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>
33 #include <linux/uprobes.h>
35 #include <asm/uaccess.h>
36 #include <asm/cacheflush.h>
38 #include <asm/mmu_context.h>
42 #ifndef arch_mmap_check
43 #define arch_mmap_check(addr, len, flags) (0)
46 #ifndef arch_rebalance_pgtables
47 #define arch_rebalance_pgtables(addr, len) (addr)
50 static void unmap_region(struct mm_struct *mm,
51 struct vm_area_struct *vma, struct vm_area_struct *prev,
52 unsigned long start, unsigned long end);
55 * WARNING: the debugging will use recursive algorithms so never enable this
56 * unless you know what you are doing.
60 /* description of effects of mapping type and prot in current implementation.
61 * this is due to the limited x86 page protection hardware. The expected
62 * behavior is in parens:
65 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
66 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
67 * w: (no) no w: (no) no w: (yes) yes w: (no) no
68 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
70 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (copy) copy w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
75 pgprot_t protection_map[16] = {
76 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
77 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
80 pgprot_t vm_get_page_prot(unsigned long vm_flags)
82 return __pgprot(pgprot_val(protection_map[vm_flags &
83 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
84 pgprot_val(arch_vm_get_page_prot(vm_flags)));
86 EXPORT_SYMBOL(vm_get_page_prot);
88 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
89 int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
90 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
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 SYSCALL_DEFINE1(brk, unsigned long, brk)
246 unsigned long rlim, retval;
247 unsigned long newbrk, oldbrk;
248 struct mm_struct *mm = current->mm;
249 unsigned long min_brk;
251 down_write(&mm->mmap_sem);
253 #ifdef CONFIG_COMPAT_BRK
255 * CONFIG_COMPAT_BRK can still be overridden by setting
256 * randomize_va_space to 2, which will still cause mm->start_brk
257 * to be arbitrarily shifted
259 if (current->brk_randomized)
260 min_brk = mm->start_brk;
262 min_brk = mm->end_data;
264 min_brk = mm->start_brk;
270 * Check against rlimit here. If this check is done later after the test
271 * of oldbrk with newbrk then it can escape the test and let the data
272 * segment grow beyond its set limit the in case where the limit is
273 * not page aligned -Ram Gupta
275 rlim = rlimit(RLIMIT_DATA);
276 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
277 (mm->end_data - mm->start_data) > rlim)
280 newbrk = PAGE_ALIGN(brk);
281 oldbrk = PAGE_ALIGN(mm->brk);
282 if (oldbrk == newbrk)
285 /* Always allow shrinking brk. */
286 if (brk <= mm->brk) {
287 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
292 /* Check against existing mmap mappings. */
293 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
296 /* Ok, looks good - let it rip. */
297 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
303 up_write(&mm->mmap_sem);
308 static int browse_rb(struct rb_root *root)
311 struct rb_node *nd, *pn = NULL;
312 unsigned long prev = 0, pend = 0;
314 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
315 struct vm_area_struct *vma;
316 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
317 if (vma->vm_start < prev)
318 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
319 if (vma->vm_start < pend)
320 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
321 if (vma->vm_start > vma->vm_end)
322 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
325 prev = vma->vm_start;
329 for (nd = pn; nd; nd = rb_prev(nd)) {
333 printk("backwards %d, forwards %d\n", j, i), i = 0;
337 void validate_mm(struct mm_struct *mm)
341 struct vm_area_struct *tmp = mm->mmap;
346 if (i != mm->map_count)
347 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
348 i = browse_rb(&mm->mm_rb);
349 if (i != mm->map_count)
350 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
354 #define validate_mm(mm) do { } while (0)
357 static struct vm_area_struct *
358 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
359 struct vm_area_struct **pprev, struct rb_node ***rb_link,
360 struct rb_node ** rb_parent)
362 struct vm_area_struct * vma;
363 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
365 __rb_link = &mm->mm_rb.rb_node;
366 rb_prev = __rb_parent = NULL;
370 struct vm_area_struct *vma_tmp;
372 __rb_parent = *__rb_link;
373 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
375 if (vma_tmp->vm_end > addr) {
377 if (vma_tmp->vm_start <= addr)
379 __rb_link = &__rb_parent->rb_left;
381 rb_prev = __rb_parent;
382 __rb_link = &__rb_parent->rb_right;
388 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
389 *rb_link = __rb_link;
390 *rb_parent = __rb_parent;
394 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
395 struct rb_node **rb_link, struct rb_node *rb_parent)
397 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
398 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
401 static void __vma_link_file(struct vm_area_struct *vma)
407 struct address_space *mapping = file->f_mapping;
409 if (vma->vm_flags & VM_DENYWRITE)
410 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
411 if (vma->vm_flags & VM_SHARED)
412 mapping->i_mmap_writable++;
414 flush_dcache_mmap_lock(mapping);
415 if (unlikely(vma->vm_flags & VM_NONLINEAR))
416 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
418 vma_prio_tree_insert(vma, &mapping->i_mmap);
419 flush_dcache_mmap_unlock(mapping);
424 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
425 struct vm_area_struct *prev, struct rb_node **rb_link,
426 struct rb_node *rb_parent)
428 __vma_link_list(mm, vma, prev, rb_parent);
429 __vma_link_rb(mm, vma, rb_link, rb_parent);
432 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
433 struct vm_area_struct *prev, struct rb_node **rb_link,
434 struct rb_node *rb_parent)
436 struct address_space *mapping = NULL;
439 mapping = vma->vm_file->f_mapping;
442 mutex_lock(&mapping->i_mmap_mutex);
444 __vma_link(mm, vma, prev, rb_link, rb_parent);
445 __vma_link_file(vma);
448 mutex_unlock(&mapping->i_mmap_mutex);
455 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
456 * mm's list and rbtree. It has already been inserted into the prio_tree.
458 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
460 struct vm_area_struct *__vma, *prev;
461 struct rb_node **rb_link, *rb_parent;
463 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
464 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
465 __vma_link(mm, vma, prev, rb_link, rb_parent);
470 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
471 struct vm_area_struct *prev)
473 struct vm_area_struct *next = vma->vm_next;
475 prev->vm_next = next;
477 next->vm_prev = prev;
478 rb_erase(&vma->vm_rb, &mm->mm_rb);
479 if (mm->mmap_cache == vma)
480 mm->mmap_cache = prev;
484 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
485 * is already present in an i_mmap tree without adjusting the tree.
486 * The following helper function should be used when such adjustments
487 * are necessary. The "insert" vma (if any) is to be inserted
488 * before we drop the necessary locks.
490 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
491 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
493 struct mm_struct *mm = vma->vm_mm;
494 struct vm_area_struct *next = vma->vm_next;
495 struct vm_area_struct *importer = NULL;
496 struct address_space *mapping = NULL;
497 struct prio_tree_root *root = NULL;
498 struct anon_vma *anon_vma = NULL;
499 struct file *file = vma->vm_file;
500 long adjust_next = 0;
503 if (next && !insert) {
504 struct vm_area_struct *exporter = NULL;
506 if (end >= next->vm_end) {
508 * vma expands, overlapping all the next, and
509 * perhaps the one after too (mprotect case 6).
511 again: remove_next = 1 + (end > next->vm_end);
515 } else if (end > next->vm_start) {
517 * vma expands, overlapping part of the next:
518 * mprotect case 5 shifting the boundary up.
520 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
523 } else if (end < vma->vm_end) {
525 * vma shrinks, and !insert tells it's not
526 * split_vma inserting another: so it must be
527 * mprotect case 4 shifting the boundary down.
529 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
535 * Easily overlooked: when mprotect shifts the boundary,
536 * make sure the expanding vma has anon_vma set if the
537 * shrinking vma had, to cover any anon pages imported.
539 if (exporter && exporter->anon_vma && !importer->anon_vma) {
540 if (anon_vma_clone(importer, exporter))
542 importer->anon_vma = exporter->anon_vma;
547 mapping = file->f_mapping;
548 if (!(vma->vm_flags & VM_NONLINEAR)) {
549 root = &mapping->i_mmap;
550 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
553 uprobe_munmap(next, next->vm_start,
557 mutex_lock(&mapping->i_mmap_mutex);
560 * Put into prio_tree now, so instantiated pages
561 * are visible to arm/parisc __flush_dcache_page
562 * throughout; but we cannot insert into address
563 * space until vma start or end is updated.
565 __vma_link_file(insert);
569 vma_adjust_trans_huge(vma, start, end, adjust_next);
572 * When changing only vma->vm_end, we don't really need anon_vma
573 * lock. This is a fairly rare case by itself, but the anon_vma
574 * lock may be shared between many sibling processes. Skipping
575 * the lock for brk adjustments makes a difference sometimes.
577 if (vma->anon_vma && (importer || start != vma->vm_start)) {
578 anon_vma = vma->anon_vma;
579 anon_vma_lock(anon_vma);
583 flush_dcache_mmap_lock(mapping);
584 vma_prio_tree_remove(vma, root);
586 vma_prio_tree_remove(next, root);
589 vma->vm_start = start;
591 vma->vm_pgoff = pgoff;
593 next->vm_start += adjust_next << PAGE_SHIFT;
594 next->vm_pgoff += adjust_next;
599 vma_prio_tree_insert(next, root);
600 vma_prio_tree_insert(vma, root);
601 flush_dcache_mmap_unlock(mapping);
606 * vma_merge has merged next into vma, and needs
607 * us to remove next before dropping the locks.
609 __vma_unlink(mm, next, vma);
611 __remove_shared_vm_struct(next, file, mapping);
614 * split_vma has split insert from vma, and needs
615 * us to insert it before dropping the locks
616 * (it may either follow vma or precede it).
618 __insert_vm_struct(mm, insert);
622 anon_vma_unlock(anon_vma);
624 mutex_unlock(&mapping->i_mmap_mutex);
635 uprobe_munmap(next, next->vm_start, next->vm_end);
637 if (next->vm_flags & VM_EXECUTABLE)
638 removed_exe_file_vma(mm);
641 anon_vma_merge(vma, next);
643 mpol_put(vma_policy(next));
644 kmem_cache_free(vm_area_cachep, next);
646 * In mprotect's case 6 (see comments on vma_merge),
647 * we must remove another next too. It would clutter
648 * up the code too much to do both in one go.
650 if (remove_next == 2) {
664 * If the vma has a ->close operation then the driver probably needs to release
665 * per-vma resources, so we don't attempt to merge those.
667 static inline int is_mergeable_vma(struct vm_area_struct *vma,
668 struct file *file, unsigned long vm_flags)
670 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
671 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
673 if (vma->vm_file != file)
675 if (vma->vm_ops && vma->vm_ops->close)
680 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
681 struct anon_vma *anon_vma2,
682 struct vm_area_struct *vma)
685 * The list_is_singular() test is to avoid merging VMA cloned from
686 * parents. This can improve scalability caused by anon_vma lock.
688 if ((!anon_vma1 || !anon_vma2) && (!vma ||
689 list_is_singular(&vma->anon_vma_chain)))
691 return anon_vma1 == anon_vma2;
695 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
696 * in front of (at a lower virtual address and file offset than) the vma.
698 * We cannot merge two vmas if they have differently assigned (non-NULL)
699 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
701 * We don't check here for the merged mmap wrapping around the end of pagecache
702 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
703 * wrap, nor mmaps which cover the final page at index -1UL.
706 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
707 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
709 if (is_mergeable_vma(vma, file, vm_flags) &&
710 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
711 if (vma->vm_pgoff == vm_pgoff)
718 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
719 * beyond (at a higher virtual address and file offset than) the vma.
721 * We cannot merge two vmas if they have differently assigned (non-NULL)
722 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
725 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
726 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
728 if (is_mergeable_vma(vma, file, vm_flags) &&
729 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
731 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
732 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
739 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
740 * whether that can be merged with its predecessor or its successor.
741 * Or both (it neatly fills a hole).
743 * In most cases - when called for mmap, brk or mremap - [addr,end) is
744 * certain not to be mapped by the time vma_merge is called; but when
745 * called for mprotect, it is certain to be already mapped (either at
746 * an offset within prev, or at the start of next), and the flags of
747 * this area are about to be changed to vm_flags - and the no-change
748 * case has already been eliminated.
750 * The following mprotect cases have to be considered, where AAAA is
751 * the area passed down from mprotect_fixup, never extending beyond one
752 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
754 * AAAA AAAA AAAA AAAA
755 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
756 * cannot merge might become might become might become
757 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
758 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
759 * mremap move: PPPPNNNNNNNN 8
761 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
762 * might become case 1 below case 2 below case 3 below
764 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
765 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
767 struct vm_area_struct *vma_merge(struct mm_struct *mm,
768 struct vm_area_struct *prev, unsigned long addr,
769 unsigned long end, unsigned long vm_flags,
770 struct anon_vma *anon_vma, struct file *file,
771 pgoff_t pgoff, struct mempolicy *policy)
773 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
774 struct vm_area_struct *area, *next;
778 * We later require that vma->vm_flags == vm_flags,
779 * so this tests vma->vm_flags & VM_SPECIAL, too.
781 if (vm_flags & VM_SPECIAL)
785 next = prev->vm_next;
789 if (next && next->vm_end == end) /* cases 6, 7, 8 */
790 next = next->vm_next;
793 * Can it merge with the predecessor?
795 if (prev && prev->vm_end == addr &&
796 mpol_equal(vma_policy(prev), policy) &&
797 can_vma_merge_after(prev, vm_flags,
798 anon_vma, file, pgoff)) {
800 * OK, it can. Can we now merge in the successor as well?
802 if (next && end == next->vm_start &&
803 mpol_equal(policy, vma_policy(next)) &&
804 can_vma_merge_before(next, vm_flags,
805 anon_vma, file, pgoff+pglen) &&
806 is_mergeable_anon_vma(prev->anon_vma,
807 next->anon_vma, NULL)) {
809 err = vma_adjust(prev, prev->vm_start,
810 next->vm_end, prev->vm_pgoff, NULL);
811 } else /* cases 2, 5, 7 */
812 err = vma_adjust(prev, prev->vm_start,
813 end, prev->vm_pgoff, NULL);
816 khugepaged_enter_vma_merge(prev);
821 * Can this new request be merged in front of next?
823 if (next && end == next->vm_start &&
824 mpol_equal(policy, vma_policy(next)) &&
825 can_vma_merge_before(next, vm_flags,
826 anon_vma, file, pgoff+pglen)) {
827 if (prev && addr < prev->vm_end) /* case 4 */
828 err = vma_adjust(prev, prev->vm_start,
829 addr, prev->vm_pgoff, NULL);
830 else /* cases 3, 8 */
831 err = vma_adjust(area, addr, next->vm_end,
832 next->vm_pgoff - pglen, NULL);
835 khugepaged_enter_vma_merge(area);
843 * Rough compatbility check to quickly see if it's even worth looking
844 * at sharing an anon_vma.
846 * They need to have the same vm_file, and the flags can only differ
847 * in things that mprotect may change.
849 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
850 * we can merge the two vma's. For example, we refuse to merge a vma if
851 * there is a vm_ops->close() function, because that indicates that the
852 * driver is doing some kind of reference counting. But that doesn't
853 * really matter for the anon_vma sharing case.
855 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
857 return a->vm_end == b->vm_start &&
858 mpol_equal(vma_policy(a), vma_policy(b)) &&
859 a->vm_file == b->vm_file &&
860 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
861 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
865 * Do some basic sanity checking to see if we can re-use the anon_vma
866 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
867 * the same as 'old', the other will be the new one that is trying
868 * to share the anon_vma.
870 * NOTE! This runs with mm_sem held for reading, so it is possible that
871 * the anon_vma of 'old' is concurrently in the process of being set up
872 * by another page fault trying to merge _that_. But that's ok: if it
873 * is being set up, that automatically means that it will be a singleton
874 * acceptable for merging, so we can do all of this optimistically. But
875 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
877 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
878 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
879 * is to return an anon_vma that is "complex" due to having gone through
882 * We also make sure that the two vma's are compatible (adjacent,
883 * and with the same memory policies). That's all stable, even with just
884 * a read lock on the mm_sem.
886 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
888 if (anon_vma_compatible(a, b)) {
889 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
891 if (anon_vma && list_is_singular(&old->anon_vma_chain))
898 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
899 * neighbouring vmas for a suitable anon_vma, before it goes off
900 * to allocate a new anon_vma. It checks because a repetitive
901 * sequence of mprotects and faults may otherwise lead to distinct
902 * anon_vmas being allocated, preventing vma merge in subsequent
905 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
907 struct anon_vma *anon_vma;
908 struct vm_area_struct *near;
914 anon_vma = reusable_anon_vma(near, vma, near);
922 anon_vma = reusable_anon_vma(near, near, vma);
927 * There's no absolute need to look only at touching neighbours:
928 * we could search further afield for "compatible" anon_vmas.
929 * But it would probably just be a waste of time searching,
930 * or lead to too many vmas hanging off the same anon_vma.
931 * We're trying to allow mprotect remerging later on,
932 * not trying to minimize memory used for anon_vmas.
937 #ifdef CONFIG_PROC_FS
938 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
939 struct file *file, long pages)
941 const unsigned long stack_flags
942 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
945 mm->shared_vm += pages;
946 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
947 mm->exec_vm += pages;
948 } else if (flags & stack_flags)
949 mm->stack_vm += pages;
950 if (flags & (VM_RESERVED|VM_IO))
951 mm->reserved_vm += pages;
953 #endif /* CONFIG_PROC_FS */
956 * If a hint addr is less than mmap_min_addr change hint to be as
957 * low as possible but still greater than mmap_min_addr
959 static inline unsigned long round_hint_to_min(unsigned long hint)
962 if (((void *)hint != NULL) &&
963 (hint < mmap_min_addr))
964 return PAGE_ALIGN(mmap_min_addr);
969 * The caller must hold down_write(¤t->mm->mmap_sem).
972 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
973 unsigned long len, unsigned long prot,
974 unsigned long flags, unsigned long pgoff)
976 struct mm_struct * mm = current->mm;
980 unsigned long reqprot = prot;
983 * Does the application expect PROT_READ to imply PROT_EXEC?
985 * (the exception is when the underlying filesystem is noexec
986 * mounted, in which case we dont add PROT_EXEC.)
988 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
989 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
995 if (!(flags & MAP_FIXED))
996 addr = round_hint_to_min(addr);
998 /* Careful about overflows.. */
999 len = PAGE_ALIGN(len);
1003 /* offset overflow? */
1004 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1007 /* Too many mappings? */
1008 if (mm->map_count > sysctl_max_map_count)
1011 /* Obtain the address to map to. we verify (or select) it and ensure
1012 * that it represents a valid section of the address space.
1014 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1015 if (addr & ~PAGE_MASK)
1018 /* Do simple checking here so the lower-level routines won't have
1019 * to. we assume access permissions have been handled by the open
1020 * of the memory object, so we don't do any here.
1022 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1023 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1025 if (flags & MAP_LOCKED)
1026 if (!can_do_mlock())
1029 /* mlock MCL_FUTURE? */
1030 if (vm_flags & VM_LOCKED) {
1031 unsigned long locked, lock_limit;
1032 locked = len >> PAGE_SHIFT;
1033 locked += mm->locked_vm;
1034 lock_limit = rlimit(RLIMIT_MEMLOCK);
1035 lock_limit >>= PAGE_SHIFT;
1036 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1040 inode = file ? file->f_path.dentry->d_inode : NULL;
1043 switch (flags & MAP_TYPE) {
1045 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1049 * Make sure we don't allow writing to an append-only
1052 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1056 * Make sure there are no mandatory locks on the file.
1058 if (locks_verify_locked(inode))
1061 vm_flags |= VM_SHARED | VM_MAYSHARE;
1062 if (!(file->f_mode & FMODE_WRITE))
1063 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1067 if (!(file->f_mode & FMODE_READ))
1069 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1070 if (vm_flags & VM_EXEC)
1072 vm_flags &= ~VM_MAYEXEC;
1075 if (!file->f_op || !file->f_op->mmap)
1083 switch (flags & MAP_TYPE) {
1089 vm_flags |= VM_SHARED | VM_MAYSHARE;
1093 * Set pgoff according to addr for anon_vma.
1095 pgoff = addr >> PAGE_SHIFT;
1102 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1106 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1108 EXPORT_SYMBOL(do_mmap_pgoff);
1110 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1111 unsigned long, prot, unsigned long, flags,
1112 unsigned long, fd, unsigned long, pgoff)
1114 struct file *file = NULL;
1115 unsigned long retval = -EBADF;
1117 if (!(flags & MAP_ANONYMOUS)) {
1118 audit_mmap_fd(fd, flags);
1119 if (unlikely(flags & MAP_HUGETLB))
1124 } else if (flags & MAP_HUGETLB) {
1125 struct user_struct *user = NULL;
1127 * VM_NORESERVE is used because the reservations will be
1128 * taken when vm_ops->mmap() is called
1129 * A dummy user value is used because we are not locking
1130 * memory so no accounting is necessary
1132 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1133 VM_NORESERVE, &user,
1134 HUGETLB_ANONHUGE_INODE);
1136 return PTR_ERR(file);
1139 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1141 down_write(¤t->mm->mmap_sem);
1142 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1143 up_write(¤t->mm->mmap_sem);
1151 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1152 struct mmap_arg_struct {
1156 unsigned long flags;
1158 unsigned long offset;
1161 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1163 struct mmap_arg_struct a;
1165 if (copy_from_user(&a, arg, sizeof(a)))
1167 if (a.offset & ~PAGE_MASK)
1170 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1171 a.offset >> PAGE_SHIFT);
1173 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1176 * Some shared mappigns will want the pages marked read-only
1177 * to track write events. If so, we'll downgrade vm_page_prot
1178 * to the private version (using protection_map[] without the
1181 int vma_wants_writenotify(struct vm_area_struct *vma)
1183 vm_flags_t vm_flags = vma->vm_flags;
1185 /* If it was private or non-writable, the write bit is already clear */
1186 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1189 /* The backer wishes to know when pages are first written to? */
1190 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1193 /* The open routine did something to the protections already? */
1194 if (pgprot_val(vma->vm_page_prot) !=
1195 pgprot_val(vm_get_page_prot(vm_flags)))
1198 /* Specialty mapping? */
1199 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1202 /* Can the mapping track the dirty pages? */
1203 return vma->vm_file && vma->vm_file->f_mapping &&
1204 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1208 * We account for memory if it's a private writeable mapping,
1209 * not hugepages and VM_NORESERVE wasn't set.
1211 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1214 * hugetlb has its own accounting separate from the core VM
1215 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1217 if (file && is_file_hugepages(file))
1220 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1223 unsigned long mmap_region(struct file *file, unsigned long addr,
1224 unsigned long len, unsigned long flags,
1225 vm_flags_t vm_flags, unsigned long pgoff)
1227 struct mm_struct *mm = current->mm;
1228 struct vm_area_struct *vma, *prev;
1229 int correct_wcount = 0;
1231 struct rb_node **rb_link, *rb_parent;
1232 unsigned long charged = 0;
1233 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1235 /* Clear old maps */
1238 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1239 if (vma && vma->vm_start < addr + len) {
1240 if (do_munmap(mm, addr, len))
1245 /* Check against address space limit. */
1246 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1250 * Set 'VM_NORESERVE' if we should not account for the
1251 * memory use of this mapping.
1253 if ((flags & MAP_NORESERVE)) {
1254 /* We honor MAP_NORESERVE if allowed to overcommit */
1255 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1256 vm_flags |= VM_NORESERVE;
1258 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1259 if (file && is_file_hugepages(file))
1260 vm_flags |= VM_NORESERVE;
1264 * Private writable mapping: check memory availability
1266 if (accountable_mapping(file, vm_flags)) {
1267 charged = len >> PAGE_SHIFT;
1268 if (security_vm_enough_memory_mm(mm, charged))
1270 vm_flags |= VM_ACCOUNT;
1274 * Can we just expand an old mapping?
1276 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1281 * Determine the object being mapped and call the appropriate
1282 * specific mapper. the address has already been validated, but
1283 * not unmapped, but the maps are removed from the list.
1285 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1292 vma->vm_start = addr;
1293 vma->vm_end = addr + len;
1294 vma->vm_flags = vm_flags;
1295 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1296 vma->vm_pgoff = pgoff;
1297 INIT_LIST_HEAD(&vma->anon_vma_chain);
1299 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1302 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1304 if (vm_flags & VM_DENYWRITE) {
1305 error = deny_write_access(file);
1310 vma->vm_file = file;
1312 error = file->f_op->mmap(file, vma);
1314 goto unmap_and_free_vma;
1315 if (vm_flags & VM_EXECUTABLE)
1316 added_exe_file_vma(mm);
1318 /* Can addr have changed??
1320 * Answer: Yes, several device drivers can do it in their
1321 * f_op->mmap method. -DaveM
1323 addr = vma->vm_start;
1324 pgoff = vma->vm_pgoff;
1325 vm_flags = vma->vm_flags;
1326 } else if (vm_flags & VM_SHARED) {
1327 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1329 error = shmem_zero_setup(vma);
1334 if (vma_wants_writenotify(vma)) {
1335 pgprot_t pprot = vma->vm_page_prot;
1337 /* Can vma->vm_page_prot have changed??
1339 * Answer: Yes, drivers may have changed it in their
1340 * f_op->mmap method.
1342 * Ensures that vmas marked as uncached stay that way.
1344 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1345 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1346 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1349 vma_link(mm, vma, prev, rb_link, rb_parent);
1350 file = vma->vm_file;
1352 /* Once vma denies write, undo our temporary denial count */
1354 atomic_inc(&inode->i_writecount);
1356 perf_event_mmap(vma);
1358 mm->total_vm += len >> PAGE_SHIFT;
1359 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1360 if (vm_flags & VM_LOCKED) {
1361 if (!mlock_vma_pages_range(vma, addr, addr + len))
1362 mm->locked_vm += (len >> PAGE_SHIFT);
1363 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1364 make_pages_present(addr, addr + len);
1366 if (file && uprobe_mmap(vma))
1367 /* matching probes but cannot insert */
1368 goto unmap_and_free_vma;
1374 atomic_inc(&inode->i_writecount);
1375 vma->vm_file = NULL;
1378 /* Undo any partial mapping done by a device driver. */
1379 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1382 kmem_cache_free(vm_area_cachep, vma);
1385 vm_unacct_memory(charged);
1389 /* Get an address range which is currently unmapped.
1390 * For shmat() with addr=0.
1392 * Ugly calling convention alert:
1393 * Return value with the low bits set means error value,
1395 * if (ret & ~PAGE_MASK)
1398 * This function "knows" that -ENOMEM has the bits set.
1400 #ifndef HAVE_ARCH_UNMAPPED_AREA
1402 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1403 unsigned long len, unsigned long pgoff, unsigned long flags)
1405 struct mm_struct *mm = current->mm;
1406 struct vm_area_struct *vma;
1407 unsigned long start_addr;
1409 if (len > TASK_SIZE)
1412 if (flags & MAP_FIXED)
1416 addr = PAGE_ALIGN(addr);
1417 vma = find_vma(mm, addr);
1418 if (TASK_SIZE - len >= addr &&
1419 (!vma || addr + len <= vma->vm_start))
1422 if (len > mm->cached_hole_size) {
1423 start_addr = addr = mm->free_area_cache;
1425 start_addr = addr = TASK_UNMAPPED_BASE;
1426 mm->cached_hole_size = 0;
1430 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1431 /* At this point: (!vma || addr < vma->vm_end). */
1432 if (TASK_SIZE - len < addr) {
1434 * Start a new search - just in case we missed
1437 if (start_addr != TASK_UNMAPPED_BASE) {
1438 addr = TASK_UNMAPPED_BASE;
1440 mm->cached_hole_size = 0;
1445 if (!vma || addr + len <= vma->vm_start) {
1447 * Remember the place where we stopped the search:
1449 mm->free_area_cache = addr + len;
1452 if (addr + mm->cached_hole_size < vma->vm_start)
1453 mm->cached_hole_size = vma->vm_start - addr;
1459 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1462 * Is this a new hole at the lowest possible address?
1464 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
1465 mm->free_area_cache = addr;
1469 * This mmap-allocator allocates new areas top-down from below the
1470 * stack's low limit (the base):
1472 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1474 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1475 const unsigned long len, const unsigned long pgoff,
1476 const unsigned long flags)
1478 struct vm_area_struct *vma;
1479 struct mm_struct *mm = current->mm;
1480 unsigned long addr = addr0, start_addr;
1482 /* requested length too big for entire address space */
1483 if (len > TASK_SIZE)
1486 if (flags & MAP_FIXED)
1489 /* requesting a specific address */
1491 addr = PAGE_ALIGN(addr);
1492 vma = find_vma(mm, addr);
1493 if (TASK_SIZE - len >= addr &&
1494 (!vma || addr + len <= vma->vm_start))
1498 /* check if free_area_cache is useful for us */
1499 if (len <= mm->cached_hole_size) {
1500 mm->cached_hole_size = 0;
1501 mm->free_area_cache = mm->mmap_base;
1505 /* either no address requested or can't fit in requested address hole */
1506 start_addr = addr = mm->free_area_cache;
1514 * Lookup failure means no vma is above this address,
1515 * else if new region fits below vma->vm_start,
1516 * return with success:
1518 vma = find_vma(mm, addr);
1519 if (!vma || addr+len <= vma->vm_start)
1520 /* remember the address as a hint for next time */
1521 return (mm->free_area_cache = addr);
1523 /* remember the largest hole we saw so far */
1524 if (addr + mm->cached_hole_size < vma->vm_start)
1525 mm->cached_hole_size = vma->vm_start - addr;
1527 /* try just below the current vma->vm_start */
1528 addr = vma->vm_start-len;
1529 } while (len < vma->vm_start);
1533 * if hint left us with no space for the requested
1534 * mapping then try again:
1536 * Note: this is different with the case of bottomup
1537 * which does the fully line-search, but we use find_vma
1538 * here that causes some holes skipped.
1540 if (start_addr != mm->mmap_base) {
1541 mm->free_area_cache = mm->mmap_base;
1542 mm->cached_hole_size = 0;
1547 * A failed mmap() very likely causes application failure,
1548 * so fall back to the bottom-up function here. This scenario
1549 * can happen with large stack limits and large mmap()
1552 mm->cached_hole_size = ~0UL;
1553 mm->free_area_cache = TASK_UNMAPPED_BASE;
1554 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1556 * Restore the topdown base:
1558 mm->free_area_cache = mm->mmap_base;
1559 mm->cached_hole_size = ~0UL;
1565 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1568 * Is this a new hole at the highest possible address?
1570 if (addr > mm->free_area_cache)
1571 mm->free_area_cache = addr;
1573 /* dont allow allocations above current base */
1574 if (mm->free_area_cache > mm->mmap_base)
1575 mm->free_area_cache = mm->mmap_base;
1579 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1580 unsigned long pgoff, unsigned long flags)
1582 unsigned long (*get_area)(struct file *, unsigned long,
1583 unsigned long, unsigned long, unsigned long);
1585 unsigned long error = arch_mmap_check(addr, len, flags);
1589 /* Careful about overflows.. */
1590 if (len > TASK_SIZE)
1593 get_area = current->mm->get_unmapped_area;
1594 if (file && file->f_op && file->f_op->get_unmapped_area)
1595 get_area = file->f_op->get_unmapped_area;
1596 addr = get_area(file, addr, len, pgoff, flags);
1597 if (IS_ERR_VALUE(addr))
1600 if (addr > TASK_SIZE - len)
1602 if (addr & ~PAGE_MASK)
1605 return arch_rebalance_pgtables(addr, len);
1608 EXPORT_SYMBOL(get_unmapped_area);
1610 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1611 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1613 struct vm_area_struct *vma = NULL;
1616 /* Check the cache first. */
1617 /* (Cache hit rate is typically around 35%.) */
1618 vma = mm->mmap_cache;
1619 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1620 struct rb_node * rb_node;
1622 rb_node = mm->mm_rb.rb_node;
1626 struct vm_area_struct * vma_tmp;
1628 vma_tmp = rb_entry(rb_node,
1629 struct vm_area_struct, vm_rb);
1631 if (vma_tmp->vm_end > addr) {
1633 if (vma_tmp->vm_start <= addr)
1635 rb_node = rb_node->rb_left;
1637 rb_node = rb_node->rb_right;
1640 mm->mmap_cache = vma;
1646 EXPORT_SYMBOL(find_vma);
1649 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1651 struct vm_area_struct *
1652 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1653 struct vm_area_struct **pprev)
1655 struct vm_area_struct *vma;
1657 vma = find_vma(mm, addr);
1659 *pprev = vma->vm_prev;
1661 struct rb_node *rb_node = mm->mm_rb.rb_node;
1664 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1665 rb_node = rb_node->rb_right;
1672 * Verify that the stack growth is acceptable and
1673 * update accounting. This is shared with both the
1674 * grow-up and grow-down cases.
1676 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1678 struct mm_struct *mm = vma->vm_mm;
1679 struct rlimit *rlim = current->signal->rlim;
1680 unsigned long new_start;
1682 /* address space limit tests */
1683 if (!may_expand_vm(mm, grow))
1686 /* Stack limit test */
1687 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1690 /* mlock limit tests */
1691 if (vma->vm_flags & VM_LOCKED) {
1692 unsigned long locked;
1693 unsigned long limit;
1694 locked = mm->locked_vm + grow;
1695 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1696 limit >>= PAGE_SHIFT;
1697 if (locked > limit && !capable(CAP_IPC_LOCK))
1701 /* Check to ensure the stack will not grow into a hugetlb-only region */
1702 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1704 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1708 * Overcommit.. This must be the final test, as it will
1709 * update security statistics.
1711 if (security_vm_enough_memory_mm(mm, grow))
1714 /* Ok, everything looks good - let it rip */
1715 mm->total_vm += grow;
1716 if (vma->vm_flags & VM_LOCKED)
1717 mm->locked_vm += grow;
1718 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1722 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1724 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1725 * vma is the last one with address > vma->vm_end. Have to extend vma.
1727 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1731 if (!(vma->vm_flags & VM_GROWSUP))
1735 * We must make sure the anon_vma is allocated
1736 * so that the anon_vma locking is not a noop.
1738 if (unlikely(anon_vma_prepare(vma)))
1740 vma_lock_anon_vma(vma);
1743 * vma->vm_start/vm_end cannot change under us because the caller
1744 * is required to hold the mmap_sem in read mode. We need the
1745 * anon_vma lock to serialize against concurrent expand_stacks.
1746 * Also guard against wrapping around to address 0.
1748 if (address < PAGE_ALIGN(address+4))
1749 address = PAGE_ALIGN(address+4);
1751 vma_unlock_anon_vma(vma);
1756 /* Somebody else might have raced and expanded it already */
1757 if (address > vma->vm_end) {
1758 unsigned long size, grow;
1760 size = address - vma->vm_start;
1761 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1764 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1765 error = acct_stack_growth(vma, size, grow);
1767 vma->vm_end = address;
1768 perf_event_mmap(vma);
1772 vma_unlock_anon_vma(vma);
1773 khugepaged_enter_vma_merge(vma);
1776 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1779 * vma is the first one with address < vma->vm_start. Have to extend vma.
1781 int expand_downwards(struct vm_area_struct *vma,
1782 unsigned long address)
1787 * We must make sure the anon_vma is allocated
1788 * so that the anon_vma locking is not a noop.
1790 if (unlikely(anon_vma_prepare(vma)))
1793 address &= PAGE_MASK;
1794 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1798 vma_lock_anon_vma(vma);
1801 * vma->vm_start/vm_end cannot change under us because the caller
1802 * is required to hold the mmap_sem in read mode. We need the
1803 * anon_vma lock to serialize against concurrent expand_stacks.
1806 /* Somebody else might have raced and expanded it already */
1807 if (address < vma->vm_start) {
1808 unsigned long size, grow;
1810 size = vma->vm_end - address;
1811 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1814 if (grow <= vma->vm_pgoff) {
1815 error = acct_stack_growth(vma, size, grow);
1817 vma->vm_start = address;
1818 vma->vm_pgoff -= grow;
1819 perf_event_mmap(vma);
1823 vma_unlock_anon_vma(vma);
1824 khugepaged_enter_vma_merge(vma);
1828 #ifdef CONFIG_STACK_GROWSUP
1829 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1831 return expand_upwards(vma, address);
1834 struct vm_area_struct *
1835 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1837 struct vm_area_struct *vma, *prev;
1840 vma = find_vma_prev(mm, addr, &prev);
1841 if (vma && (vma->vm_start <= addr))
1843 if (!prev || expand_stack(prev, addr))
1845 if (prev->vm_flags & VM_LOCKED) {
1846 mlock_vma_pages_range(prev, addr, prev->vm_end);
1851 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1853 return expand_downwards(vma, address);
1856 struct vm_area_struct *
1857 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1859 struct vm_area_struct * vma;
1860 unsigned long start;
1863 vma = find_vma(mm,addr);
1866 if (vma->vm_start <= addr)
1868 if (!(vma->vm_flags & VM_GROWSDOWN))
1870 start = vma->vm_start;
1871 if (expand_stack(vma, addr))
1873 if (vma->vm_flags & VM_LOCKED) {
1874 mlock_vma_pages_range(vma, addr, start);
1881 * Ok - we have the memory areas we should free on the vma list,
1882 * so release them, and do the vma updates.
1884 * Called with the mm semaphore held.
1886 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1888 /* Update high watermark before we lower total_vm */
1889 update_hiwater_vm(mm);
1891 long nrpages = vma_pages(vma);
1893 mm->total_vm -= nrpages;
1894 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1895 vma = remove_vma(vma);
1901 * Get rid of page table information in the indicated region.
1903 * Called with the mm semaphore held.
1905 static void unmap_region(struct mm_struct *mm,
1906 struct vm_area_struct *vma, struct vm_area_struct *prev,
1907 unsigned long start, unsigned long end)
1909 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1910 struct mmu_gather tlb;
1911 unsigned long nr_accounted = 0;
1914 tlb_gather_mmu(&tlb, mm, 0);
1915 update_hiwater_rss(mm);
1916 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1917 vm_unacct_memory(nr_accounted);
1918 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1919 next ? next->vm_start : 0);
1920 tlb_finish_mmu(&tlb, start, end);
1924 * Create a list of vma's touched by the unmap, removing them from the mm's
1925 * vma list as we go..
1928 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1929 struct vm_area_struct *prev, unsigned long end)
1931 struct vm_area_struct **insertion_point;
1932 struct vm_area_struct *tail_vma = NULL;
1935 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1936 vma->vm_prev = NULL;
1938 rb_erase(&vma->vm_rb, &mm->mm_rb);
1942 } while (vma && vma->vm_start < end);
1943 *insertion_point = vma;
1945 vma->vm_prev = prev;
1946 tail_vma->vm_next = NULL;
1947 if (mm->unmap_area == arch_unmap_area)
1948 addr = prev ? prev->vm_end : mm->mmap_base;
1950 addr = vma ? vma->vm_start : mm->mmap_base;
1951 mm->unmap_area(mm, addr);
1952 mm->mmap_cache = NULL; /* Kill the cache. */
1956 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1957 * munmap path where it doesn't make sense to fail.
1959 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1960 unsigned long addr, int new_below)
1962 struct mempolicy *pol;
1963 struct vm_area_struct *new;
1966 if (is_vm_hugetlb_page(vma) && (addr &
1967 ~(huge_page_mask(hstate_vma(vma)))))
1970 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1974 /* most fields are the same, copy all, and then fixup */
1977 INIT_LIST_HEAD(&new->anon_vma_chain);
1982 new->vm_start = addr;
1983 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1986 pol = mpol_dup(vma_policy(vma));
1991 vma_set_policy(new, pol);
1993 if (anon_vma_clone(new, vma))
1997 get_file(new->vm_file);
1998 if (vma->vm_flags & VM_EXECUTABLE)
1999 added_exe_file_vma(mm);
2002 if (new->vm_ops && new->vm_ops->open)
2003 new->vm_ops->open(new);
2006 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2007 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2009 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2015 /* Clean everything up if vma_adjust failed. */
2016 if (new->vm_ops && new->vm_ops->close)
2017 new->vm_ops->close(new);
2019 if (vma->vm_flags & VM_EXECUTABLE)
2020 removed_exe_file_vma(mm);
2023 unlink_anon_vmas(new);
2027 kmem_cache_free(vm_area_cachep, new);
2033 * Split a vma into two pieces at address 'addr', a new vma is allocated
2034 * either for the first part or the tail.
2036 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2037 unsigned long addr, int new_below)
2039 if (mm->map_count >= sysctl_max_map_count)
2042 return __split_vma(mm, vma, addr, new_below);
2045 /* Munmap is split into 2 main parts -- this part which finds
2046 * what needs doing, and the areas themselves, which do the
2047 * work. This now handles partial unmappings.
2048 * Jeremy Fitzhardinge <jeremy@goop.org>
2050 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2053 struct vm_area_struct *vma, *prev, *last;
2055 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2058 if ((len = PAGE_ALIGN(len)) == 0)
2061 /* Find the first overlapping VMA */
2062 vma = find_vma(mm, start);
2065 prev = vma->vm_prev;
2066 /* we have start < vma->vm_end */
2068 /* if it doesn't overlap, we have nothing.. */
2070 if (vma->vm_start >= end)
2074 * If we need to split any vma, do it now to save pain later.
2076 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2077 * unmapped vm_area_struct will remain in use: so lower split_vma
2078 * places tmp vma above, and higher split_vma places tmp vma below.
2080 if (start > vma->vm_start) {
2084 * Make sure that map_count on return from munmap() will
2085 * not exceed its limit; but let map_count go just above
2086 * its limit temporarily, to help free resources as expected.
2088 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2091 error = __split_vma(mm, vma, start, 0);
2097 /* Does it split the last one? */
2098 last = find_vma(mm, end);
2099 if (last && end > last->vm_start) {
2100 int error = __split_vma(mm, last, end, 1);
2104 vma = prev? prev->vm_next: mm->mmap;
2107 * unlock any mlock()ed ranges before detaching vmas
2109 if (mm->locked_vm) {
2110 struct vm_area_struct *tmp = vma;
2111 while (tmp && tmp->vm_start < end) {
2112 if (tmp->vm_flags & VM_LOCKED) {
2113 mm->locked_vm -= vma_pages(tmp);
2114 munlock_vma_pages_all(tmp);
2121 * Remove the vma's, and unmap the actual pages
2123 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2124 unmap_region(mm, vma, prev, start, end);
2126 /* Fix up all other VM information */
2127 remove_vma_list(mm, vma);
2132 EXPORT_SYMBOL(do_munmap);
2134 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2137 struct mm_struct *mm = current->mm;
2139 profile_munmap(addr);
2141 down_write(&mm->mmap_sem);
2142 ret = do_munmap(mm, addr, len);
2143 up_write(&mm->mmap_sem);
2147 static inline void verify_mm_writelocked(struct mm_struct *mm)
2149 #ifdef CONFIG_DEBUG_VM
2150 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2152 up_read(&mm->mmap_sem);
2158 * this is really a simplified "do_mmap". it only handles
2159 * anonymous maps. eventually we may be able to do some
2160 * brk-specific accounting here.
2162 unsigned long do_brk(unsigned long addr, unsigned long len)
2164 struct mm_struct * mm = current->mm;
2165 struct vm_area_struct * vma, * prev;
2166 unsigned long flags;
2167 struct rb_node ** rb_link, * rb_parent;
2168 pgoff_t pgoff = addr >> PAGE_SHIFT;
2171 len = PAGE_ALIGN(len);
2175 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2179 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2181 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2182 if (error & ~PAGE_MASK)
2188 if (mm->def_flags & VM_LOCKED) {
2189 unsigned long locked, lock_limit;
2190 locked = len >> PAGE_SHIFT;
2191 locked += mm->locked_vm;
2192 lock_limit = rlimit(RLIMIT_MEMLOCK);
2193 lock_limit >>= PAGE_SHIFT;
2194 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2199 * mm->mmap_sem is required to protect against another thread
2200 * changing the mappings in case we sleep.
2202 verify_mm_writelocked(mm);
2205 * Clear old maps. this also does some error checking for us
2208 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2209 if (vma && vma->vm_start < addr + len) {
2210 if (do_munmap(mm, addr, len))
2215 /* Check against address space limits *after* clearing old maps... */
2216 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2219 if (mm->map_count > sysctl_max_map_count)
2222 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2225 /* Can we just expand an old private anonymous mapping? */
2226 vma = vma_merge(mm, prev, addr, addr + len, flags,
2227 NULL, NULL, pgoff, NULL);
2232 * create a vma struct for an anonymous mapping
2234 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2236 vm_unacct_memory(len >> PAGE_SHIFT);
2240 INIT_LIST_HEAD(&vma->anon_vma_chain);
2242 vma->vm_start = addr;
2243 vma->vm_end = addr + len;
2244 vma->vm_pgoff = pgoff;
2245 vma->vm_flags = flags;
2246 vma->vm_page_prot = vm_get_page_prot(flags);
2247 vma_link(mm, vma, prev, rb_link, rb_parent);
2249 perf_event_mmap(vma);
2250 mm->total_vm += len >> PAGE_SHIFT;
2251 if (flags & VM_LOCKED) {
2252 if (!mlock_vma_pages_range(vma, addr, addr + len))
2253 mm->locked_vm += (len >> PAGE_SHIFT);
2258 EXPORT_SYMBOL(do_brk);
2260 /* Release all mmaps. */
2261 void exit_mmap(struct mm_struct *mm)
2263 struct mmu_gather tlb;
2264 struct vm_area_struct *vma;
2265 unsigned long nr_accounted = 0;
2267 /* mm's last user has gone, and its about to be pulled down */
2268 mmu_notifier_release(mm);
2270 if (mm->locked_vm) {
2273 if (vma->vm_flags & VM_LOCKED)
2274 munlock_vma_pages_all(vma);
2282 if (!vma) /* Can happen if dup_mmap() received an OOM */
2287 tlb_gather_mmu(&tlb, mm, 1);
2288 /* update_hiwater_rss(mm) here? but nobody should be looking */
2289 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2290 unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2291 vm_unacct_memory(nr_accounted);
2293 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2294 tlb_finish_mmu(&tlb, 0, -1);
2297 * Walk the list again, actually closing and freeing it,
2298 * with preemption enabled, without holding any MM locks.
2301 vma = remove_vma(vma);
2303 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2306 /* Insert vm structure into process list sorted by address
2307 * and into the inode's i_mmap tree. If vm_file is non-NULL
2308 * then i_mmap_mutex is taken here.
2310 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2312 struct vm_area_struct * __vma, * prev;
2313 struct rb_node ** rb_link, * rb_parent;
2316 * The vm_pgoff of a purely anonymous vma should be irrelevant
2317 * until its first write fault, when page's anon_vma and index
2318 * are set. But now set the vm_pgoff it will almost certainly
2319 * end up with (unless mremap moves it elsewhere before that
2320 * first wfault), so /proc/pid/maps tells a consistent story.
2322 * By setting it to reflect the virtual start address of the
2323 * vma, merges and splits can happen in a seamless way, just
2324 * using the existing file pgoff checks and manipulations.
2325 * Similarly in do_mmap_pgoff and in do_brk.
2327 if (!vma->vm_file) {
2328 BUG_ON(vma->anon_vma);
2329 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2331 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2332 if (__vma && __vma->vm_start < vma->vm_end)
2334 if ((vma->vm_flags & VM_ACCOUNT) &&
2335 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2338 if (vma->vm_file && uprobe_mmap(vma))
2341 vma_link(mm, vma, prev, rb_link, rb_parent);
2346 * Copy the vma structure to a new location in the same mm,
2347 * prior to moving page table entries, to effect an mremap move.
2349 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2350 unsigned long addr, unsigned long len, pgoff_t pgoff)
2352 struct vm_area_struct *vma = *vmap;
2353 unsigned long vma_start = vma->vm_start;
2354 struct mm_struct *mm = vma->vm_mm;
2355 struct vm_area_struct *new_vma, *prev;
2356 struct rb_node **rb_link, *rb_parent;
2357 struct mempolicy *pol;
2358 bool faulted_in_anon_vma = true;
2361 * If anonymous vma has not yet been faulted, update new pgoff
2362 * to match new location, to increase its chance of merging.
2364 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2365 pgoff = addr >> PAGE_SHIFT;
2366 faulted_in_anon_vma = false;
2369 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2370 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2371 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2374 * Source vma may have been merged into new_vma
2376 if (unlikely(vma_start >= new_vma->vm_start &&
2377 vma_start < new_vma->vm_end)) {
2379 * The only way we can get a vma_merge with
2380 * self during an mremap is if the vma hasn't
2381 * been faulted in yet and we were allowed to
2382 * reset the dst vma->vm_pgoff to the
2383 * destination address of the mremap to allow
2384 * the merge to happen. mremap must change the
2385 * vm_pgoff linearity between src and dst vmas
2386 * (in turn preventing a vma_merge) to be
2387 * safe. It is only safe to keep the vm_pgoff
2388 * linear if there are no pages mapped yet.
2390 VM_BUG_ON(faulted_in_anon_vma);
2393 anon_vma_moveto_tail(new_vma);
2395 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2398 pol = mpol_dup(vma_policy(vma));
2401 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2402 if (anon_vma_clone(new_vma, vma))
2403 goto out_free_mempol;
2404 vma_set_policy(new_vma, pol);
2405 new_vma->vm_start = addr;
2406 new_vma->vm_end = addr + len;
2407 new_vma->vm_pgoff = pgoff;
2408 if (new_vma->vm_file) {
2409 get_file(new_vma->vm_file);
2411 if (uprobe_mmap(new_vma))
2412 goto out_free_mempol;
2414 if (vma->vm_flags & VM_EXECUTABLE)
2415 added_exe_file_vma(mm);
2417 if (new_vma->vm_ops && new_vma->vm_ops->open)
2418 new_vma->vm_ops->open(new_vma);
2419 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2427 kmem_cache_free(vm_area_cachep, new_vma);
2432 * Return true if the calling process may expand its vm space by the passed
2435 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2437 unsigned long cur = mm->total_vm; /* pages */
2440 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2442 if (cur + npages > lim)
2448 static int special_mapping_fault(struct vm_area_struct *vma,
2449 struct vm_fault *vmf)
2452 struct page **pages;
2455 * special mappings have no vm_file, and in that case, the mm
2456 * uses vm_pgoff internally. So we have to subtract it from here.
2457 * We are allowed to do this because we are the mm; do not copy
2458 * this code into drivers!
2460 pgoff = vmf->pgoff - vma->vm_pgoff;
2462 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2466 struct page *page = *pages;
2472 return VM_FAULT_SIGBUS;
2476 * Having a close hook prevents vma merging regardless of flags.
2478 static void special_mapping_close(struct vm_area_struct *vma)
2482 static const struct vm_operations_struct special_mapping_vmops = {
2483 .close = special_mapping_close,
2484 .fault = special_mapping_fault,
2488 * Called with mm->mmap_sem held for writing.
2489 * Insert a new vma covering the given region, with the given flags.
2490 * Its pages are supplied by the given array of struct page *.
2491 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2492 * The region past the last page supplied will always produce SIGBUS.
2493 * The array pointer and the pages it points to are assumed to stay alive
2494 * for as long as this mapping might exist.
2496 int install_special_mapping(struct mm_struct *mm,
2497 unsigned long addr, unsigned long len,
2498 unsigned long vm_flags, struct page **pages)
2501 struct vm_area_struct *vma;
2503 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2504 if (unlikely(vma == NULL))
2507 INIT_LIST_HEAD(&vma->anon_vma_chain);
2509 vma->vm_start = addr;
2510 vma->vm_end = addr + len;
2512 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2513 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2515 vma->vm_ops = &special_mapping_vmops;
2516 vma->vm_private_data = pages;
2518 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2522 ret = insert_vm_struct(mm, vma);
2526 mm->total_vm += len >> PAGE_SHIFT;
2528 perf_event_mmap(vma);
2533 kmem_cache_free(vm_area_cachep, vma);
2537 static DEFINE_MUTEX(mm_all_locks_mutex);
2539 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2541 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2543 * The LSB of head.next can't change from under us
2544 * because we hold the mm_all_locks_mutex.
2546 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2548 * We can safely modify head.next after taking the
2549 * anon_vma->root->mutex. If some other vma in this mm shares
2550 * the same anon_vma we won't take it again.
2552 * No need of atomic instructions here, head.next
2553 * can't change from under us thanks to the
2554 * anon_vma->root->mutex.
2556 if (__test_and_set_bit(0, (unsigned long *)
2557 &anon_vma->root->head.next))
2562 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2564 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2566 * AS_MM_ALL_LOCKS can't change from under us because
2567 * we hold the mm_all_locks_mutex.
2569 * Operations on ->flags have to be atomic because
2570 * even if AS_MM_ALL_LOCKS is stable thanks to the
2571 * mm_all_locks_mutex, there may be other cpus
2572 * changing other bitflags in parallel to us.
2574 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2576 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2581 * This operation locks against the VM for all pte/vma/mm related
2582 * operations that could ever happen on a certain mm. This includes
2583 * vmtruncate, try_to_unmap, and all page faults.
2585 * The caller must take the mmap_sem in write mode before calling
2586 * mm_take_all_locks(). The caller isn't allowed to release the
2587 * mmap_sem until mm_drop_all_locks() returns.
2589 * mmap_sem in write mode is required in order to block all operations
2590 * that could modify pagetables and free pages without need of
2591 * altering the vma layout (for example populate_range() with
2592 * nonlinear vmas). It's also needed in write mode to avoid new
2593 * anon_vmas to be associated with existing vmas.
2595 * A single task can't take more than one mm_take_all_locks() in a row
2596 * or it would deadlock.
2598 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2599 * mapping->flags avoid to take the same lock twice, if more than one
2600 * vma in this mm is backed by the same anon_vma or address_space.
2602 * We can take all the locks in random order because the VM code
2603 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2604 * takes more than one of them in a row. Secondly we're protected
2605 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2607 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2608 * that may have to take thousand of locks.
2610 * mm_take_all_locks() can fail if it's interrupted by signals.
2612 int mm_take_all_locks(struct mm_struct *mm)
2614 struct vm_area_struct *vma;
2615 struct anon_vma_chain *avc;
2617 BUG_ON(down_read_trylock(&mm->mmap_sem));
2619 mutex_lock(&mm_all_locks_mutex);
2621 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2622 if (signal_pending(current))
2624 if (vma->vm_file && vma->vm_file->f_mapping)
2625 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2628 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2629 if (signal_pending(current))
2632 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2633 vm_lock_anon_vma(mm, avc->anon_vma);
2639 mm_drop_all_locks(mm);
2643 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2645 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2647 * The LSB of head.next can't change to 0 from under
2648 * us because we hold the mm_all_locks_mutex.
2650 * We must however clear the bitflag before unlocking
2651 * the vma so the users using the anon_vma->head will
2652 * never see our bitflag.
2654 * No need of atomic instructions here, head.next
2655 * can't change from under us until we release the
2656 * anon_vma->root->mutex.
2658 if (!__test_and_clear_bit(0, (unsigned long *)
2659 &anon_vma->root->head.next))
2661 anon_vma_unlock(anon_vma);
2665 static void vm_unlock_mapping(struct address_space *mapping)
2667 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2669 * AS_MM_ALL_LOCKS can't change to 0 from under us
2670 * because we hold the mm_all_locks_mutex.
2672 mutex_unlock(&mapping->i_mmap_mutex);
2673 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2680 * The mmap_sem cannot be released by the caller until
2681 * mm_drop_all_locks() returns.
2683 void mm_drop_all_locks(struct mm_struct *mm)
2685 struct vm_area_struct *vma;
2686 struct anon_vma_chain *avc;
2688 BUG_ON(down_read_trylock(&mm->mmap_sem));
2689 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2691 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2693 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2694 vm_unlock_anon_vma(avc->anon_vma);
2695 if (vma->vm_file && vma->vm_file->f_mapping)
2696 vm_unlock_mapping(vma->vm_file->f_mapping);
2699 mutex_unlock(&mm_all_locks_mutex);
2703 * initialise the VMA slab
2705 void __init mmap_init(void)
2709 ret = percpu_counter_init(&vm_committed_as, 0);