);
1922 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1923 vm_unacct_memory(nr_accounted);
1924 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1925 next ? next->vm_start : 0);
1926 tlb_finish_mmu(&tlb, start, end);
1930 * Create a list of vma's touched by the unmap, removing them from the mm's
1931 * vma list as we go..
1934 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1935 struct vm_area_struct *prev, unsigned long end)
1937 struct vm_area_struct **insertion_point;
1938 struct vm_area_struct *tail_vma = NULL;
1941 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1942 vma->vm_prev = NULL;
1944 rb_erase(&vma->vm_rb, &mm->mm_rb);
1948 } while (vma && vma->vm_start < end);
1949 *insertion_point = vma;
1951 vma->vm_prev = prev;
1952 tail_vma->vm_next = NULL;
1953 if (mm->unmap_area == arch_unmap_area)
1954 addr = prev ? prev->vm_end : mm->mmap_base;
1956 addr = vma ? vma->vm_start : mm->mmap_base;
1957 mm->unmap_area(mm, addr);
1958 mm->mmap_cache = NULL; /* Kill the cache. */
1962 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1963 * munmap path where it doesn't make sense to fail.
1965 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1966 unsigned long addr, int new_below)
1968 struct mempolicy *pol;
1969 struct vm_area_struct *new;
1972 if (is_vm_hugetlb_page(vma) && (addr &
1973 ~(huge_page_mask(hstate_vma(vma)))))
1976 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1980 /* most fields are the same, copy all, and then fixup */
1983 INIT_LIST_HEAD(&new->anon_vma_chain);
1988 new->vm_start = addr;
1989 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1992 pol = mpol_dup(vma_policy(vma));
1997 vma_set_policy(new, pol);
1999 if (anon_vma_clone(new, vma))
2003 get_file(new->vm_file);
2004 if (vma->vm_flags & VM_EXECUTABLE)
2005 added_exe_file_vma(mm);
2008 if (new->vm_ops && new->vm_ops->open)
2009 new->vm_ops->open(new);
2012 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2013 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2015 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2021 /* Clean everything up if vma_adjust failed. */
2022 if (new->vm_ops && new->vm_ops->close)
2023 new->vm_ops->close(new);
2025 if (vma->vm_flags & VM_EXECUTABLE)
2026 removed_exe_file_vma(mm);
2029 unlink_anon_vmas(new);
2033 kmem_cache_free(vm_area_cachep, new);
2039 * Split a vma into two pieces at address 'addr', a new vma is allocated
2040 * either for the first part or the tail.
2042 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2043 unsigned long addr, int new_below)
2045 if (mm->map_count >= sysctl_max_map_count)
2048 return __split_vma(mm, vma, addr, new_below);
2051 /* Munmap is split into 2 main parts -- this part which finds
2052 * what needs doing, and the areas themselves, which do the
2053 * work. This now handles partial unmappings.
2054 * Jeremy Fitzhardinge <jeremy@goop.org>
2056 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2059 struct vm_area_struct *vma, *prev, *last;
2061 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2064 if ((len = PAGE_ALIGN(len)) == 0)
2067 /* Find the first overlapping VMA */
2068 vma = find_vma(mm, start);
2071 prev = vma->vm_prev;
2072 /* we have start < vma->vm_end */
2074 /* if it doesn't overlap, we have nothing.. */
2076 if (vma->vm_start >= end)
2080 * If we need to split any vma, do it now to save pain later.
2082 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2083 * unmapped vm_area_struct will remain in use: so lower split_vma
2084 * places tmp vma above, and higher split_vma places tmp vma below.
2086 if (start > vma->vm_start) {
2090 * Make sure that map_count on return from munmap() will
2091 * not exceed its limit; but let map_count go just above
2092 * its limit temporarily, to help free resources as expected.
2094 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2097 error = __split_vma(mm, vma, start, 0);
2103 /* Does it split the last one? */
2104 last = find_vma(mm, end);
2105 if (last && end > last->vm_start) {
2106 int error = __split_vma(mm, last, end, 1);
2110 vma = prev? prev->vm_next: mm->mmap;
2113 * unlock any mlock()ed ranges before detaching vmas
2115 if (mm->locked_vm) {
2116 struct vm_area_struct *tmp = vma;
2117 while (tmp && tmp->vm_start < end) {
2118 if (tmp->vm_flags & VM_LOCKED) {
2119 mm->locked_vm -= vma_pages(tmp);
2120 munlock_vma_pages_all(tmp);
2127 * Remove the vma's, and unmap the actual pages
2129 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2130 unmap_region(mm, vma, prev, start, end);
2132 /* Fix up all other VM information */
2133 remove_vma_list(mm, vma);
2137 EXPORT_SYMBOL(do_munmap);
2139 int vm_munmap(unsigned long start, size_t len)
2142 struct mm_struct *mm = current->mm;
2144 down_write(&mm->mmap_sem);
2145 ret = do_munmap(mm, start, len);
2146 up_write(&mm->mmap_sem);
2149 EXPORT_SYMBOL(vm_munmap);
2151 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2153 profile_munmap(addr);
2154 return vm_munmap(addr, len);
2157 static inline void verify_mm_writelocked(struct mm_struct *mm)
2159 #ifdef CONFIG_DEBUG_VM
2160 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2162 up_read(&mm->mmap_sem);
2168 * this is really a simplified "do_mmap". it only handles
2169 * anonymous maps. eventually we may be able to do some
2170 * brk-specific accounting here.
2172 static unsigned long do_brk(unsigned long addr, unsigned long len)
2174 struct mm_struct * mm = current->mm;
2175 struct vm_area_struct * vma, * prev;
2176 unsigned long flags;
2177 struct rb_node ** rb_link, * rb_parent;
2178 pgoff_t pgoff = addr >> PAGE_SHIFT;
2181 len = PAGE_ALIGN(len);
2185 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2189 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2191 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2192 if (error & ~PAGE_MASK)
2198 if (mm->def_flags & VM_LOCKED) {
2199 unsigned long locked, lock_limit;
2200 locked = len >> PAGE_SHIFT;
2201 locked += mm->locked_vm;
2202 lock_limit = rlimit(RLIMIT_MEMLOCK);
2203 lock_limit >>= PAGE_SHIFT;
2204 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2209 * mm->mmap_sem is required to protect against another thread
2210 * changing the mappings in case we sleep.
2212 verify_mm_writelocked(mm);
2215 * Clear old maps. this also does some error checking for us
2218 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2219 if (vma && vma->vm_start < addr + len) {
2220 if (do_munmap(mm, addr, len))
2225 /* Check against address space limits *after* clearing old maps... */
2226 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2229 if (mm->map_count > sysctl_max_map_count)
2232 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2235 /* Can we just expand an old private anonymous mapping? */
2236 vma = vma_merge(mm, prev, addr, addr + len, flags,
2237 NULL, NULL, pgoff, NULL);
2242 * create a vma struct for an anonymous mapping
2244 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2246 vm_unacct_memory(len >> PAGE_SHIFT);
2250 INIT_LIST_HEAD(&vma->anon_vma_chain);
2252 vma->vm_start = addr;
2253 vma->vm_end = addr + len;
2254 vma->vm_pgoff = pgoff;
2255 vma->vm_flags = flags;
2256 vma->vm_page_prot = vm_get_page_prot(flags);
2257 vma_link(mm, vma, prev, rb_link, rb_parent);
2259 perf_event_mmap(vma);
2260 mm->total_vm += len >> PAGE_SHIFT;
2261 if (flags & VM_LOCKED) {
2262 if (!mlock_vma_pages_range(vma, addr, addr + len))
2263 mm->locked_vm += (len >> PAGE_SHIFT);
2268 unsigned long vm_brk(unsigned long addr, unsigned long len)
2270 struct mm_struct *mm = current->mm;
2273 down_write(&mm->mmap_sem);
2274 ret = do_brk(addr, len);
2275 up_write(&mm->mmap_sem);
2278 EXPORT_SYMBOL(vm_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;
2287 /* mm's last user has gone, and its about to be pulled down */
2288 mmu_notifier_release(mm);
2290 if (mm->locked_vm) {
2293 if (vma->vm_flags & VM_LOCKED)
2294 munlock_vma_pages_all(vma);
2302 if (!vma) /* Can happen if dup_mmap() received an OOM */
2307 tlb_gather_mmu(&tlb, mm, 1);
2308 /* update_hiwater_rss(mm) here? but nobody should be looking */
2309 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2310 unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2311 vm_unacct_memory(nr_accounted);
2313 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2314 tlb_finish_mmu(&tlb, 0, -1);
2317 * Walk the list again, actually closing and freeing it,
2318 * with preemption enabled, without holding any MM locks.
2321 vma = remove_vma(vma);
2323 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2326 /* Insert vm structure into process list sorted by address
2327 * and into the inode's i_mmap tree. If vm_file is non-NULL
2328 * then i_mmap_mutex is taken here.
2330 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2332 struct vm_area_struct * __vma, * prev;
2333 struct rb_node ** rb_link, * rb_parent;
2336 * The vm_pgoff of a purely anonymous vma should be irrelevant
2337 * until its first write fault, when page's anon_vma and index
2338 * are set. But now set the vm_pgoff it will almost certainly
2339 * end up with (unless mremap moves it elsewhere before that
2340 * first wfault), so /proc/pid/maps tells a consistent story.
2342 * By setting it to reflect the virtual start address of the
2343 * vma, merges and splits can happen in a seamless way, just
2344 * using the existing file pgoff checks and manipulations.
2345 * Similarly in do_mmap_pgoff and in do_brk.
2347 if (!vma->vm_file) {
2348 BUG_ON(vma->anon_vma);
2349 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2351 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2352 if (__vma && __vma->vm_start < vma->vm_end)
2354 if ((vma->vm_flags & VM_ACCOUNT) &&
2355 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2357 vma_link(mm, vma, prev, rb_link, rb_parent);
2362 * Copy the vma structure to a new location in the same mm,
2363 * prior to moving page table entries, to effect an mremap move.
2365 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2366 unsigned long addr, unsigned long len, pgoff_t pgoff)
2368 struct vm_area_struct *vma = *vmap;
2369 unsigned long vma_start = vma->vm_start;
2370 struct mm_struct *mm = vma->vm_mm;
2371 struct vm_area_struct *new_vma, *prev;
2372 struct rb_node **rb_link, *rb_parent;
2373 struct mempolicy *pol;
2374 bool faulted_in_anon_vma = true;
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 (unlikely(!vma->vm_file && !vma->anon_vma)) {
2381 pgoff = addr >> PAGE_SHIFT;
2382 faulted_in_anon_vma = false;
2385 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2386 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2387 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2390 * Source vma may have been merged into new_vma
2392 if (unlikely(vma_start >= new_vma->vm_start &&
2393 vma_start < new_vma->vm_end)) {
2395 * The only way we can get a vma_merge with
2396 * self during an mremap is if the vma hasn't
2397 * been faulted in yet and we were allowed to
2398 * reset the dst vma->vm_pgoff to the
2399 * destination address of the mremap to allow
2400 * the merge to happen. mremap must change the
2401 * vm_pgoff linearity between src and dst vmas
2402 * (in turn preventing a vma_merge) to be
2403 * safe. It is only safe to keep the vm_pgoff
2404 * linear if there are no pages mapped yet.
2406 VM_BUG_ON(faulted_in_anon_vma);
2409 anon_vma_moveto_tail(new_vma);
2411 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2414 pol = mpol_dup(vma_policy(vma));
2417 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2418 if (anon_vma_clone(new_vma, vma))
2419 goto out_free_mempol;
2420 vma_set_policy(new_vma, pol);
2421 new_vma->vm_start = addr;
2422 new_vma->vm_end = addr + len;
2423 new_vma->vm_pgoff = pgoff;
2424 if (new_vma->vm_file) {
2425 get_file(new_vma->vm_file);
2426 if (vma->vm_flags & VM_EXECUTABLE)
2427 added_exe_file_vma(mm);
2429 if (new_vma->vm_ops && new_vma->vm_ops->open)
2430 new_vma->vm_ops->open(new_vma);
2431 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2439 kmem_cache_free(vm_area_cachep, new_vma);
2444 * Return true if the calling process may expand its vm space by the passed
2447 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2449 unsigned long cur = mm->total_vm; /* pages */
2452 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2454 if (cur + npages > lim)
2460 static int special_mapping_fault(struct vm_area_struct *vma,
2461 struct vm_fault *vmf)
2464 struct page **pages;
2467 * special mappings have no vm_file, and in that case, the mm
2468 * uses vm_pgoff internally. So we have to subtract it from here.
2469 * We are allowed to do this because we are the mm; do not copy
2470 * this code into drivers!
2472 pgoff = vmf->pgoff - vma->vm_pgoff;
2474 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2478 struct page *page = *pages;
2484 return VM_FAULT_SIGBUS;
2488 * Having a close hook prevents vma merging regardless of flags.
2490 static void special_mapping_close(struct vm_area_struct *vma)
2494 static const struct vm_operations_struct special_mapping_vmops = {
2495 .close = special_mapping_close,
2496 .fault = special_mapping_fault,
2500 * Called with mm->mmap_sem held for writing.
2501 * Insert a new vma covering the given region, with the given flags.
2502 * Its pages are supplied by the given array of struct page *.
2503 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2504 * The region past the last page supplied will always produce SIGBUS.
2505 * The array pointer and the pages it points to are assumed to stay alive
2506 * for as long as this mapping might exist.
2508 int install_special_mapping(struct mm_struct *mm,
2509 unsigned long addr, unsigned long len,
2510 unsigned long vm_flags, struct page **pages)
2513 struct vm_area_struct *vma;
2515 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2516 if (unlikely(vma == NULL))
2519 INIT_LIST_HEAD(&vma->anon_vma_chain);
2521 vma->vm_start = addr;
2522 vma->vm_end = addr + len;
2524 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2525 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2527 vma->vm_ops = &special_mapping_vmops;
2528 vma->vm_private_data = pages;
2530 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2534 ret = insert_vm_struct(mm, vma);
2538 mm->total_vm += len >> PAGE_SHIFT;
2540 perf_event_mmap(vma);
2545 kmem_cache_free(vm_area_cachep, vma);
2549 static DEFINE_MUTEX(mm_all_locks_mutex);
2551 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2553 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2555 * The LSB of head.next can't change from under us
2556 * because we hold the mm_all_locks_mutex.
2558 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2560 * We can safely modify head.next after taking the
2561 * anon_vma->root->mutex. If some other vma in this mm shares
2562 * the same anon_vma we won't take it again.
2564 * No need of atomic instructions here, head.next
2565 * can't change from under us thanks to the
2566 * anon_vma->root->mutex.
2568 if (__test_and_set_bit(0, (unsigned long *)
2569 &anon_vma->root->head.next))
2574 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2576 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2578 * AS_MM_ALL_LOCKS can't change from under us because
2579 * we hold the mm_all_locks_mutex.
2581 * Operations on ->flags have to be atomic because
2582 * even if AS_MM_ALL_LOCKS is stable thanks to the
2583 * mm_all_locks_mutex, there may be other cpus
2584 * changing other bitflags in parallel to us.
2586 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2588 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2593 * This operation locks against the VM for all pte/vma/mm related
2594 * operations that could ever happen on a certain mm. This includes
2595 * vmtruncate, try_to_unmap, and all page faults.
2597 * The caller must take the mmap_sem in write mode before calling
2598 * mm_take_all_locks(). The caller isn't allowed to release the
2599 * mmap_sem until mm_drop_all_locks() returns.
2601 * mmap_sem in write mode is required in order to block all operations
2602 * that could modify pagetables and free pages without need of
2603 * altering the vma layout (for example populate_range() with
2604 * nonlinear vmas). It's also needed in write mode to avoid new
2605 * anon_vmas to be associated with existing vmas.
2607 * A single task can't take more than one mm_take_all_locks() in a row
2608 * or it would deadlock.
2610 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2611 * mapping->flags avoid to take the same lock twice, if more than one
2612 * vma in this mm is backed by the same anon_vma or address_space.
2614 * We can take all the locks in random order because the VM code
2615 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2616 * takes more than one of them in a row. Secondly we're protected
2617 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2619 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2620 * that may have to take thousand of locks.
2622 * mm_take_all_locks() can fail if it's interrupted by signals.
2624 int mm_take_all_locks(struct mm_struct *mm)
2626 struct vm_area_struct *vma;
2627 struct anon_vma_chain *avc;
2629 BUG_ON(down_read_trylock(&mm->mmap_sem));
2631 mutex_lock(&mm_all_locks_mutex);
2633 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2634 if (signal_pending(current))
2636 if (vma->vm_file && vma->vm_file->f_mapping)
2637 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2640 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2641 if (signal_pending(current))
2644 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2645 vm_lock_anon_vma(mm, avc->anon_vma);
2651 mm_drop_all_locks(mm);
2655 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2657 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2659 * The LSB of head.next can't change to 0 from under
2660 * us because we hold the mm_all_locks_mutex.
2662 * We must however clear the bitflag before unlocking
2663 * the vma so the users using the anon_vma->head will
2664 * never see our bitflag.
2666 * No need of atomic instructions here, head.next
2667 * can't change from under us until we release the
2668 * anon_vma->root->mutex.
2670 if (!__test_and_clear_bit(0, (unsigned long *)
2671 &anon_vma->root->head.next))
2673 anon_vma_unlock(anon_vma);
2677 static void vm_unlock_mapping(struct address_space *mapping)
2679 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2681 * AS_MM_ALL_LOCKS can't change to 0 from under us
2682 * because we hold the mm_all_locks_mutex.
2684 mutex_unlock(&mapping->i_mmap_mutex);
2685 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2692 * The mmap_sem cannot be released by the caller until
2693 * mm_drop_all_locks() returns.
2695 void mm_drop_all_locks(struct mm_struct *mm)
2697 struct vm_area_struct *vma;
2698 struct anon_vma_chain *avc;
2700 BUG_ON(down_read_trylock(&mm->mmap_sem));
2701 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2703 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2705 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2706 vm_unlock_anon_vma(avc->anon_vma);
2707 if (vma->vm_file && vma->vm_file->f_mapping)
2708 vm_unlock_mapping(vma->vm_file->f_mapping);
2711 mutex_unlock(&mm_all_locks_mutex);
2715 * initialise the VMA slab
2717 void __init mmap_init(void)
2721 ret = percpu_counter_init(&vm_committed_as, 0);