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);
54 /* description of effects of mapping type and prot in current implementation.
55 * this is due to the limited x86 page protection hardware. The expected
56 * behavior is in parens:
59 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
60 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
61 * w: (no) no w: (no) no w: (yes) yes w: (no) no
62 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
64 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
65 * w: (no) no w: (no) no w: (copy) copy w: (no) no
66 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
69 pgprot_t protection_map[16] = {
70 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
71 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
74 pgprot_t vm_get_page_prot(unsigned long vm_flags)
76 return __pgprot(pgprot_val(protection_map[vm_flags &
77 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
78 pgprot_val(arch_vm_get_page_prot(vm_flags)));
80 EXPORT_SYMBOL(vm_get_page_prot);
82 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
83 int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
84 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
86 * Make sure vm_committed_as in one cacheline and not cacheline shared with
87 * other variables. It can be updated by several CPUs frequently.
89 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
92 * Check that a process has enough memory to allocate a new virtual
93 * mapping. 0 means there is enough memory for the allocation to
94 * succeed and -ENOMEM implies there is not.
96 * We currently support three overcommit policies, which are set via the
97 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
99 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
100 * Additional code 2002 Jul 20 by Robert Love.
102 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
104 * Note this is a helper function intended to be used by LSMs which
105 * wish to use this logic.
107 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
109 unsigned long free, allowed;
111 vm_acct_memory(pages);
114 * Sometimes we want to use more memory than we have
116 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
119 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
120 free = global_page_state(NR_FREE_PAGES);
121 free += global_page_state(NR_FILE_PAGES);
124 * shmem pages shouldn't be counted as free in this
125 * case, they can't be purged, only swapped out, and
126 * that won't affect the overall amount of available
127 * memory in the system.
129 free -= global_page_state(NR_SHMEM);
131 free += nr_swap_pages;
134 * Any slabs which are created with the
135 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
136 * which are reclaimable, under pressure. The dentry
137 * cache and most inode caches should fall into this
139 free += global_page_state(NR_SLAB_RECLAIMABLE);
142 * Leave reserved pages. The pages are not for anonymous pages.
144 if (free <= totalreserve_pages)
147 free -= totalreserve_pages;
150 * Leave the last 3% for root
161 allowed = (totalram_pages - hugetlb_total_pages())
162 * sysctl_overcommit_ratio / 100;
164 * Leave the last 3% for root
167 allowed -= allowed / 32;
168 allowed += total_swap_pages;
170 /* Don't let a single process grow too big:
171 leave 3% of the size of this process for other processes */
173 allowed -= mm->total_vm / 32;
175 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
178 vm_unacct_memory(pages);
184 * Requires inode->i_mapping->i_mmap_mutex
186 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
187 struct file *file, struct address_space *mapping)
189 if (vma->vm_flags & VM_DENYWRITE)
190 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
191 if (vma->vm_flags & VM_SHARED)
192 mapping->i_mmap_writable--;
194 flush_dcache_mmap_lock(mapping);
195 if (unlikely(vma->vm_flags & VM_NONLINEAR))
196 list_del_init(&vma->shared.nonlinear);
198 vma_interval_tree_remove(vma, &mapping->i_mmap);
199 flush_dcache_mmap_unlock(mapping);
203 * Unlink a file-based vm structure from its interval tree, to hide
204 * vma from rmap and vmtruncate before freeing its page tables.
206 void unlink_file_vma(struct vm_area_struct *vma)
208 struct file *file = vma->vm_file;
211 struct address_space *mapping = file->f_mapping;
212 mutex_lock(&mapping->i_mmap_mutex);
213 __remove_shared_vm_struct(vma, file, mapping);
214 mutex_unlock(&mapping->i_mmap_mutex);
219 * Close a vm structure and free it, returning the next.
221 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
223 struct vm_area_struct *next = vma->vm_next;
226 if (vma->vm_ops && vma->vm_ops->close)
227 vma->vm_ops->close(vma);
230 mpol_put(vma_policy(vma));
231 kmem_cache_free(vm_area_cachep, vma);
235 static unsigned long do_brk(unsigned long addr, unsigned long len);
237 SYSCALL_DEFINE1(brk, unsigned long, brk)
239 unsigned long rlim, retval;
240 unsigned long newbrk, oldbrk;
241 struct mm_struct *mm = current->mm;
242 unsigned long min_brk;
244 down_write(&mm->mmap_sem);
246 #ifdef CONFIG_COMPAT_BRK
248 * CONFIG_COMPAT_BRK can still be overridden by setting
249 * randomize_va_space to 2, which will still cause mm->start_brk
250 * to be arbitrarily shifted
252 if (current->brk_randomized)
253 min_brk = mm->start_brk;
255 min_brk = mm->end_data;
257 min_brk = mm->start_brk;
263 * Check against rlimit here. If this check is done later after the test
264 * of oldbrk with newbrk then it can escape the test and let the data
265 * segment grow beyond its set limit the in case where the limit is
266 * not page aligned -Ram Gupta
268 rlim = rlimit(RLIMIT_DATA);
269 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
270 (mm->end_data - mm->start_data) > rlim)
273 newbrk = PAGE_ALIGN(brk);
274 oldbrk = PAGE_ALIGN(mm->brk);
275 if (oldbrk == newbrk)
278 /* Always allow shrinking brk. */
279 if (brk <= mm->brk) {
280 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
285 /* Check against existing mmap mappings. */
286 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
289 /* Ok, looks good - let it rip. */
290 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
296 up_write(&mm->mmap_sem);
300 #ifdef CONFIG_DEBUG_VM_RB
301 static int browse_rb(struct rb_root *root)
304 struct rb_node *nd, *pn = NULL;
305 unsigned long prev = 0, pend = 0;
307 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
308 struct vm_area_struct *vma;
309 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
310 if (vma->vm_start < prev)
311 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
312 if (vma->vm_start < pend)
313 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
314 if (vma->vm_start > vma->vm_end)
315 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
318 prev = vma->vm_start;
322 for (nd = pn; nd; nd = rb_prev(nd)) {
326 printk("backwards %d, forwards %d\n", j, i), i = 0;
330 void validate_mm(struct mm_struct *mm)
334 struct vm_area_struct *vma = mm->mmap;
336 struct anon_vma_chain *avc;
337 vma_lock_anon_vma(vma);
338 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
339 anon_vma_interval_tree_verify(avc);
340 vma_unlock_anon_vma(vma);
344 if (i != mm->map_count)
345 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
346 i = browse_rb(&mm->mm_rb);
347 if (i != mm->map_count)
348 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
352 #define validate_mm(mm) do { } while (0)
356 * vma has some anon_vma assigned, and is already inserted on that
357 * anon_vma's interval trees.
359 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
360 * vma must be removed from the anon_vma's interval trees using
361 * anon_vma_interval_tree_pre_update_vma().
363 * After the update, the vma will be reinserted using
364 * anon_vma_interval_tree_post_update_vma().
366 * The entire update must be protected by exclusive mmap_sem and by
367 * the root anon_vma's mutex.
370 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
372 struct anon_vma_chain *avc;
374 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
375 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
379 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
381 struct anon_vma_chain *avc;
383 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
384 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
387 static int find_vma_links(struct mm_struct *mm, unsigned long addr,
388 unsigned long end, struct vm_area_struct **pprev,
389 struct rb_node ***rb_link, struct rb_node **rb_parent)
391 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
393 __rb_link = &mm->mm_rb.rb_node;
394 rb_prev = __rb_parent = NULL;
397 struct vm_area_struct *vma_tmp;
399 __rb_parent = *__rb_link;
400 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
402 if (vma_tmp->vm_end > addr) {
403 /* Fail if an existing vma overlaps the area */
404 if (vma_tmp->vm_start < end)
406 __rb_link = &__rb_parent->rb_left;
408 rb_prev = __rb_parent;
409 __rb_link = &__rb_parent->rb_right;
415 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
416 *rb_link = __rb_link;
417 *rb_parent = __rb_parent;
421 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
422 struct rb_node **rb_link, struct rb_node *rb_parent)
424 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
425 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
428 static void __vma_link_file(struct vm_area_struct *vma)
434 struct address_space *mapping = file->f_mapping;
436 if (vma->vm_flags & VM_DENYWRITE)
437 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
438 if (vma->vm_flags & VM_SHARED)
439 mapping->i_mmap_writable++;
441 flush_dcache_mmap_lock(mapping);
442 if (unlikely(vma->vm_flags & VM_NONLINEAR))
443 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
445 vma_interval_tree_insert(vma, &mapping->i_mmap);
446 flush_dcache_mmap_unlock(mapping);
451 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
452 struct vm_area_struct *prev, struct rb_node **rb_link,
453 struct rb_node *rb_parent)
455 __vma_link_list(mm, vma, prev, rb_parent);
456 __vma_link_rb(mm, vma, rb_link, rb_parent);
459 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
460 struct vm_area_struct *prev, struct rb_node **rb_link,
461 struct rb_node *rb_parent)
463 struct address_space *mapping = NULL;
466 mapping = vma->vm_file->f_mapping;
469 mutex_lock(&mapping->i_mmap_mutex);
471 __vma_link(mm, vma, prev, rb_link, rb_parent);
472 __vma_link_file(vma);
475 mutex_unlock(&mapping->i_mmap_mutex);
482 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
483 * mm's list and rbtree. It has already been inserted into the interval tree.
485 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
487 struct vm_area_struct *prev;
488 struct rb_node **rb_link, *rb_parent;
490 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
491 &prev, &rb_link, &rb_parent))
493 __vma_link(mm, vma, prev, rb_link, rb_parent);
498 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
499 struct vm_area_struct *prev)
501 struct vm_area_struct *next = vma->vm_next;
503 prev->vm_next = next;
505 next->vm_prev = prev;
506 rb_erase(&vma->vm_rb, &mm->mm_rb);
507 if (mm->mmap_cache == vma)
508 mm->mmap_cache = prev;
512 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
513 * is already present in an i_mmap tree without adjusting the tree.
514 * The following helper function should be used when such adjustments
515 * are necessary. The "insert" vma (if any) is to be inserted
516 * before we drop the necessary locks.
518 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
519 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
521 struct mm_struct *mm = vma->vm_mm;
522 struct vm_area_struct *next = vma->vm_next;
523 struct vm_area_struct *importer = NULL;
524 struct address_space *mapping = NULL;
525 struct rb_root *root = NULL;
526 struct anon_vma *anon_vma = NULL;
527 struct file *file = vma->vm_file;
528 long adjust_next = 0;
531 if (next && !insert) {
532 struct vm_area_struct *exporter = NULL;
534 if (end >= next->vm_end) {
536 * vma expands, overlapping all the next, and
537 * perhaps the one after too (mprotect case 6).
539 again: remove_next = 1 + (end > next->vm_end);
543 } else if (end > next->vm_start) {
545 * vma expands, overlapping part of the next:
546 * mprotect case 5 shifting the boundary up.
548 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
551 } else if (end < vma->vm_end) {
553 * vma shrinks, and !insert tells it's not
554 * split_vma inserting another: so it must be
555 * mprotect case 4 shifting the boundary down.
557 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
563 * Easily overlooked: when mprotect shifts the boundary,
564 * make sure the expanding vma has anon_vma set if the
565 * shrinking vma had, to cover any anon pages imported.
567 if (exporter && exporter->anon_vma && !importer->anon_vma) {
568 if (anon_vma_clone(importer, exporter))
570 importer->anon_vma = exporter->anon_vma;
575 mapping = file->f_mapping;
576 if (!(vma->vm_flags & VM_NONLINEAR)) {
577 root = &mapping->i_mmap;
578 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
581 uprobe_munmap(next, next->vm_start,
585 mutex_lock(&mapping->i_mmap_mutex);
588 * Put into interval tree now, so instantiated pages
589 * are visible to arm/parisc __flush_dcache_page
590 * throughout; but we cannot insert into address
591 * space until vma start or end is updated.
593 __vma_link_file(insert);
597 vma_adjust_trans_huge(vma, start, end, adjust_next);
599 anon_vma = vma->anon_vma;
600 if (!anon_vma && adjust_next)
601 anon_vma = next->anon_vma;
603 VM_BUG_ON(adjust_next && next->anon_vma &&
604 anon_vma != next->anon_vma);
605 anon_vma_lock(anon_vma);
606 anon_vma_interval_tree_pre_update_vma(vma);
608 anon_vma_interval_tree_pre_update_vma(next);
612 flush_dcache_mmap_lock(mapping);
613 vma_interval_tree_remove(vma, root);
615 vma_interval_tree_remove(next, root);
618 vma->vm_start = start;
620 vma->vm_pgoff = pgoff;
622 next->vm_start += adjust_next << PAGE_SHIFT;
623 next->vm_pgoff += adjust_next;
628 vma_interval_tree_insert(next, root);
629 vma_interval_tree_insert(vma, root);
630 flush_dcache_mmap_unlock(mapping);
635 * vma_merge has merged next into vma, and needs
636 * us to remove next before dropping the locks.
638 __vma_unlink(mm, next, vma);
640 __remove_shared_vm_struct(next, file, mapping);
643 * split_vma has split insert from vma, and needs
644 * us to insert it before dropping the locks
645 * (it may either follow vma or precede it).
647 __insert_vm_struct(mm, insert);
651 anon_vma_interval_tree_post_update_vma(vma);
653 anon_vma_interval_tree_post_update_vma(next);
654 anon_vma_unlock(anon_vma);
657 mutex_unlock(&mapping->i_mmap_mutex);
668 uprobe_munmap(next, next->vm_start, next->vm_end);
672 anon_vma_merge(vma, next);
674 mpol_put(vma_policy(next));
675 kmem_cache_free(vm_area_cachep, next);
677 * In mprotect's case 6 (see comments on vma_merge),
678 * we must remove another next too. It would clutter
679 * up the code too much to do both in one go.
681 if (remove_next == 2) {
695 * If the vma has a ->close operation then the driver probably needs to release
696 * per-vma resources, so we don't attempt to merge those.
698 static inline int is_mergeable_vma(struct vm_area_struct *vma,
699 struct file *file, unsigned long vm_flags)
701 if (vma->vm_flags ^ vm_flags)
703 if (vma->vm_file != file)
705 if (vma->vm_ops && vma->vm_ops->close)
710 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
711 struct anon_vma *anon_vma2,
712 struct vm_area_struct *vma)
715 * The list_is_singular() test is to avoid merging VMA cloned from
716 * parents. This can improve scalability caused by anon_vma lock.
718 if ((!anon_vma1 || !anon_vma2) && (!vma ||
719 list_is_singular(&vma->anon_vma_chain)))
721 return anon_vma1 == anon_vma2;
725 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
726 * in front of (at a lower virtual address and file offset than) the vma.
728 * We cannot merge two vmas if they have differently assigned (non-NULL)
729 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
731 * We don't check here for the merged mmap wrapping around the end of pagecache
732 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
733 * wrap, nor mmaps which cover the final page at index -1UL.
736 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
737 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
739 if (is_mergeable_vma(vma, file, vm_flags) &&
740 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
741 if (vma->vm_pgoff == vm_pgoff)
748 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
749 * beyond (at a higher virtual address and file offset than) the vma.
751 * We cannot merge two vmas if they have differently assigned (non-NULL)
752 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
755 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
756 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
758 if (is_mergeable_vma(vma, file, vm_flags) &&
759 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
761 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
762 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
769 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
770 * whether that can be merged with its predecessor or its successor.
771 * Or both (it neatly fills a hole).
773 * In most cases - when called for mmap, brk or mremap - [addr,end) is
774 * certain not to be mapped by the time vma_merge is called; but when
775 * called for mprotect, it is certain to be already mapped (either at
776 * an offset within prev, or at the start of next), and the flags of
777 * this area are about to be changed to vm_flags - and the no-change
778 * case has already been eliminated.
780 * The following mprotect cases have to be considered, where AAAA is
781 * the area passed down from mprotect_fixup, never extending beyond one
782 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
784 * AAAA AAAA AAAA AAAA
785 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
786 * cannot merge might become might become might become
787 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
788 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
789 * mremap move: PPPPNNNNNNNN 8
791 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
792 * might become case 1 below case 2 below case 3 below
794 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
795 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
797 struct vm_area_struct *vma_merge(struct mm_struct *mm,
798 struct vm_area_struct *prev, unsigned long addr,
799 unsigned long end, unsigned long vm_flags,
800 struct anon_vma *anon_vma, struct file *file,
801 pgoff_t pgoff, struct mempolicy *policy)
803 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
804 struct vm_area_struct *area, *next;
808 * We later require that vma->vm_flags == vm_flags,
809 * so this tests vma->vm_flags & VM_SPECIAL, too.
811 if (vm_flags & VM_SPECIAL)
815 next = prev->vm_next;
819 if (next && next->vm_end == end) /* cases 6, 7, 8 */
820 next = next->vm_next;
823 * Can it merge with the predecessor?
825 if (prev && prev->vm_end == addr &&
826 mpol_equal(vma_policy(prev), policy) &&
827 can_vma_merge_after(prev, vm_flags,
828 anon_vma, file, pgoff)) {
830 * OK, it can. Can we now merge in the successor as well?
832 if (next && end == next->vm_start &&
833 mpol_equal(policy, vma_policy(next)) &&
834 can_vma_merge_before(next, vm_flags,
835 anon_vma, file, pgoff+pglen) &&
836 is_mergeable_anon_vma(prev->anon_vma,
837 next->anon_vma, NULL)) {
839 err = vma_adjust(prev, prev->vm_start,
840 next->vm_end, prev->vm_pgoff, NULL);
841 } else /* cases 2, 5, 7 */
842 err = vma_adjust(prev, prev->vm_start,
843 end, prev->vm_pgoff, NULL);
846 khugepaged_enter_vma_merge(prev);
851 * Can this new request be merged in front of next?
853 if (next && end == next->vm_start &&
854 mpol_equal(policy, vma_policy(next)) &&
855 can_vma_merge_before(next, vm_flags,
856 anon_vma, file, pgoff+pglen)) {
857 if (prev && addr < prev->vm_end) /* case 4 */
858 err = vma_adjust(prev, prev->vm_start,
859 addr, prev->vm_pgoff, NULL);
860 else /* cases 3, 8 */
861 err = vma_adjust(area, addr, next->vm_end,
862 next->vm_pgoff - pglen, NULL);
865 khugepaged_enter_vma_merge(area);
873 * Rough compatbility check to quickly see if it's even worth looking
874 * at sharing an anon_vma.
876 * They need to have the same vm_file, and the flags can only differ
877 * in things that mprotect may change.
879 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
880 * we can merge the two vma's. For example, we refuse to merge a vma if
881 * there is a vm_ops->close() function, because that indicates that the
882 * driver is doing some kind of reference counting. But that doesn't
883 * really matter for the anon_vma sharing case.
885 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
887 return a->vm_end == b->vm_start &&
888 mpol_equal(vma_policy(a), vma_policy(b)) &&
889 a->vm_file == b->vm_file &&
890 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
891 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
895 * Do some basic sanity checking to see if we can re-use the anon_vma
896 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
897 * the same as 'old', the other will be the new one that is trying
898 * to share the anon_vma.
900 * NOTE! This runs with mm_sem held for reading, so it is possible that
901 * the anon_vma of 'old' is concurrently in the process of being set up
902 * by another page fault trying to merge _that_. But that's ok: if it
903 * is being set up, that automatically means that it will be a singleton
904 * acceptable for merging, so we can do all of this optimistically. But
905 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
907 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
908 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
909 * is to return an anon_vma that is "complex" due to having gone through
912 * We also make sure that the two vma's are compatible (adjacent,
913 * and with the same memory policies). That's all stable, even with just
914 * a read lock on the mm_sem.
916 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
918 if (anon_vma_compatible(a, b)) {
919 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
921 if (anon_vma && list_is_singular(&old->anon_vma_chain))
928 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
929 * neighbouring vmas for a suitable anon_vma, before it goes off
930 * to allocate a new anon_vma. It checks because a repetitive
931 * sequence of mprotects and faults may otherwise lead to distinct
932 * anon_vmas being allocated, preventing vma merge in subsequent
935 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
937 struct anon_vma *anon_vma;
938 struct vm_area_struct *near;
944 anon_vma = reusable_anon_vma(near, vma, near);
952 anon_vma = reusable_anon_vma(near, near, vma);
957 * There's no absolute need to look only at touching neighbours:
958 * we could search further afield for "compatible" anon_vmas.
959 * But it would probably just be a waste of time searching,
960 * or lead to too many vmas hanging off the same anon_vma.
961 * We're trying to allow mprotect remerging later on,
962 * not trying to minimize memory used for anon_vmas.
967 #ifdef CONFIG_PROC_FS
968 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
969 struct file *file, long pages)
971 const unsigned long stack_flags
972 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
974 mm->total_vm += pages;
977 mm->shared_vm += pages;
978 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
979 mm->exec_vm += pages;
980 } else if (flags & stack_flags)
981 mm->stack_vm += pages;
983 #endif /* CONFIG_PROC_FS */
986 * If a hint addr is less than mmap_min_addr change hint to be as
987 * low as possible but still greater than mmap_min_addr
989 static inline unsigned long round_hint_to_min(unsigned long hint)
992 if (((void *)hint != NULL) &&
993 (hint < mmap_min_addr))
994 return PAGE_ALIGN(mmap_min_addr);
999 * The caller must hold down_write(¤t->mm->mmap_sem).
1002 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
1003 unsigned long len, unsigned long prot,
1004 unsigned long flags, unsigned long pgoff)
1006 struct mm_struct * mm = current->mm;
1007 struct inode *inode;
1008 vm_flags_t vm_flags;
1011 * Does the application expect PROT_READ to imply PROT_EXEC?
1013 * (the exception is when the underlying filesystem is noexec
1014 * mounted, in which case we dont add PROT_EXEC.)
1016 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1017 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
1023 if (!(flags & MAP_FIXED))
1024 addr = round_hint_to_min(addr);
1026 /* Careful about overflows.. */
1027 len = PAGE_ALIGN(len);
1031 /* offset overflow? */
1032 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1035 /* Too many mappings? */
1036 if (mm->map_count > sysctl_max_map_count)
1039 /* Obtain the address to map to. we verify (or select) it and ensure
1040 * that it represents a valid section of the address space.
1042 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1043 if (addr & ~PAGE_MASK)
1046 /* Do simple checking here so the lower-level routines won't have
1047 * to. we assume access permissions have been handled by the open
1048 * of the memory object, so we don't do any here.
1050 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1051 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1053 if (flags & MAP_LOCKED)
1054 if (!can_do_mlock())
1057 /* mlock MCL_FUTURE? */
1058 if (vm_flags & VM_LOCKED) {
1059 unsigned long locked, lock_limit;
1060 locked = len >> PAGE_SHIFT;
1061 locked += mm->locked_vm;
1062 lock_limit = rlimit(RLIMIT_MEMLOCK);
1063 lock_limit >>= PAGE_SHIFT;
1064 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1068 inode = file ? file->f_path.dentry->d_inode : NULL;
1071 switch (flags & MAP_TYPE) {
1073 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1077 * Make sure we don't allow writing to an append-only
1080 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1084 * Make sure there are no mandatory locks on the file.
1086 if (locks_verify_locked(inode))
1089 vm_flags |= VM_SHARED | VM_MAYSHARE;
1090 if (!(file->f_mode & FMODE_WRITE))
1091 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1095 if (!(file->f_mode & FMODE_READ))
1097 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1098 if (vm_flags & VM_EXEC)
1100 vm_flags &= ~VM_MAYEXEC;
1103 if (!file->f_op || !file->f_op->mmap)
1111 switch (flags & MAP_TYPE) {
1117 vm_flags |= VM_SHARED | VM_MAYSHARE;
1121 * Set pgoff according to addr for anon_vma.
1123 pgoff = addr >> PAGE_SHIFT;
1130 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1133 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1134 unsigned long, prot, unsigned long, flags,
1135 unsigned long, fd, unsigned long, pgoff)
1137 struct file *file = NULL;
1138 unsigned long retval = -EBADF;
1140 if (!(flags & MAP_ANONYMOUS)) {
1141 audit_mmap_fd(fd, flags);
1142 if (unlikely(flags & MAP_HUGETLB))
1147 } else if (flags & MAP_HUGETLB) {
1148 struct user_struct *user = NULL;
1150 * VM_NORESERVE is used because the reservations will be
1151 * taken when vm_ops->mmap() is called
1152 * A dummy user value is used because we are not locking
1153 * memory so no accounting is necessary
1155 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1157 &user, HUGETLB_ANONHUGE_INODE,
1158 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1160 return PTR_ERR(file);
1163 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1165 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1172 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1173 struct mmap_arg_struct {
1177 unsigned long flags;
1179 unsigned long offset;
1182 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1184 struct mmap_arg_struct a;
1186 if (copy_from_user(&a, arg, sizeof(a)))
1188 if (a.offset & ~PAGE_MASK)
1191 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1192 a.offset >> PAGE_SHIFT);
1194 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1197 * Some shared mappigns will want the pages marked read-only
1198 * to track write events. If so, we'll downgrade vm_page_prot
1199 * to the private version (using protection_map[] without the
1202 int vma_wants_writenotify(struct vm_area_struct *vma)
1204 vm_flags_t vm_flags = vma->vm_flags;
1206 /* If it was private or non-writable, the write bit is already clear */
1207 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1210 /* The backer wishes to know when pages are first written to? */
1211 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1214 /* The open routine did something to the protections already? */
1215 if (pgprot_val(vma->vm_page_prot) !=
1216 pgprot_val(vm_get_page_prot(vm_flags)))
1219 /* Specialty mapping? */
1220 if (vm_flags & VM_PFNMAP)
1223 /* Can the mapping track the dirty pages? */
1224 return vma->vm_file && vma->vm_file->f_mapping &&
1225 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1229 * We account for memory if it's a private writeable mapping,
1230 * not hugepages and VM_NORESERVE wasn't set.
1232 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1235 * hugetlb has its own accounting separate from the core VM
1236 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1238 if (file && is_file_hugepages(file))
1241 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1244 unsigned long mmap_region(struct file *file, unsigned long addr,
1245 unsigned long len, unsigned long flags,
1246 vm_flags_t vm_flags, unsigned long pgoff)
1248 struct mm_struct *mm = current->mm;
1249 struct vm_area_struct *vma, *prev;
1250 int correct_wcount = 0;
1252 struct rb_node **rb_link, *rb_parent;
1253 unsigned long charged = 0;
1254 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1256 /* Clear old maps */
1259 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
1260 if (do_munmap(mm, addr, len))
1265 /* Check against address space limit. */
1266 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1270 * Set 'VM_NORESERVE' if we should not account for the
1271 * memory use of this mapping.
1273 if ((flags & MAP_NORESERVE)) {
1274 /* We honor MAP_NORESERVE if allowed to overcommit */
1275 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1276 vm_flags |= VM_NORESERVE;
1278 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1279 if (file && is_file_hugepages(file))
1280 vm_flags |= VM_NORESERVE;
1284 * Private writable mapping: check memory availability
1286 if (accountable_mapping(file, vm_flags)) {
1287 charged = len >> PAGE_SHIFT;
1288 if (security_vm_enough_memory_mm(mm, charged))
1290 vm_flags |= VM_ACCOUNT;
1294 * Can we just expand an old mapping?
1296 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1301 * Determine the object being mapped and call the appropriate
1302 * specific mapper. the address has already been validated, but
1303 * not unmapped, but the maps are removed from the list.
1305 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1312 vma->vm_start = addr;
1313 vma->vm_end = addr + len;
1314 vma->vm_flags = vm_flags;
1315 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1316 vma->vm_pgoff = pgoff;
1317 INIT_LIST_HEAD(&vma->anon_vma_chain);
1319 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1322 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1324 if (vm_flags & VM_DENYWRITE) {
1325 error = deny_write_access(file);
1330 vma->vm_file = get_file(file);
1331 error = file->f_op->mmap(file, vma);
1333 goto unmap_and_free_vma;
1335 /* Can addr have changed??
1337 * Answer: Yes, several device drivers can do it in their
1338 * f_op->mmap method. -DaveM
1340 addr = vma->vm_start;
1341 pgoff = vma->vm_pgoff;
1342 vm_flags = vma->vm_flags;
1343 } else if (vm_flags & VM_SHARED) {
1344 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1346 error = shmem_zero_setup(vma);
1351 if (vma_wants_writenotify(vma)) {
1352 pgprot_t pprot = vma->vm_page_prot;
1354 /* Can vma->vm_page_prot have changed??
1356 * Answer: Yes, drivers may have changed it in their
1357 * f_op->mmap method.
1359 * Ensures that vmas marked as uncached stay that way.
1361 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1362 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1363 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1366 vma_link(mm, vma, prev, rb_link, rb_parent);
1367 file = vma->vm_file;
1369 /* Once vma denies write, undo our temporary denial count */
1371 atomic_inc(&inode->i_writecount);
1373 perf_event_mmap(vma);
1375 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1376 if (vm_flags & VM_LOCKED) {
1377 if (!mlock_vma_pages_range(vma, addr, addr + len))
1378 mm->locked_vm += (len >> PAGE_SHIFT);
1379 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1380 make_pages_present(addr, addr + len);
1389 atomic_inc(&inode->i_writecount);
1390 vma->vm_file = NULL;
1393 /* Undo any partial mapping done by a device driver. */
1394 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1397 kmem_cache_free(vm_area_cachep, vma);
1400 vm_unacct_memory(charged);
1404 /* Get an address range which is currently unmapped.
1405 * For shmat() with addr=0.
1407 * Ugly calling convention alert:
1408 * Return value with the low bits set means error value,
1410 * if (ret & ~PAGE_MASK)
1413 * This function "knows" that -ENOMEM has the bits set.
1415 #ifndef HAVE_ARCH_UNMAPPED_AREA
1417 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1418 unsigned long len, unsigned long pgoff, unsigned long flags)
1420 struct mm_struct *mm = current->mm;
1421 struct vm_area_struct *vma;
1422 unsigned long start_addr;
1424 if (len > TASK_SIZE)
1427 if (flags & MAP_FIXED)
1431 addr = PAGE_ALIGN(addr);
1432 vma = find_vma(mm, addr);
1433 if (TASK_SIZE - len >= addr &&
1434 (!vma || addr + len <= vma->vm_start))
1437 if (len > mm->cached_hole_size) {
1438 start_addr = addr = mm->free_area_cache;
1440 start_addr = addr = TASK_UNMAPPED_BASE;
1441 mm->cached_hole_size = 0;
1445 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1446 /* At this point: (!vma || addr < vma->vm_end). */
1447 if (TASK_SIZE - len < addr) {
1449 * Start a new search - just in case we missed
1452 if (start_addr != TASK_UNMAPPED_BASE) {
1453 addr = TASK_UNMAPPED_BASE;
1455 mm->cached_hole_size = 0;
1460 if (!vma || addr + len <= vma->vm_start) {
1462 * Remember the place where we stopped the search:
1464 mm->free_area_cache = addr + len;
1467 if (addr + mm->cached_hole_size < vma->vm_start)
1468 mm->cached_hole_size = vma->vm_start - addr;
1474 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1477 * Is this a new hole at the lowest possible address?
1479 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
1480 mm->free_area_cache = addr;
1484 * This mmap-allocator allocates new areas top-down from below the
1485 * stack's low limit (the base):
1487 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1489 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1490 const unsigned long len, const unsigned long pgoff,
1491 const unsigned long flags)
1493 struct vm_area_struct *vma;
1494 struct mm_struct *mm = current->mm;
1495 unsigned long addr = addr0, start_addr;
1497 /* requested length too big for entire address space */
1498 if (len > TASK_SIZE)
1501 if (flags & MAP_FIXED)
1504 /* requesting a specific address */
1506 addr = PAGE_ALIGN(addr);
1507 vma = find_vma(mm, addr);
1508 if (TASK_SIZE - len >= addr &&
1509 (!vma || addr + len <= vma->vm_start))
1513 /* check if free_area_cache is useful for us */
1514 if (len <= mm->cached_hole_size) {
1515 mm->cached_hole_size = 0;
1516 mm->free_area_cache = mm->mmap_base;
1520 /* either no address requested or can't fit in requested address hole */
1521 start_addr = addr = mm->free_area_cache;
1529 * Lookup failure means no vma is above this address,
1530 * else if new region fits below vma->vm_start,
1531 * return with success:
1533 vma = find_vma(mm, addr);
1534 if (!vma || addr+len <= vma->vm_start)
1535 /* remember the address as a hint for next time */
1536 return (mm->free_area_cache = addr);
1538 /* remember the largest hole we saw so far */
1539 if (addr + mm->cached_hole_size < vma->vm_start)
1540 mm->cached_hole_size = vma->vm_start - addr;
1542 /* try just below the current vma->vm_start */
1543 addr = vma->vm_start-len;
1544 } while (len < vma->vm_start);
1548 * if hint left us with no space for the requested
1549 * mapping then try again:
1551 * Note: this is different with the case of bottomup
1552 * which does the fully line-search, but we use find_vma
1553 * here that causes some holes skipped.
1555 if (start_addr != mm->mmap_base) {
1556 mm->free_area_cache = mm->mmap_base;
1557 mm->cached_hole_size = 0;
1562 * A failed mmap() very likely causes application failure,
1563 * so fall back to the bottom-up function here. This scenario
1564 * can happen with large stack limits and large mmap()
1567 mm->cached_hole_size = ~0UL;
1568 mm->free_area_cache = TASK_UNMAPPED_BASE;
1569 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1571 * Restore the topdown base:
1573 mm->free_area_cache = mm->mmap_base;
1574 mm->cached_hole_size = ~0UL;
1580 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1583 * Is this a new hole at the highest possible address?
1585 if (addr > mm->free_area_cache)
1586 mm->free_area_cache = addr;
1588 /* dont allow allocations above current base */
1589 if (mm->free_area_cache > mm->mmap_base)
1590 mm->free_area_cache = mm->mmap_base;
1594 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1595 unsigned long pgoff, unsigned long flags)
1597 unsigned long (*get_area)(struct file *, unsigned long,
1598 unsigned long, unsigned long, unsigned long);
1600 unsigned long error = arch_mmap_check(addr, len, flags);
1604 /* Careful about overflows.. */
1605 if (len > TASK_SIZE)
1608 get_area = current->mm->get_unmapped_area;
1609 if (file && file->f_op && file->f_op->get_unmapped_area)
1610 get_area = file->f_op->get_unmapped_area;
1611 addr = get_area(file, addr, len, pgoff, flags);
1612 if (IS_ERR_VALUE(addr))
1615 if (addr > TASK_SIZE - len)
1617 if (addr & ~PAGE_MASK)
1620 addr = arch_rebalance_pgtables(addr, len);
1621 error = security_mmap_addr(addr);
1622 return error ? error : addr;
1625 EXPORT_SYMBOL(get_unmapped_area);
1627 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1628 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1630 struct vm_area_struct *vma = NULL;
1632 if (WARN_ON_ONCE(!mm)) /* Remove this in linux-3.6 */
1635 /* Check the cache first. */
1636 /* (Cache hit rate is typically around 35%.) */
1637 vma = mm->mmap_cache;
1638 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1639 struct rb_node *rb_node;
1641 rb_node = mm->mm_rb.rb_node;
1645 struct vm_area_struct *vma_tmp;
1647 vma_tmp = rb_entry(rb_node,
1648 struct vm_area_struct, vm_rb);
1650 if (vma_tmp->vm_end > addr) {
1652 if (vma_tmp->vm_start <= addr)
1654 rb_node = rb_node->rb_left;
1656 rb_node = rb_node->rb_right;
1659 mm->mmap_cache = vma;
1664 EXPORT_SYMBOL(find_vma);
1667 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1669 struct vm_area_struct *
1670 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1671 struct vm_area_struct **pprev)
1673 struct vm_area_struct *vma;
1675 vma = find_vma(mm, addr);
1677 *pprev = vma->vm_prev;
1679 struct rb_node *rb_node = mm->mm_rb.rb_node;
1682 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1683 rb_node = rb_node->rb_right;
1690 * Verify that the stack growth is acceptable and
1691 * update accounting. This is shared with both the
1692 * grow-up and grow-down cases.
1694 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1696 struct mm_struct *mm = vma->vm_mm;
1697 struct rlimit *rlim = current->signal->rlim;
1698 unsigned long new_start;
1700 /* address space limit tests */
1701 if (!may_expand_vm(mm, grow))
1704 /* Stack limit test */
1705 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1708 /* mlock limit tests */
1709 if (vma->vm_flags & VM_LOCKED) {
1710 unsigned long locked;
1711 unsigned long limit;
1712 locked = mm->locked_vm + grow;
1713 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1714 limit >>= PAGE_SHIFT;
1715 if (locked > limit && !capable(CAP_IPC_LOCK))
1719 /* Check to ensure the stack will not grow into a hugetlb-only region */
1720 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1722 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1726 * Overcommit.. This must be the final test, as it will
1727 * update security statistics.
1729 if (security_vm_enough_memory_mm(mm, grow))
1732 /* Ok, everything looks good - let it rip */
1733 if (vma->vm_flags & VM_LOCKED)
1734 mm->locked_vm += grow;
1735 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1739 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1741 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1742 * vma is the last one with address > vma->vm_end. Have to extend vma.
1744 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1748 if (!(vma->vm_flags & VM_GROWSUP))
1752 * We must make sure the anon_vma is allocated
1753 * so that the anon_vma locking is not a noop.
1755 if (unlikely(anon_vma_prepare(vma)))
1757 vma_lock_anon_vma(vma);
1760 * vma->vm_start/vm_end cannot change under us because the caller
1761 * is required to hold the mmap_sem in read mode. We need the
1762 * anon_vma lock to serialize against concurrent expand_stacks.
1763 * Also guard against wrapping around to address 0.
1765 if (address < PAGE_ALIGN(address+4))
1766 address = PAGE_ALIGN(address+4);
1768 vma_unlock_anon_vma(vma);
1773 /* Somebody else might have raced and expanded it already */
1774 if (address > vma->vm_end) {
1775 unsigned long size, grow;
1777 size = address - vma->vm_start;
1778 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1781 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1782 error = acct_stack_growth(vma, size, grow);
1784 anon_vma_interval_tree_pre_update_vma(vma);
1785 vma->vm_end = address;
1786 anon_vma_interval_tree_post_update_vma(vma);
1787 perf_event_mmap(vma);
1791 vma_unlock_anon_vma(vma);
1792 khugepaged_enter_vma_merge(vma);
1793 validate_mm(vma->vm_mm);
1796 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1799 * vma is the first one with address < vma->vm_start. Have to extend vma.
1801 int expand_downwards(struct vm_area_struct *vma,
1802 unsigned long address)
1807 * We must make sure the anon_vma is allocated
1808 * so that the anon_vma locking is not a noop.
1810 if (unlikely(anon_vma_prepare(vma)))
1813 address &= PAGE_MASK;
1814 error = security_mmap_addr(address);
1818 vma_lock_anon_vma(vma);
1821 * vma->vm_start/vm_end cannot change under us because the caller
1822 * is required to hold the mmap_sem in read mode. We need the
1823 * anon_vma lock to serialize against concurrent expand_stacks.
1826 /* Somebody else might have raced and expanded it already */
1827 if (address < vma->vm_start) {
1828 unsigned long size, grow;
1830 size = vma->vm_end - address;
1831 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1834 if (grow <= vma->vm_pgoff) {
1835 error = acct_stack_growth(vma, size, grow);
1837 anon_vma_interval_tree_pre_update_vma(vma);
1838 vma->vm_start = address;
1839 vma->vm_pgoff -= grow;
1840 anon_vma_interval_tree_post_update_vma(vma);
1841 perf_event_mmap(vma);
1845 vma_unlock_anon_vma(vma);
1846 khugepaged_enter_vma_merge(vma);
1847 validate_mm(vma->vm_mm);
1851 #ifdef CONFIG_STACK_GROWSUP
1852 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1854 return expand_upwards(vma, address);
1857 struct vm_area_struct *
1858 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1860 struct vm_area_struct *vma, *prev;
1863 vma = find_vma_prev(mm, addr, &prev);
1864 if (vma && (vma->vm_start <= addr))
1866 if (!prev || expand_stack(prev, addr))
1868 if (prev->vm_flags & VM_LOCKED) {
1869 mlock_vma_pages_range(prev, addr, prev->vm_end);
1874 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1876 return expand_downwards(vma, address);
1879 struct vm_area_struct *
1880 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1882 struct vm_area_struct * vma;
1883 unsigned long start;
1886 vma = find_vma(mm,addr);
1889 if (vma->vm_start <= addr)
1891 if (!(vma->vm_flags & VM_GROWSDOWN))
1893 start = vma->vm_start;
1894 if (expand_stack(vma, addr))
1896 if (vma->vm_flags & VM_LOCKED) {
1897 mlock_vma_pages_range(vma, addr, start);
1904 * Ok - we have the memory areas we should free on the vma list,
1905 * so release them, and do the vma updates.
1907 * Called with the mm semaphore held.
1909 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1911 unsigned long nr_accounted = 0;
1913 /* Update high watermark before we lower total_vm */
1914 update_hiwater_vm(mm);
1916 long nrpages = vma_pages(vma);
1918 if (vma->vm_flags & VM_ACCOUNT)
1919 nr_accounted += nrpages;
1920 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1921 vma = remove_vma(vma);
1923 vm_unacct_memory(nr_accounted);
1928 * Get rid of page table information in the indicated region.
1930 * Called with the mm semaphore held.
1932 static void unmap_region(struct mm_struct *mm,
1933 struct vm_area_struct *vma, struct vm_area_struct *prev,
1934 unsigned long start, unsigned long end)
1936 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1937 struct mmu_gather tlb;
1940 tlb_gather_mmu(&tlb, mm, 0);
1941 update_hiwater_rss(mm);
1942 unmap_vmas(&tlb, vma, start, end);
1943 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1944 next ? next->vm_start : 0);
1945 tlb_finish_mmu(&tlb, start, end);
1949 * Create a list of vma's touched by the unmap, removing them from the mm's
1950 * vma list as we go..
1953 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1954 struct vm_area_struct *prev, unsigned long end)
1956 struct vm_area_struct **insertion_point;
1957 struct vm_area_struct *tail_vma = NULL;
1960 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1961 vma->vm_prev = NULL;
1963 rb_erase(&vma->vm_rb, &mm->mm_rb);
1967 } while (vma && vma->vm_start < end);
1968 *insertion_point = vma;
1970 vma->vm_prev = prev;
1971 tail_vma->vm_next = NULL;
1972 if (mm->unmap_area == arch_unmap_area)
1973 addr = prev ? prev->vm_end : mm->mmap_base;
1975 addr = vma ? vma->vm_start : mm->mmap_base;
1976 mm->unmap_area(mm, addr);
1977 mm->mmap_cache = NULL; /* Kill the cache. */
1981 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1982 * munmap path where it doesn't make sense to fail.
1984 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1985 unsigned long addr, int new_below)
1987 struct mempolicy *pol;
1988 struct vm_area_struct *new;
1991 if (is_vm_hugetlb_page(vma) && (addr &
1992 ~(huge_page_mask(hstate_vma(vma)))))
1995 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1999 /* most fields are the same, copy all, and then fixup */
2002 INIT_LIST_HEAD(&new->anon_vma_chain);
2007 new->vm_start = addr;
2008 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2011 pol = mpol_dup(vma_policy(vma));
2016 vma_set_policy(new, pol);
2018 if (anon_vma_clone(new, vma))
2022 get_file(new->vm_file);
2024 if (new->vm_ops && new->vm_ops->open)
2025 new->vm_ops->open(new);
2028 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2029 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2031 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2037 /* Clean everything up if vma_adjust failed. */
2038 if (new->vm_ops && new->vm_ops->close)
2039 new->vm_ops->close(new);
2042 unlink_anon_vmas(new);
2046 kmem_cache_free(vm_area_cachep, new);
2052 * Split a vma into two pieces at address 'addr', a new vma is allocated
2053 * either for the first part or the tail.
2055 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2056 unsigned long addr, int new_below)
2058 if (mm->map_count >= sysctl_max_map_count)
2061 return __split_vma(mm, vma, addr, new_below);
2064 /* Munmap is split into 2 main parts -- this part which finds
2065 * what needs doing, and the areas themselves, which do the
2066 * work. This now handles partial unmappings.
2067 * Jeremy Fitzhardinge <jeremy@goop.org>
2069 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2072 struct vm_area_struct *vma, *prev, *last;
2074 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2077 if ((len = PAGE_ALIGN(len)) == 0)
2080 /* Find the first overlapping VMA */
2081 vma = find_vma(mm, start);
2084 prev = vma->vm_prev;
2085 /* we have start < vma->vm_end */
2087 /* if it doesn't overlap, we have nothing.. */
2089 if (vma->vm_start >= end)
2093 * If we need to split any vma, do it now to save pain later.
2095 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2096 * unmapped vm_area_struct will remain in use: so lower split_vma
2097 * places tmp vma above, and higher split_vma places tmp vma below.
2099 if (start > vma->vm_start) {
2103 * Make sure that map_count on return from munmap() will
2104 * not exceed its limit; but let map_count go just above
2105 * its limit temporarily, to help free resources as expected.
2107 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2110 error = __split_vma(mm, vma, start, 0);
2116 /* Does it split the last one? */
2117 last = find_vma(mm, end);
2118 if (last && end > last->vm_start) {
2119 int error = __split_vma(mm, last, end, 1);
2123 vma = prev? prev->vm_next: mm->mmap;
2126 * unlock any mlock()ed ranges before detaching vmas
2128 if (mm->locked_vm) {
2129 struct vm_area_struct *tmp = vma;
2130 while (tmp && tmp->vm_start < end) {
2131 if (tmp->vm_flags & VM_LOCKED) {
2132 mm->locked_vm -= vma_pages(tmp);
2133 munlock_vma_pages_all(tmp);
2140 * Remove the vma's, and unmap the actual pages
2142 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2143 unmap_region(mm, vma, prev, start, end);
2145 /* Fix up all other VM information */
2146 remove_vma_list(mm, vma);
2151 int vm_munmap(unsigned long start, size_t len)
2154 struct mm_struct *mm = current->mm;
2156 down_write(&mm->mmap_sem);
2157 ret = do_munmap(mm, start, len);
2158 up_write(&mm->mmap_sem);
2161 EXPORT_SYMBOL(vm_munmap);
2163 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2165 profile_munmap(addr);
2166 return vm_munmap(addr, len);
2169 static inline void verify_mm_writelocked(struct mm_struct *mm)
2171 #ifdef CONFIG_DEBUG_VM
2172 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2174 up_read(&mm->mmap_sem);
2180 * this is really a simplified "do_mmap". it only handles
2181 * anonymous maps. eventually we may be able to do some
2182 * brk-specific accounting here.
2184 static unsigned long do_brk(unsigned long addr, unsigned long len)
2186 struct mm_struct * mm = current->mm;
2187 struct vm_area_struct * vma, * prev;
2188 unsigned long flags;
2189 struct rb_node ** rb_link, * rb_parent;
2190 pgoff_t pgoff = addr >> PAGE_SHIFT;
2193 len = PAGE_ALIGN(len);
2197 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2199 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2200 if (error & ~PAGE_MASK)
2206 if (mm->def_flags & VM_LOCKED) {
2207 unsigned long locked, lock_limit;
2208 locked = len >> PAGE_SHIFT;
2209 locked += mm->locked_vm;
2210 lock_limit = rlimit(RLIMIT_MEMLOCK);
2211 lock_limit >>= PAGE_SHIFT;
2212 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2217 * mm->mmap_sem is required to protect against another thread
2218 * changing the mappings in case we sleep.
2220 verify_mm_writelocked(mm);
2223 * Clear old maps. this also does some error checking for us
2226 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
2227 if (do_munmap(mm, addr, len))
2232 /* Check against address space limits *after* clearing old maps... */
2233 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2236 if (mm->map_count > sysctl_max_map_count)
2239 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2242 /* Can we just expand an old private anonymous mapping? */
2243 vma = vma_merge(mm, prev, addr, addr + len, flags,
2244 NULL, NULL, pgoff, NULL);
2249 * create a vma struct for an anonymous mapping
2251 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2253 vm_unacct_memory(len >> PAGE_SHIFT);
2257 INIT_LIST_HEAD(&vma->anon_vma_chain);
2259 vma->vm_start = addr;
2260 vma->vm_end = addr + len;
2261 vma->vm_pgoff = pgoff;
2262 vma->vm_flags = flags;
2263 vma->vm_page_prot = vm_get_page_prot(flags);
2264 vma_link(mm, vma, prev, rb_link, rb_parent);
2266 perf_event_mmap(vma);
2267 mm->total_vm += len >> PAGE_SHIFT;
2268 if (flags & VM_LOCKED) {
2269 if (!mlock_vma_pages_range(vma, addr, addr + len))
2270 mm->locked_vm += (len >> PAGE_SHIFT);
2275 unsigned long vm_brk(unsigned long addr, unsigned long len)
2277 struct mm_struct *mm = current->mm;
2280 down_write(&mm->mmap_sem);
2281 ret = do_brk(addr, len);
2282 up_write(&mm->mmap_sem);
2285 EXPORT_SYMBOL(vm_brk);
2287 /* Release all mmaps. */
2288 void exit_mmap(struct mm_struct *mm)
2290 struct mmu_gather tlb;
2291 struct vm_area_struct *vma;
2292 unsigned long nr_accounted = 0;
2294 /* mm's last user has gone, and its about to be pulled down */
2295 mmu_notifier_release(mm);
2297 if (mm->locked_vm) {
2300 if (vma->vm_flags & VM_LOCKED)
2301 munlock_vma_pages_all(vma);
2309 if (!vma) /* Can happen if dup_mmap() received an OOM */
2314 tlb_gather_mmu(&tlb, mm, 1);
2315 /* update_hiwater_rss(mm) here? but nobody should be looking */
2316 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2317 unmap_vmas(&tlb, vma, 0, -1);
2319 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2320 tlb_finish_mmu(&tlb, 0, -1);
2323 * Walk the list again, actually closing and freeing it,
2324 * with preemption enabled, without holding any MM locks.
2327 if (vma->vm_flags & VM_ACCOUNT)
2328 nr_accounted += vma_pages(vma);
2329 vma = remove_vma(vma);
2331 vm_unacct_memory(nr_accounted);
2333 WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2336 /* Insert vm structure into process list sorted by address
2337 * and into the inode's i_mmap tree. If vm_file is non-NULL
2338 * then i_mmap_mutex is taken here.
2340 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
2342 struct vm_area_struct *prev;
2343 struct rb_node **rb_link, *rb_parent;
2346 * The vm_pgoff of a purely anonymous vma should be irrelevant
2347 * until its first write fault, when page's anon_vma and index
2348 * are set. But now set the vm_pgoff it will almost certainly
2349 * end up with (unless mremap moves it elsewhere before that
2350 * first wfault), so /proc/pid/maps tells a consistent story.
2352 * By setting it to reflect the virtual start address of the
2353 * vma, merges and splits can happen in a seamless way, just
2354 * using the existing file pgoff checks and manipulations.
2355 * Similarly in do_mmap_pgoff and in do_brk.
2357 if (!vma->vm_file) {
2358 BUG_ON(vma->anon_vma);
2359 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2361 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2362 &prev, &rb_link, &rb_parent))
2364 if ((vma->vm_flags & VM_ACCOUNT) &&
2365 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2368 vma_link(mm, vma, prev, rb_link, rb_parent);
2373 * Copy the vma structure to a new location in the same mm,
2374 * prior to moving page table entries, to effect an mremap move.
2376 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2377 unsigned long addr, unsigned long len, pgoff_t pgoff,
2378 bool *need_rmap_locks)
2380 struct vm_area_struct *vma = *vmap;
2381 unsigned long vma_start = vma->vm_start;
2382 struct mm_struct *mm = vma->vm_mm;
2383 struct vm_area_struct *new_vma, *prev;
2384 struct rb_node **rb_link, *rb_parent;
2385 struct mempolicy *pol;
2386 bool faulted_in_anon_vma = true;
2389 * If anonymous vma has not yet been faulted, update new pgoff
2390 * to match new location, to increase its chance of merging.
2392 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2393 pgoff = addr >> PAGE_SHIFT;
2394 faulted_in_anon_vma = false;
2397 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2398 return NULL; /* should never get here */
2399 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2400 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2403 * Source vma may have been merged into new_vma
2405 if (unlikely(vma_start >= new_vma->vm_start &&
2406 vma_start < new_vma->vm_end)) {
2408 * The only way we can get a vma_merge with
2409 * self during an mremap is if the vma hasn't
2410 * been faulted in yet and we were allowed to
2411 * reset the dst vma->vm_pgoff to the
2412 * destination address of the mremap to allow
2413 * the merge to happen. mremap must change the
2414 * vm_pgoff linearity between src and dst vmas
2415 * (in turn preventing a vma_merge) to be
2416 * safe. It is only safe to keep the vm_pgoff
2417 * linear if there are no pages mapped yet.
2419 VM_BUG_ON(faulted_in_anon_vma);
2420 *vmap = vma = new_vma;
2422 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
2424 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2427 new_vma->vm_start = addr;
2428 new_vma->vm_end = addr + len;
2429 new_vma->vm_pgoff = pgoff;
2430 pol = mpol_dup(vma_policy(vma));
2433 vma_set_policy(new_vma, pol);
2434 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2435 if (anon_vma_clone(new_vma, vma))
2436 goto out_free_mempol;
2437 if (new_vma->vm_file)
2438 get_file(new_vma->vm_file);
2439 if (new_vma->vm_ops && new_vma->vm_ops->open)
2440 new_vma->vm_ops->open(new_vma);
2441 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2442 *need_rmap_locks = false;
2450 kmem_cache_free(vm_area_cachep, new_vma);
2455 * Return true if the calling process may expand its vm space by the passed
2458 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2460 unsigned long cur = mm->total_vm; /* pages */
2463 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2465 if (cur + npages > lim)
2471 static int special_mapping_fault(struct vm_area_struct *vma,
2472 struct vm_fault *vmf)
2475 struct page **pages;
2478 * special mappings have no vm_file, and in that case, the mm
2479 * uses vm_pgoff internally. So we have to subtract it from here.
2480 * We are allowed to do this because we are the mm; do not copy
2481 * this code into drivers!
2483 pgoff = vmf->pgoff - vma->vm_pgoff;
2485 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2489 struct page *page = *pages;
2495 return VM_FAULT_SIGBUS;
2499 * Having a close hook prevents vma merging regardless of flags.
2501 static void special_mapping_close(struct vm_area_struct *vma)
2505 static const struct vm_operations_struct special_mapping_vmops = {
2506 .close = special_mapping_close,
2507 .fault = special_mapping_fault,
2511 * Called with mm->mmap_sem held for writing.
2512 * Insert a new vma covering the given region, with the given flags.
2513 * Its pages are supplied by the given array of struct page *.
2514 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2515 * The region past the last page supplied will always produce SIGBUS.
2516 * The array pointer and the pages it points to are assumed to stay alive
2517 * for as long as this mapping might exist.
2519 int install_special_mapping(struct mm_struct *mm,
2520 unsigned long addr, unsigned long len,
2521 unsigned long vm_flags, struct page **pages)
2524 struct vm_area_struct *vma;
2526 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2527 if (unlikely(vma == NULL))
2530 INIT_LIST_HEAD(&vma->anon_vma_chain);
2532 vma->vm_start = addr;
2533 vma->vm_end = addr + len;
2535 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2536 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2538 vma->vm_ops = &special_mapping_vmops;
2539 vma->vm_private_data = pages;
2541 ret = insert_vm_struct(mm, vma);
2545 mm->total_vm += len >> PAGE_SHIFT;
2547 perf_event_mmap(vma);
2552 kmem_cache_free(vm_area_cachep, vma);
2556 static DEFINE_MUTEX(mm_all_locks_mutex);
2558 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2560 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
2562 * The LSB of head.next can't change from under us
2563 * because we hold the mm_all_locks_mutex.
2565 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2567 * We can safely modify head.next after taking the
2568 * anon_vma->root->mutex. If some other vma in this mm shares
2569 * the same anon_vma we won't take it again.
2571 * No need of atomic instructions here, head.next
2572 * can't change from under us thanks to the
2573 * anon_vma->root->mutex.
2575 if (__test_and_set_bit(0, (unsigned long *)
2576 &anon_vma->root->rb_root.rb_node))
2581 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2583 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2585 * AS_MM_ALL_LOCKS can't change from under us because
2586 * we hold the mm_all_locks_mutex.
2588 * Operations on ->flags have to be atomic because
2589 * even if AS_MM_ALL_LOCKS is stable thanks to the
2590 * mm_all_locks_mutex, there may be other cpus
2591 * changing other bitflags in parallel to us.
2593 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2595 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2600 * This operation locks against the VM for all pte/vma/mm related
2601 * operations that could ever happen on a certain mm. This includes
2602 * vmtruncate, try_to_unmap, and all page faults.
2604 * The caller must take the mmap_sem in write mode before calling
2605 * mm_take_all_locks(). The caller isn't allowed to release the
2606 * mmap_sem until mm_drop_all_locks() returns.
2608 * mmap_sem in write mode is required in order to block all operations
2609 * that could modify pagetables and free pages without need of
2610 * altering the vma layout (for example populate_range() with
2611 * nonlinear vmas). It's also needed in write mode to avoid new
2612 * anon_vmas to be associated with existing vmas.
2614 * A single task can't take more than one mm_take_all_locks() in a row
2615 * or it would deadlock.
2617 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
2618 * mapping->flags avoid to take the same lock twice, if more than one
2619 * vma in this mm is backed by the same anon_vma or address_space.
2621 * We can take all the locks in random order because the VM code
2622 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2623 * takes more than one of them in a row. Secondly we're protected
2624 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2626 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2627 * that may have to take thousand of locks.
2629 * mm_take_all_locks() can fail if it's interrupted by signals.
2631 int mm_take_all_locks(struct mm_struct *mm)
2633 struct vm_area_struct *vma;
2634 struct anon_vma_chain *avc;
2636 BUG_ON(down_read_trylock(&mm->mmap_sem));
2638 mutex_lock(&mm_all_locks_mutex);
2640 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2641 if (signal_pending(current))
2643 if (vma->vm_file && vma->vm_file->f_mapping)
2644 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2647 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2648 if (signal_pending(current))
2651 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2652 vm_lock_anon_vma(mm, avc->anon_vma);
2658 mm_drop_all_locks(mm);
2662 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2664 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
2666 * The LSB of head.next can't change to 0 from under
2667 * us because we hold the mm_all_locks_mutex.
2669 * We must however clear the bitflag before unlocking
2670 * the vma so the users using the anon_vma->rb_root will
2671 * never see our bitflag.
2673 * No need of atomic instructions here, head.next
2674 * can't change from under us until we release the
2675 * anon_vma->root->mutex.
2677 if (!__test_and_clear_bit(0, (unsigned long *)
2678 &anon_vma->root->rb_root.rb_node))
2680 anon_vma_unlock(anon_vma);
2684 static void vm_unlock_mapping(struct address_space *mapping)
2686 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2688 * AS_MM_ALL_LOCKS can't change to 0 from under us
2689 * because we hold the mm_all_locks_mutex.
2691 mutex_unlock(&mapping->i_mmap_mutex);
2692 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2699 * The mmap_sem cannot be released by the caller until
2700 * mm_drop_all_locks() returns.
2702 void mm_drop_all_locks(struct mm_struct *mm)
2704 struct vm_area_struct *vma;
2705 struct anon_vma_chain *avc;
2707 BUG_ON(down_read_trylock(&mm->mmap_sem));
2708 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2710 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2712 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2713 vm_unlock_anon_vma(avc->anon_vma);
2714 if (vma->vm_file && vma->vm_file->f_mapping)
2715 vm_unlock_mapping(vma->vm_file->f_mapping);
2718 mutex_unlock(&mm_all_locks_mutex);
2722 * initialise the VMA slab
2724 void __init mmap_init(void)
2728 ret = percpu_counter_init(&vm_committed_as, 0);