6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/backing-dev.h>
13 #include <linux/shm.h>
14 #include <linux/mman.h>
15 #include <linux/pagemap.h>
16 #include <linux/swap.h>
17 #include <linux/syscalls.h>
18 #include <linux/capability.h>
19 #include <linux/init.h>
20 #include <linux/file.h>
22 #include <linux/personality.h>
23 #include <linux/security.h>
24 #include <linux/hugetlb.h>
25 #include <linux/profile.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/mempolicy.h>
29 #include <linux/rmap.h>
30 #include <linux/mmu_notifier.h>
31 #include <linux/perf_event.h>
32 #include <linux/audit.h>
33 #include <linux/khugepaged.h>
34 #include <linux/uprobes.h>
35 #include <linux/rbtree_augmented.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/notifier.h>
38 #include <linux/memory.h>
40 #include <asm/uaccess.h>
41 #include <asm/cacheflush.h>
43 #include <asm/mmu_context.h>
46 #ifdef CONFIG_SDCARD_FS
47 #include "../fs/sdcardfs/sdcardfs.h"
50 #ifndef arch_mmap_check
51 #define arch_mmap_check(addr, len, flags) (0)
54 #ifndef arch_rebalance_pgtables
55 #define arch_rebalance_pgtables(addr, len) (addr)
58 static void unmap_region(struct mm_struct *mm,
59 struct vm_area_struct *vma, struct vm_area_struct *prev,
60 unsigned long start, unsigned long end);
62 /* description of effects of mapping type and prot in current implementation.
63 * this is due to the limited x86 page protection hardware. The expected
64 * behavior is in parens:
67 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
68 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
69 * w: (no) no w: (no) no w: (yes) yes w: (no) no
70 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
73 * w: (no) no w: (no) no w: (copy) copy w: (no) no
74 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
77 pgprot_t protection_map[16] = {
78 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
79 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
82 pgprot_t vm_get_page_prot(unsigned long vm_flags)
84 return __pgprot(pgprot_val(protection_map[vm_flags &
85 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
86 pgprot_val(arch_vm_get_page_prot(vm_flags)));
88 EXPORT_SYMBOL(vm_get_page_prot);
90 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
91 int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
92 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
93 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
94 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
96 * Make sure vm_committed_as in one cacheline and not cacheline shared with
97 * other variables. It can be updated by several CPUs frequently.
99 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
102 * The global memory commitment made in the system can be a metric
103 * that can be used to drive ballooning decisions when Linux is hosted
104 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
105 * balancing memory across competing virtual machines that are hosted.
106 * Several metrics drive this policy engine including the guest reported
109 unsigned long vm_memory_committed(void)
111 return percpu_counter_read_positive(&vm_committed_as);
113 EXPORT_SYMBOL_GPL(vm_memory_committed);
116 * Check that a process has enough memory to allocate a new virtual
117 * mapping. 0 means there is enough memory for the allocation to
118 * succeed and -ENOMEM implies there is not.
120 * We currently support three overcommit policies, which are set via the
121 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
123 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
124 * Additional code 2002 Jul 20 by Robert Love.
126 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
128 * Note this is a helper function intended to be used by LSMs which
129 * wish to use this logic.
131 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
133 long free, allowed, reserve;
135 vm_acct_memory(pages);
138 * Sometimes we want to use more memory than we have
140 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
143 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
144 free = global_page_state(NR_FREE_PAGES);
145 free += global_page_state(NR_FILE_PAGES);
148 * shmem pages shouldn't be counted as free in this
149 * case, they can't be purged, only swapped out, and
150 * that won't affect the overall amount of available
151 * memory in the system.
153 free -= global_page_state(NR_SHMEM);
155 free += get_nr_swap_pages();
158 * Any slabs which are created with the
159 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
160 * which are reclaimable, under pressure. The dentry
161 * cache and most inode caches should fall into this
163 free += global_page_state(NR_SLAB_RECLAIMABLE);
166 * Leave reserved pages. The pages are not for anonymous pages.
168 if (free <= totalreserve_pages)
171 free -= totalreserve_pages;
174 * Reserve some for root
177 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
185 allowed = (totalram_pages - hugetlb_total_pages())
186 * sysctl_overcommit_ratio / 100;
188 * Reserve some for root
191 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
192 allowed += total_swap_pages;
195 * Don't let a single process grow so big a user can't recover
198 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
199 allowed -= min_t(long, mm->total_vm / 32, reserve);
202 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
205 vm_unacct_memory(pages);
211 * Requires inode->i_mapping->i_mmap_mutex
213 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
214 struct file *file, struct address_space *mapping)
216 if (vma->vm_flags & VM_DENYWRITE)
217 atomic_inc(&file_inode(file)->i_writecount);
218 if (vma->vm_flags & VM_SHARED)
219 mapping->i_mmap_writable--;
221 flush_dcache_mmap_lock(mapping);
222 if (unlikely(vma->vm_flags & VM_NONLINEAR))
223 list_del_init(&vma->shared.nonlinear);
225 vma_interval_tree_remove(vma, &mapping->i_mmap);
226 flush_dcache_mmap_unlock(mapping);
230 * Unlink a file-based vm structure from its interval tree, to hide
231 * vma from rmap and vmtruncate before freeing its page tables.
233 void unlink_file_vma(struct vm_area_struct *vma)
235 struct file *file = vma->vm_file;
238 struct address_space *mapping = file->f_mapping;
239 mutex_lock(&mapping->i_mmap_mutex);
240 __remove_shared_vm_struct(vma, file, mapping);
241 mutex_unlock(&mapping->i_mmap_mutex);
246 * Close a vm structure and free it, returning the next.
248 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
250 struct vm_area_struct *next = vma->vm_next;
253 if (vma->vm_ops && vma->vm_ops->close)
254 vma->vm_ops->close(vma);
257 mpol_put(vma_policy(vma));
258 kmem_cache_free(vm_area_cachep, vma);
262 static unsigned long do_brk(unsigned long addr, unsigned long len);
264 SYSCALL_DEFINE1(brk, unsigned long, brk)
266 unsigned long rlim, retval;
267 unsigned long newbrk, oldbrk;
268 struct mm_struct *mm = current->mm;
269 unsigned long min_brk;
272 down_write(&mm->mmap_sem);
274 #ifdef CONFIG_COMPAT_BRK
276 * CONFIG_COMPAT_BRK can still be overridden by setting
277 * randomize_va_space to 2, which will still cause mm->start_brk
278 * to be arbitrarily shifted
280 if (current->brk_randomized)
281 min_brk = mm->start_brk;
283 min_brk = mm->end_data;
285 min_brk = mm->start_brk;
291 * Check against rlimit here. If this check is done later after the test
292 * of oldbrk with newbrk then it can escape the test and let the data
293 * segment grow beyond its set limit the in case where the limit is
294 * not page aligned -Ram Gupta
296 rlim = rlimit(RLIMIT_DATA);
297 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
298 (mm->end_data - mm->start_data) > rlim)
301 newbrk = PAGE_ALIGN(brk);
302 oldbrk = PAGE_ALIGN(mm->brk);
303 if (oldbrk == newbrk)
306 /* Always allow shrinking brk. */
307 if (brk <= mm->brk) {
308 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
313 /* Check against existing mmap mappings. */
314 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
317 /* Ok, looks good - let it rip. */
318 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
323 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
324 up_write(&mm->mmap_sem);
326 mm_populate(oldbrk, newbrk - oldbrk);
331 up_write(&mm->mmap_sem);
335 static long vma_compute_subtree_gap(struct vm_area_struct *vma)
337 unsigned long max, subtree_gap;
340 max -= vma->vm_prev->vm_end;
341 if (vma->vm_rb.rb_left) {
342 subtree_gap = rb_entry(vma->vm_rb.rb_left,
343 struct vm_area_struct, vm_rb)->rb_subtree_gap;
344 if (subtree_gap > max)
347 if (vma->vm_rb.rb_right) {
348 subtree_gap = rb_entry(vma->vm_rb.rb_right,
349 struct vm_area_struct, vm_rb)->rb_subtree_gap;
350 if (subtree_gap > max)
356 #ifdef CONFIG_DEBUG_VM_RB
357 static int browse_rb(struct rb_root *root)
359 int i = 0, j, bug = 0;
360 struct rb_node *nd, *pn = NULL;
361 unsigned long prev = 0, pend = 0;
363 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
364 struct vm_area_struct *vma;
365 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
366 if (vma->vm_start < prev) {
367 printk("vm_start %lx prev %lx\n", vma->vm_start, prev);
370 if (vma->vm_start < pend) {
371 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
374 if (vma->vm_start > vma->vm_end) {
375 printk("vm_end %lx < vm_start %lx\n",
376 vma->vm_end, vma->vm_start);
379 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
380 printk("free gap %lx, correct %lx\n",
382 vma_compute_subtree_gap(vma));
387 prev = vma->vm_start;
391 for (nd = pn; nd; nd = rb_prev(nd))
394 printk("backwards %d, forwards %d\n", j, i);
400 static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
404 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
405 struct vm_area_struct *vma;
406 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
407 BUG_ON(vma != ignore &&
408 vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
412 void validate_mm(struct mm_struct *mm)
416 unsigned long highest_address = 0;
417 struct vm_area_struct *vma = mm->mmap;
419 struct anon_vma_chain *avc;
420 vma_lock_anon_vma(vma);
421 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
422 anon_vma_interval_tree_verify(avc);
423 vma_unlock_anon_vma(vma);
424 highest_address = vma->vm_end;
428 if (i != mm->map_count) {
429 printk("map_count %d vm_next %d\n", mm->map_count, i);
432 if (highest_address != mm->highest_vm_end) {
433 printk("mm->highest_vm_end %lx, found %lx\n",
434 mm->highest_vm_end, highest_address);
437 i = browse_rb(&mm->mm_rb);
438 if (i != mm->map_count) {
439 printk("map_count %d rb %d\n", mm->map_count, i);
445 #define validate_mm_rb(root, ignore) do { } while (0)
446 #define validate_mm(mm) do { } while (0)
449 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
450 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
453 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
454 * vma->vm_prev->vm_end values changed, without modifying the vma's position
457 static void vma_gap_update(struct vm_area_struct *vma)
460 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
461 * function that does exacltly what we want.
463 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
466 static inline void vma_rb_insert(struct vm_area_struct *vma,
467 struct rb_root *root)
469 /* All rb_subtree_gap values must be consistent prior to insertion */
470 validate_mm_rb(root, NULL);
472 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
475 static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
478 * All rb_subtree_gap values must be consistent prior to erase,
479 * with the possible exception of the vma being erased.
481 validate_mm_rb(root, vma);
484 * Note rb_erase_augmented is a fairly large inline function,
485 * so make sure we instantiate it only once with our desired
486 * augmented rbtree callbacks.
488 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
492 * vma has some anon_vma assigned, and is already inserted on that
493 * anon_vma's interval trees.
495 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
496 * vma must be removed from the anon_vma's interval trees using
497 * anon_vma_interval_tree_pre_update_vma().
499 * After the update, the vma will be reinserted using
500 * anon_vma_interval_tree_post_update_vma().
502 * The entire update must be protected by exclusive mmap_sem and by
503 * the root anon_vma's mutex.
506 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
508 struct anon_vma_chain *avc;
510 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
511 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
515 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
517 struct anon_vma_chain *avc;
519 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
520 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
523 static int find_vma_links(struct mm_struct *mm, unsigned long addr,
524 unsigned long end, struct vm_area_struct **pprev,
525 struct rb_node ***rb_link, struct rb_node **rb_parent)
527 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
529 __rb_link = &mm->mm_rb.rb_node;
530 rb_prev = __rb_parent = NULL;
533 struct vm_area_struct *vma_tmp;
535 __rb_parent = *__rb_link;
536 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
538 if (vma_tmp->vm_end > addr) {
539 /* Fail if an existing vma overlaps the area */
540 if (vma_tmp->vm_start < end)
542 __rb_link = &__rb_parent->rb_left;
544 rb_prev = __rb_parent;
545 __rb_link = &__rb_parent->rb_right;
551 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
552 *rb_link = __rb_link;
553 *rb_parent = __rb_parent;
557 static unsigned long count_vma_pages_range(struct mm_struct *mm,
558 unsigned long addr, unsigned long end)
560 unsigned long nr_pages = 0;
561 struct vm_area_struct *vma;
563 /* Find first overlaping mapping */
564 vma = find_vma_intersection(mm, addr, end);
568 nr_pages = (min(end, vma->vm_end) -
569 max(addr, vma->vm_start)) >> PAGE_SHIFT;
571 /* Iterate over the rest of the overlaps */
572 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
573 unsigned long overlap_len;
575 if (vma->vm_start > end)
578 overlap_len = min(end, vma->vm_end) - vma->vm_start;
579 nr_pages += overlap_len >> PAGE_SHIFT;
585 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
586 struct rb_node **rb_link, struct rb_node *rb_parent)
588 /* Update tracking information for the gap following the new vma. */
590 vma_gap_update(vma->vm_next);
592 mm->highest_vm_end = vma->vm_end;
595 * vma->vm_prev wasn't known when we followed the rbtree to find the
596 * correct insertion point for that vma. As a result, we could not
597 * update the vma vm_rb parents rb_subtree_gap values on the way down.
598 * So, we first insert the vma with a zero rb_subtree_gap value
599 * (to be consistent with what we did on the way down), and then
600 * immediately update the gap to the correct value. Finally we
601 * rebalance the rbtree after all augmented values have been set.
603 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
604 vma->rb_subtree_gap = 0;
606 vma_rb_insert(vma, &mm->mm_rb);
609 static void __vma_link_file(struct vm_area_struct *vma)
615 struct address_space *mapping = file->f_mapping;
617 if (vma->vm_flags & VM_DENYWRITE)
618 atomic_dec(&file_inode(file)->i_writecount);
619 if (vma->vm_flags & VM_SHARED)
620 mapping->i_mmap_writable++;
622 flush_dcache_mmap_lock(mapping);
623 if (unlikely(vma->vm_flags & VM_NONLINEAR))
624 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
626 vma_interval_tree_insert(vma, &mapping->i_mmap);
627 flush_dcache_mmap_unlock(mapping);
632 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
633 struct vm_area_struct *prev, struct rb_node **rb_link,
634 struct rb_node *rb_parent)
636 __vma_link_list(mm, vma, prev, rb_parent);
637 __vma_link_rb(mm, vma, rb_link, rb_parent);
640 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
641 struct vm_area_struct *prev, struct rb_node **rb_link,
642 struct rb_node *rb_parent)
644 struct address_space *mapping = NULL;
647 mapping = vma->vm_file->f_mapping;
650 mutex_lock(&mapping->i_mmap_mutex);
652 __vma_link(mm, vma, prev, rb_link, rb_parent);
653 __vma_link_file(vma);
656 mutex_unlock(&mapping->i_mmap_mutex);
663 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
664 * mm's list and rbtree. It has already been inserted into the interval tree.
666 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
668 struct vm_area_struct *prev;
669 struct rb_node **rb_link, *rb_parent;
671 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
672 &prev, &rb_link, &rb_parent))
674 __vma_link(mm, vma, prev, rb_link, rb_parent);
679 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
680 struct vm_area_struct *prev)
682 struct vm_area_struct *next;
684 vma_rb_erase(vma, &mm->mm_rb);
685 prev->vm_next = next = vma->vm_next;
687 next->vm_prev = prev;
688 if (mm->mmap_cache == vma)
689 mm->mmap_cache = prev;
693 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
694 * is already present in an i_mmap tree without adjusting the tree.
695 * The following helper function should be used when such adjustments
696 * are necessary. The "insert" vma (if any) is to be inserted
697 * before we drop the necessary locks.
699 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
700 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
702 struct mm_struct *mm = vma->vm_mm;
703 struct vm_area_struct *next = vma->vm_next;
704 struct vm_area_struct *importer = NULL;
705 struct address_space *mapping = NULL;
706 struct rb_root *root = NULL;
707 struct anon_vma *anon_vma = NULL;
708 struct file *file = vma->vm_file;
709 bool start_changed = false, end_changed = false;
710 long adjust_next = 0;
713 if (next && !insert) {
714 struct vm_area_struct *exporter = NULL;
716 if (end >= next->vm_end) {
718 * vma expands, overlapping all the next, and
719 * perhaps the one after too (mprotect case 6).
721 again: remove_next = 1 + (end > next->vm_end);
725 } else if (end > next->vm_start) {
727 * vma expands, overlapping part of the next:
728 * mprotect case 5 shifting the boundary up.
730 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
733 } else if (end < vma->vm_end) {
735 * vma shrinks, and !insert tells it's not
736 * split_vma inserting another: so it must be
737 * mprotect case 4 shifting the boundary down.
739 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
745 * Easily overlooked: when mprotect shifts the boundary,
746 * make sure the expanding vma has anon_vma set if the
747 * shrinking vma had, to cover any anon pages imported.
749 if (exporter && exporter->anon_vma && !importer->anon_vma) {
750 if (anon_vma_clone(importer, exporter))
752 importer->anon_vma = exporter->anon_vma;
757 mapping = file->f_mapping;
758 if (!(vma->vm_flags & VM_NONLINEAR)) {
759 root = &mapping->i_mmap;
760 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
763 uprobe_munmap(next, next->vm_start,
767 mutex_lock(&mapping->i_mmap_mutex);
770 * Put into interval tree now, so instantiated pages
771 * are visible to arm/parisc __flush_dcache_page
772 * throughout; but we cannot insert into address
773 * space until vma start or end is updated.
775 __vma_link_file(insert);
779 vma_adjust_trans_huge(vma, start, end, adjust_next);
781 anon_vma = vma->anon_vma;
782 if (!anon_vma && adjust_next)
783 anon_vma = next->anon_vma;
785 VM_BUG_ON(adjust_next && next->anon_vma &&
786 anon_vma != next->anon_vma);
787 anon_vma_lock_write(anon_vma);
788 anon_vma_interval_tree_pre_update_vma(vma);
790 anon_vma_interval_tree_pre_update_vma(next);
794 flush_dcache_mmap_lock(mapping);
795 vma_interval_tree_remove(vma, root);
797 vma_interval_tree_remove(next, root);
800 if (start != vma->vm_start) {
801 vma->vm_start = start;
802 start_changed = true;
804 if (end != vma->vm_end) {
808 vma->vm_pgoff = pgoff;
810 next->vm_start += adjust_next << PAGE_SHIFT;
811 next->vm_pgoff += adjust_next;
816 vma_interval_tree_insert(next, root);
817 vma_interval_tree_insert(vma, root);
818 flush_dcache_mmap_unlock(mapping);
823 * vma_merge has merged next into vma, and needs
824 * us to remove next before dropping the locks.
826 __vma_unlink(mm, next, vma);
828 __remove_shared_vm_struct(next, file, mapping);
831 * split_vma has split insert from vma, and needs
832 * us to insert it before dropping the locks
833 * (it may either follow vma or precede it).
835 __insert_vm_struct(mm, insert);
841 mm->highest_vm_end = end;
842 else if (!adjust_next)
843 vma_gap_update(next);
848 anon_vma_interval_tree_post_update_vma(vma);
850 anon_vma_interval_tree_post_update_vma(next);
851 anon_vma_unlock_write(anon_vma);
854 mutex_unlock(&mapping->i_mmap_mutex);
865 uprobe_munmap(next, next->vm_start, next->vm_end);
869 anon_vma_merge(vma, next);
871 mpol_put(vma_policy(next));
872 kmem_cache_free(vm_area_cachep, next);
874 * In mprotect's case 6 (see comments on vma_merge),
875 * we must remove another next too. It would clutter
876 * up the code too much to do both in one go.
879 if (remove_next == 2)
882 vma_gap_update(next);
884 mm->highest_vm_end = end;
895 * If the vma has a ->close operation then the driver probably needs to release
896 * per-vma resources, so we don't attempt to merge those.
898 static inline int is_mergeable_vma(struct vm_area_struct *vma,
899 struct file *file, unsigned long vm_flags,
900 const char __user *anon_name)
902 if (vma->vm_flags ^ vm_flags)
904 if (vma->vm_file != file)
906 if (vma->vm_ops && vma->vm_ops->close)
908 if (vma_get_anon_name(vma) != anon_name)
913 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
914 struct anon_vma *anon_vma2,
915 struct vm_area_struct *vma)
918 * The list_is_singular() test is to avoid merging VMA cloned from
919 * parents. This can improve scalability caused by anon_vma lock.
921 if ((!anon_vma1 || !anon_vma2) && (!vma ||
922 list_is_singular(&vma->anon_vma_chain)))
924 return anon_vma1 == anon_vma2;
928 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
929 * in front of (at a lower virtual address and file offset than) the vma.
931 * We cannot merge two vmas if they have differently assigned (non-NULL)
932 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
934 * We don't check here for the merged mmap wrapping around the end of pagecache
935 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
936 * wrap, nor mmaps which cover the final page at index -1UL.
939 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
940 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff,
941 const char __user *anon_name)
943 if (is_mergeable_vma(vma, file, vm_flags, anon_name) &&
944 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
945 if (vma->vm_pgoff == vm_pgoff)
952 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
953 * beyond (at a higher virtual address and file offset than) the vma.
955 * We cannot merge two vmas if they have differently assigned (non-NULL)
956 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
959 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
960 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff,
961 const char __user *anon_name)
963 if (is_mergeable_vma(vma, file, vm_flags, anon_name) &&
964 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
966 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
967 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
974 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
975 * figure out whether that can be merged with its predecessor or its
976 * successor. Or both (it neatly fills a hole).
978 * In most cases - when called for mmap, brk or mremap - [addr,end) is
979 * certain not to be mapped by the time vma_merge is called; but when
980 * called for mprotect, it is certain to be already mapped (either at
981 * an offset within prev, or at the start of next), and the flags of
982 * this area are about to be changed to vm_flags - and the no-change
983 * case has already been eliminated.
985 * The following mprotect cases have to be considered, where AAAA is
986 * the area passed down from mprotect_fixup, never extending beyond one
987 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
989 * AAAA AAAA AAAA AAAA
990 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
991 * cannot merge might become might become might become
992 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
993 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
994 * mremap move: PPPPNNNNNNNN 8
996 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
997 * might become case 1 below case 2 below case 3 below
999 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1000 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1002 struct vm_area_struct *vma_merge(struct mm_struct *mm,
1003 struct vm_area_struct *prev, unsigned long addr,
1004 unsigned long end, unsigned long vm_flags,
1005 struct anon_vma *anon_vma, struct file *file,
1006 pgoff_t pgoff, struct mempolicy *policy,
1007 const char __user *anon_name)
1009 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1010 struct vm_area_struct *area, *next;
1014 * We later require that vma->vm_flags == vm_flags,
1015 * so this tests vma->vm_flags & VM_SPECIAL, too.
1017 if (vm_flags & VM_SPECIAL)
1021 next = prev->vm_next;
1025 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1026 next = next->vm_next;
1029 * Can it merge with the predecessor?
1031 if (prev && prev->vm_end == addr &&
1032 mpol_equal(vma_policy(prev), policy) &&
1033 can_vma_merge_after(prev, vm_flags, anon_vma,
1034 file, pgoff, anon_name)) {
1036 * OK, it can. Can we now merge in the successor as well?
1038 if (next && end == next->vm_start &&
1039 mpol_equal(policy, vma_policy(next)) &&
1040 can_vma_merge_before(next, vm_flags, anon_vma,
1041 file, pgoff+pglen, anon_name) &&
1042 is_mergeable_anon_vma(prev->anon_vma,
1043 next->anon_vma, NULL)) {
1045 err = vma_adjust(prev, prev->vm_start,
1046 next->vm_end, prev->vm_pgoff, NULL);
1047 } else /* cases 2, 5, 7 */
1048 err = vma_adjust(prev, prev->vm_start,
1049 end, prev->vm_pgoff, NULL);
1052 khugepaged_enter_vma_merge(prev);
1057 * Can this new request be merged in front of next?
1059 if (next && end == next->vm_start &&
1060 mpol_equal(policy, vma_policy(next)) &&
1061 can_vma_merge_before(next, vm_flags, anon_vma,
1062 file, pgoff+pglen, anon_name)) {
1063 if (prev && addr < prev->vm_end) /* case 4 */
1064 err = vma_adjust(prev, prev->vm_start,
1065 addr, prev->vm_pgoff, NULL);
1066 else /* cases 3, 8 */
1067 err = vma_adjust(area, addr, next->vm_end,
1068 next->vm_pgoff - pglen, NULL);
1071 khugepaged_enter_vma_merge(area);
1079 * Rough compatbility check to quickly see if it's even worth looking
1080 * at sharing an anon_vma.
1082 * They need to have the same vm_file, and the flags can only differ
1083 * in things that mprotect may change.
1085 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1086 * we can merge the two vma's. For example, we refuse to merge a vma if
1087 * there is a vm_ops->close() function, because that indicates that the
1088 * driver is doing some kind of reference counting. But that doesn't
1089 * really matter for the anon_vma sharing case.
1091 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1093 return a->vm_end == b->vm_start &&
1094 mpol_equal(vma_policy(a), vma_policy(b)) &&
1095 a->vm_file == b->vm_file &&
1096 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
1097 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1101 * Do some basic sanity checking to see if we can re-use the anon_vma
1102 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1103 * the same as 'old', the other will be the new one that is trying
1104 * to share the anon_vma.
1106 * NOTE! This runs with mm_sem held for reading, so it is possible that
1107 * the anon_vma of 'old' is concurrently in the process of being set up
1108 * by another page fault trying to merge _that_. But that's ok: if it
1109 * is being set up, that automatically means that it will be a singleton
1110 * acceptable for merging, so we can do all of this optimistically. But
1111 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1113 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1114 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1115 * is to return an anon_vma that is "complex" due to having gone through
1118 * We also make sure that the two vma's are compatible (adjacent,
1119 * and with the same memory policies). That's all stable, even with just
1120 * a read lock on the mm_sem.
1122 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1124 if (anon_vma_compatible(a, b)) {
1125 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1127 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1134 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1135 * neighbouring vmas for a suitable anon_vma, before it goes off
1136 * to allocate a new anon_vma. It checks because a repetitive
1137 * sequence of mprotects and faults may otherwise lead to distinct
1138 * anon_vmas being allocated, preventing vma merge in subsequent
1141 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1143 struct anon_vma *anon_vma;
1144 struct vm_area_struct *near;
1146 near = vma->vm_next;
1150 anon_vma = reusable_anon_vma(near, vma, near);
1154 near = vma->vm_prev;
1158 anon_vma = reusable_anon_vma(near, near, vma);
1163 * There's no absolute need to look only at touching neighbours:
1164 * we could search further afield for "compatible" anon_vmas.
1165 * But it would probably just be a waste of time searching,
1166 * or lead to too many vmas hanging off the same anon_vma.
1167 * We're trying to allow mprotect remerging later on,
1168 * not trying to minimize memory used for anon_vmas.
1173 #ifdef CONFIG_PROC_FS
1174 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
1175 struct file *file, long pages)
1177 const unsigned long stack_flags
1178 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1180 mm->total_vm += pages;
1183 mm->shared_vm += pages;
1184 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1185 mm->exec_vm += pages;
1186 } else if (flags & stack_flags)
1187 mm->stack_vm += pages;
1189 #endif /* CONFIG_PROC_FS */
1192 * If a hint addr is less than mmap_min_addr change hint to be as
1193 * low as possible but still greater than mmap_min_addr
1195 static inline unsigned long round_hint_to_min(unsigned long hint)
1198 if (((void *)hint != NULL) &&
1199 (hint < mmap_min_addr))
1200 return PAGE_ALIGN(mmap_min_addr);
1205 * The caller must hold down_write(¤t->mm->mmap_sem).
1208 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
1209 unsigned long len, unsigned long prot,
1210 unsigned long flags, unsigned long pgoff,
1211 unsigned long *populate)
1213 struct mm_struct * mm = current->mm;
1214 struct inode *inode;
1215 vm_flags_t vm_flags;
1218 #ifdef CONFIG_SDCARD_FS
1219 if (file && (file->f_path.mnt->mnt_sb->s_magic == SDCARDFS_SUPER_MAGIC))
1220 file = sdcardfs_lower_file(file);
1224 * Does the application expect PROT_READ to imply PROT_EXEC?
1226 * (the exception is when the underlying filesystem is noexec
1227 * mounted, in which case we dont add PROT_EXEC.)
1229 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1230 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
1236 if (!(flags & MAP_FIXED))
1237 addr = round_hint_to_min(addr);
1239 /* Careful about overflows.. */
1240 len = PAGE_ALIGN(len);
1244 /* offset overflow? */
1245 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1248 /* Too many mappings? */
1249 if (mm->map_count > sysctl_max_map_count)
1252 /* Obtain the address to map to. we verify (or select) it and ensure
1253 * that it represents a valid section of the address space.
1255 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1256 if (addr & ~PAGE_MASK)
1259 /* Do simple checking here so the lower-level routines won't have
1260 * to. we assume access permissions have been handled by the open
1261 * of the memory object, so we don't do any here.
1263 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1264 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1266 if (flags & MAP_LOCKED)
1267 if (!can_do_mlock())
1270 /* mlock MCL_FUTURE? */
1271 if (vm_flags & VM_LOCKED) {
1272 unsigned long locked, lock_limit;
1273 locked = len >> PAGE_SHIFT;
1274 locked += mm->locked_vm;
1275 lock_limit = rlimit(RLIMIT_MEMLOCK);
1276 lock_limit >>= PAGE_SHIFT;
1277 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1281 inode = file ? file_inode(file) : NULL;
1284 switch (flags & MAP_TYPE) {
1286 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1290 * Make sure we don't allow writing to an append-only
1293 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1297 * Make sure there are no mandatory locks on the file.
1299 if (locks_verify_locked(inode))
1302 vm_flags |= VM_SHARED | VM_MAYSHARE;
1303 if (!(file->f_mode & FMODE_WRITE))
1304 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1308 if (!(file->f_mode & FMODE_READ))
1310 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1311 if (vm_flags & VM_EXEC)
1313 vm_flags &= ~VM_MAYEXEC;
1316 if (!file->f_op || !file->f_op->mmap)
1324 switch (flags & MAP_TYPE) {
1330 vm_flags |= VM_SHARED | VM_MAYSHARE;
1334 * Set pgoff according to addr for anon_vma.
1336 pgoff = addr >> PAGE_SHIFT;
1344 * Set 'VM_NORESERVE' if we should not account for the
1345 * memory use of this mapping.
1347 if (flags & MAP_NORESERVE) {
1348 /* We honor MAP_NORESERVE if allowed to overcommit */
1349 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1350 vm_flags |= VM_NORESERVE;
1352 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1353 if (file && is_file_hugepages(file))
1354 vm_flags |= VM_NORESERVE;
1357 addr = mmap_region(file, addr, len, vm_flags, pgoff);
1358 if (!IS_ERR_VALUE(addr) &&
1359 ((vm_flags & VM_LOCKED) ||
1360 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1365 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1366 unsigned long, prot, unsigned long, flags,
1367 unsigned long, fd, unsigned long, pgoff)
1369 struct file *file = NULL;
1370 unsigned long retval = -EBADF;
1372 if (!(flags & MAP_ANONYMOUS)) {
1373 audit_mmap_fd(fd, flags);
1374 if (unlikely(flags & MAP_HUGETLB))
1379 if (is_file_hugepages(file))
1380 len = ALIGN(len, huge_page_size(hstate_file(file)));
1381 } else if (flags & MAP_HUGETLB) {
1382 struct user_struct *user = NULL;
1383 struct hstate *hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) &
1389 len = ALIGN(len, huge_page_size(hs));
1391 * VM_NORESERVE is used because the reservations will be
1392 * taken when vm_ops->mmap() is called
1393 * A dummy user value is used because we are not locking
1394 * memory so no accounting is necessary
1396 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1398 &user, HUGETLB_ANONHUGE_INODE,
1399 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1401 return PTR_ERR(file);
1404 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1406 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1413 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1414 struct mmap_arg_struct {
1418 unsigned long flags;
1420 unsigned long offset;
1423 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1425 struct mmap_arg_struct a;
1427 if (copy_from_user(&a, arg, sizeof(a)))
1429 if (a.offset & ~PAGE_MASK)
1432 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1433 a.offset >> PAGE_SHIFT);
1435 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1438 * Some shared mappigns will want the pages marked read-only
1439 * to track write events. If so, we'll downgrade vm_page_prot
1440 * to the private version (using protection_map[] without the
1443 int vma_wants_writenotify(struct vm_area_struct *vma)
1445 vm_flags_t vm_flags = vma->vm_flags;
1447 /* If it was private or non-writable, the write bit is already clear */
1448 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1451 /* The backer wishes to know when pages are first written to? */
1452 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1455 /* The open routine did something to the protections already? */
1456 if (pgprot_val(vma->vm_page_prot) !=
1457 pgprot_val(vm_get_page_prot(vm_flags)))
1460 /* Specialty mapping? */
1461 if (vm_flags & VM_PFNMAP)
1464 /* Can the mapping track the dirty pages? */
1465 return vma->vm_file && vma->vm_file->f_mapping &&
1466 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1470 * We account for memory if it's a private writeable mapping,
1471 * not hugepages and VM_NORESERVE wasn't set.
1473 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1476 * hugetlb has its own accounting separate from the core VM
1477 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1479 if (file && is_file_hugepages(file))
1482 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1485 unsigned long mmap_region(struct file *file, unsigned long addr,
1486 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
1488 struct mm_struct *mm = current->mm;
1489 struct vm_area_struct *vma, *prev;
1490 int correct_wcount = 0;
1492 struct rb_node **rb_link, *rb_parent;
1493 unsigned long charged = 0;
1494 struct inode *inode = file ? file_inode(file) : NULL;
1496 /* Check against address space limit. */
1497 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1498 unsigned long nr_pages;
1501 * MAP_FIXED may remove pages of mappings that intersects with
1502 * requested mapping. Account for the pages it would unmap.
1504 if (!(vm_flags & MAP_FIXED))
1507 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1509 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1513 /* Clear old maps */
1516 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
1517 if (do_munmap(mm, addr, len))
1523 * Private writable mapping: check memory availability
1525 if (accountable_mapping(file, vm_flags)) {
1526 charged = len >> PAGE_SHIFT;
1527 if (security_vm_enough_memory_mm(mm, charged))
1529 vm_flags |= VM_ACCOUNT;
1533 * Can we just expand an old mapping?
1535 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff,
1541 * Determine the object being mapped and call the appropriate
1542 * specific mapper. the address has already been validated, but
1543 * not unmapped, but the maps are removed from the list.
1545 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1552 vma->vm_start = addr;
1553 vma->vm_end = addr + len;
1554 vma->vm_flags = vm_flags;
1555 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1556 vma->vm_pgoff = pgoff;
1557 INIT_LIST_HEAD(&vma->anon_vma_chain);
1559 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1562 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1564 if (vm_flags & VM_DENYWRITE) {
1565 error = deny_write_access(file);
1570 vma->vm_file = get_file(file);
1571 error = file->f_op->mmap(file, vma);
1573 goto unmap_and_free_vma;
1575 /* Can addr have changed??
1577 * Answer: Yes, several device drivers can do it in their
1578 * f_op->mmap method. -DaveM
1579 * Bug: If addr is changed, prev, rb_link, rb_parent should
1580 * be updated for vma_link()
1582 WARN_ON_ONCE(addr != vma->vm_start);
1584 addr = vma->vm_start;
1585 pgoff = vma->vm_pgoff;
1586 vm_flags = vma->vm_flags;
1587 } else if (vm_flags & VM_SHARED) {
1588 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1590 error = shmem_zero_setup(vma);
1595 if (vma_wants_writenotify(vma)) {
1596 pgprot_t pprot = vma->vm_page_prot;
1598 /* Can vma->vm_page_prot have changed??
1600 * Answer: Yes, drivers may have changed it in their
1601 * f_op->mmap method.
1603 * Ensures that vmas marked as uncached stay that way.
1605 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1606 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1607 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1610 vma_link(mm, vma, prev, rb_link, rb_parent);
1611 file = vma->vm_file;
1613 /* Once vma denies write, undo our temporary denial count */
1615 atomic_inc(&inode->i_writecount);
1617 perf_event_mmap(vma);
1619 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1620 if (vm_flags & VM_LOCKED) {
1621 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1622 vma == get_gate_vma(current->mm)))
1623 mm->locked_vm += (len >> PAGE_SHIFT);
1625 vma->vm_flags &= ~VM_LOCKED;
1635 atomic_inc(&inode->i_writecount);
1636 vma->vm_file = NULL;
1639 /* Undo any partial mapping done by a device driver. */
1640 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1643 kmem_cache_free(vm_area_cachep, vma);
1646 vm_unacct_memory(charged);
1650 unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1653 * We implement the search by looking for an rbtree node that
1654 * immediately follows a suitable gap. That is,
1655 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1656 * - gap_end = vma->vm_start >= info->low_limit + length;
1657 * - gap_end - gap_start >= length
1660 struct mm_struct *mm = current->mm;
1661 struct vm_area_struct *vma;
1662 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1664 /* Adjust search length to account for worst case alignment overhead */
1665 length = info->length + info->align_mask;
1666 if (length < info->length)
1669 /* Adjust search limits by the desired length */
1670 if (info->high_limit < length)
1672 high_limit = info->high_limit - length;
1674 if (info->low_limit > high_limit)
1676 low_limit = info->low_limit + length;
1678 /* Check if rbtree root looks promising */
1679 if (RB_EMPTY_ROOT(&mm->mm_rb))
1681 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1682 if (vma->rb_subtree_gap < length)
1686 /* Visit left subtree if it looks promising */
1687 gap_end = vma->vm_start;
1688 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1689 struct vm_area_struct *left =
1690 rb_entry(vma->vm_rb.rb_left,
1691 struct vm_area_struct, vm_rb);
1692 if (left->rb_subtree_gap >= length) {
1698 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1700 /* Check if current node has a suitable gap */
1701 if (gap_start > high_limit)
1703 if (gap_end >= low_limit && gap_end - gap_start >= length)
1706 /* Visit right subtree if it looks promising */
1707 if (vma->vm_rb.rb_right) {
1708 struct vm_area_struct *right =
1709 rb_entry(vma->vm_rb.rb_right,
1710 struct vm_area_struct, vm_rb);
1711 if (right->rb_subtree_gap >= length) {
1717 /* Go back up the rbtree to find next candidate node */
1719 struct rb_node *prev = &vma->vm_rb;
1720 if (!rb_parent(prev))
1722 vma = rb_entry(rb_parent(prev),
1723 struct vm_area_struct, vm_rb);
1724 if (prev == vma->vm_rb.rb_left) {
1725 gap_start = vma->vm_prev->vm_end;
1726 gap_end = vma->vm_start;
1733 /* Check highest gap, which does not precede any rbtree node */
1734 gap_start = mm->highest_vm_end;
1735 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1736 if (gap_start > high_limit)
1740 /* We found a suitable gap. Clip it with the original low_limit. */
1741 if (gap_start < info->low_limit)
1742 gap_start = info->low_limit;
1744 /* Adjust gap address to the desired alignment */
1745 gap_start += (info->align_offset - gap_start) & info->align_mask;
1747 VM_BUG_ON(gap_start + info->length > info->high_limit);
1748 VM_BUG_ON(gap_start + info->length > gap_end);
1752 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1754 struct mm_struct *mm = current->mm;
1755 struct vm_area_struct *vma;
1756 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1758 /* Adjust search length to account for worst case alignment overhead */
1759 length = info->length + info->align_mask;
1760 if (length < info->length)
1764 * Adjust search limits by the desired length.
1765 * See implementation comment at top of unmapped_area().
1767 gap_end = info->high_limit;
1768 if (gap_end < length)
1770 high_limit = gap_end - length;
1772 if (info->low_limit > high_limit)
1774 low_limit = info->low_limit + length;
1776 /* Check highest gap, which does not precede any rbtree node */
1777 gap_start = mm->highest_vm_end;
1778 if (gap_start <= high_limit)
1781 /* Check if rbtree root looks promising */
1782 if (RB_EMPTY_ROOT(&mm->mm_rb))
1784 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1785 if (vma->rb_subtree_gap < length)
1789 /* Visit right subtree if it looks promising */
1790 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1791 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1792 struct vm_area_struct *right =
1793 rb_entry(vma->vm_rb.rb_right,
1794 struct vm_area_struct, vm_rb);
1795 if (right->rb_subtree_gap >= length) {
1802 /* Check if current node has a suitable gap */
1803 gap_end = vma->vm_start;
1804 if (gap_end < low_limit)
1806 if (gap_start <= high_limit && gap_end - gap_start >= length)
1809 /* Visit left subtree if it looks promising */
1810 if (vma->vm_rb.rb_left) {
1811 struct vm_area_struct *left =
1812 rb_entry(vma->vm_rb.rb_left,
1813 struct vm_area_struct, vm_rb);
1814 if (left->rb_subtree_gap >= length) {
1820 /* Go back up the rbtree to find next candidate node */
1822 struct rb_node *prev = &vma->vm_rb;
1823 if (!rb_parent(prev))
1825 vma = rb_entry(rb_parent(prev),
1826 struct vm_area_struct, vm_rb);
1827 if (prev == vma->vm_rb.rb_right) {
1828 gap_start = vma->vm_prev ?
1829 vma->vm_prev->vm_end : 0;
1836 /* We found a suitable gap. Clip it with the original high_limit. */
1837 if (gap_end > info->high_limit)
1838 gap_end = info->high_limit;
1841 /* Compute highest gap address at the desired alignment */
1842 gap_end -= info->length;
1843 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1845 VM_BUG_ON(gap_end < info->low_limit);
1846 VM_BUG_ON(gap_end < gap_start);
1850 /* Get an address range which is currently unmapped.
1851 * For shmat() with addr=0.
1853 * Ugly calling convention alert:
1854 * Return value with the low bits set means error value,
1856 * if (ret & ~PAGE_MASK)
1859 * This function "knows" that -ENOMEM has the bits set.
1861 #ifndef HAVE_ARCH_UNMAPPED_AREA
1863 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1864 unsigned long len, unsigned long pgoff, unsigned long flags)
1866 struct mm_struct *mm = current->mm;
1867 struct vm_area_struct *vma;
1868 struct vm_unmapped_area_info info;
1870 if (len > TASK_SIZE - mmap_min_addr)
1873 if (flags & MAP_FIXED)
1877 addr = PAGE_ALIGN(addr);
1878 vma = find_vma(mm, addr);
1879 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1880 (!vma || addr + len <= vma->vm_start))
1886 info.low_limit = TASK_UNMAPPED_BASE;
1887 info.high_limit = TASK_SIZE;
1888 info.align_mask = 0;
1889 return vm_unmapped_area(&info);
1893 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1896 * Is this a new hole at the lowest possible address?
1898 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
1899 mm->free_area_cache = addr;
1903 * This mmap-allocator allocates new areas top-down from below the
1904 * stack's low limit (the base):
1906 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1908 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1909 const unsigned long len, const unsigned long pgoff,
1910 const unsigned long flags)
1912 struct vm_area_struct *vma;
1913 struct mm_struct *mm = current->mm;
1914 unsigned long addr = addr0;
1915 struct vm_unmapped_area_info info;
1917 /* requested length too big for entire address space */
1918 if (len > TASK_SIZE - mmap_min_addr)
1921 if (flags & MAP_FIXED)
1924 /* requesting a specific address */
1926 addr = PAGE_ALIGN(addr);
1927 vma = find_vma(mm, addr);
1928 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1929 (!vma || addr + len <= vma->vm_start))
1933 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1935 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
1936 info.high_limit = mm->mmap_base;
1937 info.align_mask = 0;
1938 addr = vm_unmapped_area(&info);
1941 * A failed mmap() very likely causes application failure,
1942 * so fall back to the bottom-up function here. This scenario
1943 * can happen with large stack limits and large mmap()
1946 if (addr & ~PAGE_MASK) {
1947 VM_BUG_ON(addr != -ENOMEM);
1949 info.low_limit = TASK_UNMAPPED_BASE;
1950 info.high_limit = TASK_SIZE;
1951 addr = vm_unmapped_area(&info);
1958 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1961 * Is this a new hole at the highest possible address?
1963 if (addr > mm->free_area_cache)
1964 mm->free_area_cache = addr;
1966 /* dont allow allocations above current base */
1967 if (mm->free_area_cache > mm->mmap_base)
1968 mm->free_area_cache = mm->mmap_base;
1972 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1973 unsigned long pgoff, unsigned long flags)
1975 unsigned long (*get_area)(struct file *, unsigned long,
1976 unsigned long, unsigned long, unsigned long);
1978 unsigned long error = arch_mmap_check(addr, len, flags);
1982 /* Careful about overflows.. */
1983 if (len > TASK_SIZE)
1986 get_area = current->mm->get_unmapped_area;
1987 if (file && file->f_op && file->f_op->get_unmapped_area)
1988 get_area = file->f_op->get_unmapped_area;
1989 addr = get_area(file, addr, len, pgoff, flags);
1990 if (IS_ERR_VALUE(addr))
1993 if (addr > TASK_SIZE - len)
1995 if (addr & ~PAGE_MASK)
1998 addr = arch_rebalance_pgtables(addr, len);
1999 error = security_mmap_addr(addr);
2000 return error ? error : addr;
2003 EXPORT_SYMBOL(get_unmapped_area);
2005 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2006 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
2008 struct vm_area_struct *vma = NULL;
2010 /* Check the cache first. */
2011 /* (Cache hit rate is typically around 35%.) */
2012 vma = ACCESS_ONCE(mm->mmap_cache);
2013 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
2014 struct rb_node *rb_node;
2016 rb_node = mm->mm_rb.rb_node;
2020 struct vm_area_struct *vma_tmp;
2022 vma_tmp = rb_entry(rb_node,
2023 struct vm_area_struct, vm_rb);
2025 if (vma_tmp->vm_end > addr) {
2027 if (vma_tmp->vm_start <= addr)
2029 rb_node = rb_node->rb_left;
2031 rb_node = rb_node->rb_right;
2034 mm->mmap_cache = vma;
2039 EXPORT_SYMBOL(find_vma);
2042 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2044 struct vm_area_struct *
2045 find_vma_prev(struct mm_struct *mm, unsigned long addr,
2046 struct vm_area_struct **pprev)
2048 struct vm_area_struct *vma;
2050 vma = find_vma(mm, addr);
2052 *pprev = vma->vm_prev;
2054 struct rb_node *rb_node = mm->mm_rb.rb_node;
2057 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2058 rb_node = rb_node->rb_right;
2065 * Verify that the stack growth is acceptable and
2066 * update accounting. This is shared with both the
2067 * grow-up and grow-down cases.
2069 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
2071 struct mm_struct *mm = vma->vm_mm;
2072 struct rlimit *rlim = current->signal->rlim;
2073 unsigned long new_start, actual_size;
2075 /* address space limit tests */
2076 if (!may_expand_vm(mm, grow))
2079 /* Stack limit test */
2081 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2082 actual_size -= PAGE_SIZE;
2083 if (actual_size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
2086 /* mlock limit tests */
2087 if (vma->vm_flags & VM_LOCKED) {
2088 unsigned long locked;
2089 unsigned long limit;
2090 locked = mm->locked_vm + grow;
2091 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2092 limit >>= PAGE_SHIFT;
2093 if (locked > limit && !capable(CAP_IPC_LOCK))
2097 /* Check to ensure the stack will not grow into a hugetlb-only region */
2098 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2100 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2104 * Overcommit.. This must be the final test, as it will
2105 * update security statistics.
2107 if (security_vm_enough_memory_mm(mm, grow))
2110 /* Ok, everything looks good - let it rip */
2111 if (vma->vm_flags & VM_LOCKED)
2112 mm->locked_vm += grow;
2113 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
2117 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2119 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2120 * vma is the last one with address > vma->vm_end. Have to extend vma.
2122 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2126 if (!(vma->vm_flags & VM_GROWSUP))
2130 * We must make sure the anon_vma is allocated
2131 * so that the anon_vma locking is not a noop.
2133 if (unlikely(anon_vma_prepare(vma)))
2135 vma_lock_anon_vma(vma);
2138 * vma->vm_start/vm_end cannot change under us because the caller
2139 * is required to hold the mmap_sem in read mode. We need the
2140 * anon_vma lock to serialize against concurrent expand_stacks.
2141 * Also guard against wrapping around to address 0.
2143 if (address < PAGE_ALIGN(address+4))
2144 address = PAGE_ALIGN(address+4);
2146 vma_unlock_anon_vma(vma);
2151 /* Somebody else might have raced and expanded it already */
2152 if (address > vma->vm_end) {
2153 unsigned long size, grow;
2155 size = address - vma->vm_start;
2156 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2159 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2160 error = acct_stack_growth(vma, size, grow);
2163 * vma_gap_update() doesn't support concurrent
2164 * updates, but we only hold a shared mmap_sem
2165 * lock here, so we need to protect against
2166 * concurrent vma expansions.
2167 * vma_lock_anon_vma() doesn't help here, as
2168 * we don't guarantee that all growable vmas
2169 * in a mm share the same root anon vma.
2170 * So, we reuse mm->page_table_lock to guard
2171 * against concurrent vma expansions.
2173 spin_lock(&vma->vm_mm->page_table_lock);
2174 anon_vma_interval_tree_pre_update_vma(vma);
2175 vma->vm_end = address;
2176 anon_vma_interval_tree_post_update_vma(vma);
2178 vma_gap_update(vma->vm_next);
2180 vma->vm_mm->highest_vm_end = address;
2181 spin_unlock(&vma->vm_mm->page_table_lock);
2183 perf_event_mmap(vma);
2187 vma_unlock_anon_vma(vma);
2188 khugepaged_enter_vma_merge(vma);
2189 validate_mm(vma->vm_mm);
2192 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2195 * vma is the first one with address < vma->vm_start. Have to extend vma.
2197 int expand_downwards(struct vm_area_struct *vma,
2198 unsigned long address)
2203 * We must make sure the anon_vma is allocated
2204 * so that the anon_vma locking is not a noop.
2206 if (unlikely(anon_vma_prepare(vma)))
2209 address &= PAGE_MASK;
2210 error = security_mmap_addr(address);
2214 vma_lock_anon_vma(vma);
2217 * vma->vm_start/vm_end cannot change under us because the caller
2218 * is required to hold the mmap_sem in read mode. We need the
2219 * anon_vma lock to serialize against concurrent expand_stacks.
2222 /* Somebody else might have raced and expanded it already */
2223 if (address < vma->vm_start) {
2224 unsigned long size, grow;
2226 size = vma->vm_end - address;
2227 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2230 if (grow <= vma->vm_pgoff) {
2231 error = acct_stack_growth(vma, size, grow);
2234 * vma_gap_update() doesn't support concurrent
2235 * updates, but we only hold a shared mmap_sem
2236 * lock here, so we need to protect against
2237 * concurrent vma expansions.
2238 * vma_lock_anon_vma() doesn't help here, as
2239 * we don't guarantee that all growable vmas
2240 * in a mm share the same root anon vma.
2241 * So, we reuse mm->page_table_lock to guard
2242 * against concurrent vma expansions.
2244 spin_lock(&vma->vm_mm->page_table_lock);
2245 anon_vma_interval_tree_pre_update_vma(vma);
2246 vma->vm_start = address;
2247 vma->vm_pgoff -= grow;
2248 anon_vma_interval_tree_post_update_vma(vma);
2249 vma_gap_update(vma);
2250 spin_unlock(&vma->vm_mm->page_table_lock);
2252 perf_event_mmap(vma);
2256 vma_unlock_anon_vma(vma);
2257 khugepaged_enter_vma_merge(vma);
2258 validate_mm(vma->vm_mm);
2263 * Note how expand_stack() refuses to expand the stack all the way to
2264 * abut the next virtual mapping, *unless* that mapping itself is also
2265 * a stack mapping. We want to leave room for a guard page, after all
2266 * (the guard page itself is not added here, that is done by the
2267 * actual page faulting logic)
2269 * This matches the behavior of the guard page logic (see mm/memory.c:
2270 * check_stack_guard_page()), which only allows the guard page to be
2271 * removed under these circumstances.
2273 #ifdef CONFIG_STACK_GROWSUP
2274 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2276 struct vm_area_struct *next;
2278 address &= PAGE_MASK;
2279 next = vma->vm_next;
2280 if (next && next->vm_start == address + PAGE_SIZE) {
2281 if (!(next->vm_flags & VM_GROWSUP))
2284 return expand_upwards(vma, address);
2287 struct vm_area_struct *
2288 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2290 struct vm_area_struct *vma, *prev;
2293 vma = find_vma_prev(mm, addr, &prev);
2294 if (vma && (vma->vm_start <= addr))
2296 if (!prev || expand_stack(prev, addr))
2298 if (prev->vm_flags & VM_LOCKED)
2299 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
2303 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2305 struct vm_area_struct *prev;
2307 address &= PAGE_MASK;
2308 prev = vma->vm_prev;
2309 if (prev && prev->vm_end == address) {
2310 if (!(prev->vm_flags & VM_GROWSDOWN))
2313 return expand_downwards(vma, address);
2316 struct vm_area_struct *
2317 find_extend_vma(struct mm_struct * mm, unsigned long addr)
2319 struct vm_area_struct * vma;
2320 unsigned long start;
2323 vma = find_vma(mm,addr);
2326 if (vma->vm_start <= addr)
2328 if (!(vma->vm_flags & VM_GROWSDOWN))
2330 start = vma->vm_start;
2331 if (expand_stack(vma, addr))
2333 if (vma->vm_flags & VM_LOCKED)
2334 __mlock_vma_pages_range(vma, addr, start, NULL);
2340 * Ok - we have the memory areas we should free on the vma list,
2341 * so release them, and do the vma updates.
2343 * Called with the mm semaphore held.
2345 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2347 unsigned long nr_accounted = 0;
2349 /* Update high watermark before we lower total_vm */
2350 update_hiwater_vm(mm);
2352 long nrpages = vma_pages(vma);
2354 if (vma->vm_flags & VM_ACCOUNT)
2355 nr_accounted += nrpages;
2356 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
2357 vma = remove_vma(vma);
2359 vm_unacct_memory(nr_accounted);
2364 * Get rid of page table information in the indicated region.
2366 * Called with the mm semaphore held.
2368 static void unmap_region(struct mm_struct *mm,
2369 struct vm_area_struct *vma, struct vm_area_struct *prev,
2370 unsigned long start, unsigned long end)
2372 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
2373 struct mmu_gather tlb;
2376 tlb_gather_mmu(&tlb, mm, start, end);
2377 update_hiwater_rss(mm);
2378 unmap_vmas(&tlb, vma, start, end);
2379 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2380 next ? next->vm_start : USER_PGTABLES_CEILING);
2381 tlb_finish_mmu(&tlb, start, end);
2385 * Create a list of vma's touched by the unmap, removing them from the mm's
2386 * vma list as we go..
2389 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2390 struct vm_area_struct *prev, unsigned long end)
2392 struct vm_area_struct **insertion_point;
2393 struct vm_area_struct *tail_vma = NULL;
2396 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2397 vma->vm_prev = NULL;
2399 vma_rb_erase(vma, &mm->mm_rb);
2403 } while (vma && vma->vm_start < end);
2404 *insertion_point = vma;
2406 vma->vm_prev = prev;
2407 vma_gap_update(vma);
2409 mm->highest_vm_end = prev ? prev->vm_end : 0;
2410 tail_vma->vm_next = NULL;
2411 if (mm->unmap_area == arch_unmap_area)
2412 addr = prev ? prev->vm_end : mm->mmap_base;
2414 addr = vma ? vma->vm_start : mm->mmap_base;
2415 mm->unmap_area(mm, addr);
2416 mm->mmap_cache = NULL; /* Kill the cache. */
2420 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2421 * munmap path where it doesn't make sense to fail.
2423 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
2424 unsigned long addr, int new_below)
2426 struct mempolicy *pol;
2427 struct vm_area_struct *new;
2430 if (is_vm_hugetlb_page(vma) && (addr &
2431 ~(huge_page_mask(hstate_vma(vma)))))
2434 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2438 /* most fields are the same, copy all, and then fixup */
2441 INIT_LIST_HEAD(&new->anon_vma_chain);
2446 new->vm_start = addr;
2447 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2450 pol = mpol_dup(vma_policy(vma));
2455 vma_set_policy(new, pol);
2457 if (anon_vma_clone(new, vma))
2461 get_file(new->vm_file);
2463 if (new->vm_ops && new->vm_ops->open)
2464 new->vm_ops->open(new);
2467 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2468 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2470 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2476 /* Clean everything up if vma_adjust failed. */
2477 if (new->vm_ops && new->vm_ops->close)
2478 new->vm_ops->close(new);
2481 unlink_anon_vmas(new);
2485 kmem_cache_free(vm_area_cachep, new);
2491 * Split a vma into two pieces at address 'addr', a new vma is allocated
2492 * either for the first part or the tail.
2494 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2495 unsigned long addr, int new_below)
2497 if (mm->map_count >= sysctl_max_map_count)
2500 return __split_vma(mm, vma, addr, new_below);
2503 /* Munmap is split into 2 main parts -- this part which finds
2504 * what needs doing, and the areas themselves, which do the
2505 * work. This now handles partial unmappings.
2506 * Jeremy Fitzhardinge <jeremy@goop.org>
2508 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2511 struct vm_area_struct *vma, *prev, *last;
2513 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2516 if ((len = PAGE_ALIGN(len)) == 0)
2519 /* Find the first overlapping VMA */
2520 vma = find_vma(mm, start);
2523 prev = vma->vm_prev;
2524 /* we have start < vma->vm_end */
2526 /* if it doesn't overlap, we have nothing.. */
2528 if (vma->vm_start >= end)
2532 * If we need to split any vma, do it now to save pain later.
2534 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2535 * unmapped vm_area_struct will remain in use: so lower split_vma
2536 * places tmp vma above, and higher split_vma places tmp vma below.
2538 if (start > vma->vm_start) {
2542 * Make sure that map_count on return from munmap() will
2543 * not exceed its limit; but let map_count go just above
2544 * its limit temporarily, to help free resources as expected.
2546 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2549 error = __split_vma(mm, vma, start, 0);
2555 /* Does it split the last one? */
2556 last = find_vma(mm, end);
2557 if (last && end > last->vm_start) {
2558 int error = __split_vma(mm, last, end, 1);
2562 vma = prev? prev->vm_next: mm->mmap;
2565 * unlock any mlock()ed ranges before detaching vmas
2567 if (mm->locked_vm) {
2568 struct vm_area_struct *tmp = vma;
2569 while (tmp && tmp->vm_start < end) {
2570 if (tmp->vm_flags & VM_LOCKED) {
2571 mm->locked_vm -= vma_pages(tmp);
2572 munlock_vma_pages_all(tmp);
2579 * Remove the vma's, and unmap the actual pages
2581 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2582 unmap_region(mm, vma, prev, start, end);
2584 /* Fix up all other VM information */
2585 remove_vma_list(mm, vma);
2590 int vm_munmap(unsigned long start, size_t len)
2593 struct mm_struct *mm = current->mm;
2595 down_write(&mm->mmap_sem);
2596 ret = do_munmap(mm, start, len);
2597 up_write(&mm->mmap_sem);
2600 EXPORT_SYMBOL(vm_munmap);
2602 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2604 profile_munmap(addr);
2605 return vm_munmap(addr, len);
2608 static inline void verify_mm_writelocked(struct mm_struct *mm)
2610 #ifdef CONFIG_DEBUG_VM
2611 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2613 up_read(&mm->mmap_sem);
2619 * this is really a simplified "do_mmap". it only handles
2620 * anonymous maps. eventually we may be able to do some
2621 * brk-specific accounting here.
2623 static unsigned long do_brk(unsigned long addr, unsigned long len)
2625 struct mm_struct * mm = current->mm;
2626 struct vm_area_struct * vma, * prev;
2627 unsigned long flags;
2628 struct rb_node ** rb_link, * rb_parent;
2629 pgoff_t pgoff = addr >> PAGE_SHIFT;
2632 len = PAGE_ALIGN(len);
2636 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2638 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2639 if (error & ~PAGE_MASK)
2645 if (mm->def_flags & VM_LOCKED) {
2646 unsigned long locked, lock_limit;
2647 locked = len >> PAGE_SHIFT;
2648 locked += mm->locked_vm;
2649 lock_limit = rlimit(RLIMIT_MEMLOCK);
2650 lock_limit >>= PAGE_SHIFT;
2651 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2656 * mm->mmap_sem is required to protect against another thread
2657 * changing the mappings in case we sleep.
2659 verify_mm_writelocked(mm);
2662 * Clear old maps. this also does some error checking for us
2665 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
2666 if (do_munmap(mm, addr, len))
2671 /* Check against address space limits *after* clearing old maps... */
2672 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2675 if (mm->map_count > sysctl_max_map_count)
2678 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2681 /* Can we just expand an old private anonymous mapping? */
2682 vma = vma_merge(mm, prev, addr, addr + len, flags,
2683 NULL, NULL, pgoff, NULL, NULL);
2688 * create a vma struct for an anonymous mapping
2690 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2692 vm_unacct_memory(len >> PAGE_SHIFT);
2696 INIT_LIST_HEAD(&vma->anon_vma_chain);
2698 vma->vm_start = addr;
2699 vma->vm_end = addr + len;
2700 vma->vm_pgoff = pgoff;
2701 vma->vm_flags = flags;
2702 vma->vm_page_prot = vm_get_page_prot(flags);
2703 vma_link(mm, vma, prev, rb_link, rb_parent);
2705 perf_event_mmap(vma);
2706 mm->total_vm += len >> PAGE_SHIFT;
2707 if (flags & VM_LOCKED)
2708 mm->locked_vm += (len >> PAGE_SHIFT);
2712 unsigned long vm_brk(unsigned long addr, unsigned long len)
2714 struct mm_struct *mm = current->mm;
2718 down_write(&mm->mmap_sem);
2719 ret = do_brk(addr, len);
2720 populate = ((mm->def_flags & VM_LOCKED) != 0);
2721 up_write(&mm->mmap_sem);
2723 mm_populate(addr, len);
2726 EXPORT_SYMBOL(vm_brk);
2728 /* Release all mmaps. */
2729 void exit_mmap(struct mm_struct *mm)
2731 struct mmu_gather tlb;
2732 struct vm_area_struct *vma;
2733 unsigned long nr_accounted = 0;
2735 /* mm's last user has gone, and its about to be pulled down */
2736 mmu_notifier_release(mm);
2738 if (mm->locked_vm) {
2741 if (vma->vm_flags & VM_LOCKED)
2742 munlock_vma_pages_all(vma);
2750 if (!vma) /* Can happen if dup_mmap() received an OOM */
2755 tlb_gather_mmu(&tlb, mm, 0, -1);
2756 /* update_hiwater_rss(mm) here? but nobody should be looking */
2757 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2758 unmap_vmas(&tlb, vma, 0, -1);
2760 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
2761 tlb_finish_mmu(&tlb, 0, -1);
2764 * Walk the list again, actually closing and freeing it,
2765 * with preemption enabled, without holding any MM locks.
2768 if (vma->vm_flags & VM_ACCOUNT)
2769 nr_accounted += vma_pages(vma);
2770 vma = remove_vma(vma);
2772 vm_unacct_memory(nr_accounted);
2774 WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2777 /* Insert vm structure into process list sorted by address
2778 * and into the inode's i_mmap tree. If vm_file is non-NULL
2779 * then i_mmap_mutex is taken here.
2781 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
2783 struct vm_area_struct *prev;
2784 struct rb_node **rb_link, *rb_parent;
2787 * The vm_pgoff of a purely anonymous vma should be irrelevant
2788 * until its first write fault, when page's anon_vma and index
2789 * are set. But now set the vm_pgoff it will almost certainly
2790 * end up with (unless mremap moves it elsewhere before that
2791 * first wfault), so /proc/pid/maps tells a consistent story.
2793 * By setting it to reflect the virtual start address of the
2794 * vma, merges and splits can happen in a seamless way, just
2795 * using the existing file pgoff checks and manipulations.
2796 * Similarly in do_mmap_pgoff and in do_brk.
2798 if (!vma->vm_file) {
2799 BUG_ON(vma->anon_vma);
2800 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2802 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2803 &prev, &rb_link, &rb_parent))
2805 if ((vma->vm_flags & VM_ACCOUNT) &&
2806 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2809 vma_link(mm, vma, prev, rb_link, rb_parent);
2814 * Copy the vma structure to a new location in the same mm,
2815 * prior to moving page table entries, to effect an mremap move.
2817 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2818 unsigned long addr, unsigned long len, pgoff_t pgoff,
2819 bool *need_rmap_locks)
2821 struct vm_area_struct *vma = *vmap;
2822 unsigned long vma_start = vma->vm_start;
2823 struct mm_struct *mm = vma->vm_mm;
2824 struct vm_area_struct *new_vma, *prev;
2825 struct rb_node **rb_link, *rb_parent;
2826 struct mempolicy *pol;
2827 bool faulted_in_anon_vma = true;
2830 * If anonymous vma has not yet been faulted, update new pgoff
2831 * to match new location, to increase its chance of merging.
2833 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2834 pgoff = addr >> PAGE_SHIFT;
2835 faulted_in_anon_vma = false;
2838 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2839 return NULL; /* should never get here */
2840 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2841 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2842 vma_get_anon_name(vma));
2845 * Source vma may have been merged into new_vma
2847 if (unlikely(vma_start >= new_vma->vm_start &&
2848 vma_start < new_vma->vm_end)) {
2850 * The only way we can get a vma_merge with
2851 * self during an mremap is if the vma hasn't
2852 * been faulted in yet and we were allowed to
2853 * reset the dst vma->vm_pgoff to the
2854 * destination address of the mremap to allow
2855 * the merge to happen. mremap must change the
2856 * vm_pgoff linearity between src and dst vmas
2857 * (in turn preventing a vma_merge) to be
2858 * safe. It is only safe to keep the vm_pgoff
2859 * linear if there are no pages mapped yet.
2861 VM_BUG_ON(faulted_in_anon_vma);
2862 *vmap = vma = new_vma;
2864 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
2866 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2869 new_vma->vm_start = addr;
2870 new_vma->vm_end = addr + len;
2871 new_vma->vm_pgoff = pgoff;
2872 pol = mpol_dup(vma_policy(vma));
2875 vma_set_policy(new_vma, pol);
2876 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2877 if (anon_vma_clone(new_vma, vma))
2878 goto out_free_mempol;
2879 if (new_vma->vm_file)
2880 get_file(new_vma->vm_file);
2881 if (new_vma->vm_ops && new_vma->vm_ops->open)
2882 new_vma->vm_ops->open(new_vma);
2883 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2884 *need_rmap_locks = false;
2892 kmem_cache_free(vm_area_cachep, new_vma);
2897 * Return true if the calling process may expand its vm space by the passed
2900 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2902 unsigned long cur = mm->total_vm; /* pages */
2905 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2907 if (cur + npages > lim)
2913 static int special_mapping_fault(struct vm_area_struct *vma,
2914 struct vm_fault *vmf)
2917 struct page **pages;
2920 * special mappings have no vm_file, and in that case, the mm
2921 * uses vm_pgoff internally. So we have to subtract it from here.
2922 * We are allowed to do this because we are the mm; do not copy
2923 * this code into drivers!
2925 pgoff = vmf->pgoff - vma->vm_pgoff;
2927 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2931 struct page *page = *pages;
2937 return VM_FAULT_SIGBUS;
2941 * Having a close hook prevents vma merging regardless of flags.
2943 static void special_mapping_close(struct vm_area_struct *vma)
2947 static const struct vm_operations_struct special_mapping_vmops = {
2948 .close = special_mapping_close,
2949 .fault = special_mapping_fault,
2953 * Called with mm->mmap_sem held for writing.
2954 * Insert a new vma covering the given region, with the given flags.
2955 * Its pages are supplied by the given array of struct page *.
2956 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2957 * The region past the last page supplied will always produce SIGBUS.
2958 * The array pointer and the pages it points to are assumed to stay alive
2959 * for as long as this mapping might exist.
2961 int install_special_mapping(struct mm_struct *mm,
2962 unsigned long addr, unsigned long len,
2963 unsigned long vm_flags, struct page **pages)
2966 struct vm_area_struct *vma;
2968 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2969 if (unlikely(vma == NULL))
2972 INIT_LIST_HEAD(&vma->anon_vma_chain);
2974 vma->vm_start = addr;
2975 vma->vm_end = addr + len;
2977 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2978 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2980 vma->vm_ops = &special_mapping_vmops;
2981 vma->vm_private_data = pages;
2983 ret = insert_vm_struct(mm, vma);
2987 mm->total_vm += len >> PAGE_SHIFT;
2989 perf_event_mmap(vma);
2994 kmem_cache_free(vm_area_cachep, vma);
2998 static DEFINE_MUTEX(mm_all_locks_mutex);
3000 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3002 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3004 * The LSB of head.next can't change from under us
3005 * because we hold the mm_all_locks_mutex.
3007 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
3009 * We can safely modify head.next after taking the
3010 * anon_vma->root->rwsem. If some other vma in this mm shares
3011 * the same anon_vma we won't take it again.
3013 * No need of atomic instructions here, head.next
3014 * can't change from under us thanks to the
3015 * anon_vma->root->rwsem.
3017 if (__test_and_set_bit(0, (unsigned long *)
3018 &anon_vma->root->rb_root.rb_node))
3023 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3025 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3027 * AS_MM_ALL_LOCKS can't change from under us because
3028 * we hold the mm_all_locks_mutex.
3030 * Operations on ->flags have to be atomic because
3031 * even if AS_MM_ALL_LOCKS is stable thanks to the
3032 * mm_all_locks_mutex, there may be other cpus
3033 * changing other bitflags in parallel to us.
3035 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3037 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
3042 * This operation locks against the VM for all pte/vma/mm related
3043 * operations that could ever happen on a certain mm. This includes
3044 * vmtruncate, try_to_unmap, and all page faults.
3046 * The caller must take the mmap_sem in write mode before calling
3047 * mm_take_all_locks(). The caller isn't allowed to release the
3048 * mmap_sem until mm_drop_all_locks() returns.
3050 * mmap_sem in write mode is required in order to block all operations
3051 * that could modify pagetables and free pages without need of
3052 * altering the vma layout (for example populate_range() with
3053 * nonlinear vmas). It's also needed in write mode to avoid new
3054 * anon_vmas to be associated with existing vmas.
3056 * A single task can't take more than one mm_take_all_locks() in a row
3057 * or it would deadlock.
3059 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3060 * mapping->flags avoid to take the same lock twice, if more than one
3061 * vma in this mm is backed by the same anon_vma or address_space.
3063 * We can take all the locks in random order because the VM code
3064 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
3065 * takes more than one of them in a row. Secondly we're protected
3066 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3068 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3069 * that may have to take thousand of locks.
3071 * mm_take_all_locks() can fail if it's interrupted by signals.
3073 int mm_take_all_locks(struct mm_struct *mm)
3075 struct vm_area_struct *vma;
3076 struct anon_vma_chain *avc;
3078 BUG_ON(down_read_trylock(&mm->mmap_sem));
3080 mutex_lock(&mm_all_locks_mutex);
3082 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3083 if (signal_pending(current))
3085 if (vma->vm_file && vma->vm_file->f_mapping)
3086 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3089 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3090 if (signal_pending(current))
3093 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3094 vm_lock_anon_vma(mm, avc->anon_vma);
3100 mm_drop_all_locks(mm);
3104 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3106 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3108 * The LSB of head.next can't change to 0 from under
3109 * us because we hold the mm_all_locks_mutex.
3111 * We must however clear the bitflag before unlocking
3112 * the vma so the users using the anon_vma->rb_root will
3113 * never see our bitflag.
3115 * No need of atomic instructions here, head.next
3116 * can't change from under us until we release the
3117 * anon_vma->root->rwsem.
3119 if (!__test_and_clear_bit(0, (unsigned long *)
3120 &anon_vma->root->rb_root.rb_node))
3122 anon_vma_unlock_write(anon_vma);
3126 static void vm_unlock_mapping(struct address_space *mapping)
3128 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3130 * AS_MM_ALL_LOCKS can't change to 0 from under us
3131 * because we hold the mm_all_locks_mutex.
3133 mutex_unlock(&mapping->i_mmap_mutex);
3134 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3141 * The mmap_sem cannot be released by the caller until
3142 * mm_drop_all_locks() returns.
3144 void mm_drop_all_locks(struct mm_struct *mm)
3146 struct vm_area_struct *vma;
3147 struct anon_vma_chain *avc;
3149 BUG_ON(down_read_trylock(&mm->mmap_sem));
3150 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3152 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3154 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3155 vm_unlock_anon_vma(avc->anon_vma);
3156 if (vma->vm_file && vma->vm_file->f_mapping)
3157 vm_unlock_mapping(vma->vm_file->f_mapping);
3160 mutex_unlock(&mm_all_locks_mutex);
3164 * initialise the VMA slab
3166 void __init mmap_init(void)
3170 ret = percpu_counter_init(&vm_committed_as, 0);
3175 * Initialise sysctl_user_reserve_kbytes.
3177 * This is intended to prevent a user from starting a single memory hogging
3178 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3181 * The default value is min(3% of free memory, 128MB)
3182 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3184 static int init_user_reserve(void)
3186 unsigned long free_kbytes;
3188 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3190 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3193 module_init(init_user_reserve)
3196 * Initialise sysctl_admin_reserve_kbytes.
3198 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3199 * to log in and kill a memory hogging process.
3201 * Systems with more than 256MB will reserve 8MB, enough to recover
3202 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3203 * only reserve 3% of free pages by default.
3205 static int init_admin_reserve(void)
3207 unsigned long free_kbytes;
3209 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3211 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3214 module_init(init_admin_reserve)
3217 * Reinititalise user and admin reserves if memory is added or removed.
3219 * The default user reserve max is 128MB, and the default max for the
3220 * admin reserve is 8MB. These are usually, but not always, enough to
3221 * enable recovery from a memory hogging process using login/sshd, a shell,
3222 * and tools like top. It may make sense to increase or even disable the
3223 * reserve depending on the existence of swap or variations in the recovery
3224 * tools. So, the admin may have changed them.
3226 * If memory is added and the reserves have been eliminated or increased above
3227 * the default max, then we'll trust the admin.
3229 * If memory is removed and there isn't enough free memory, then we
3230 * need to reset the reserves.
3232 * Otherwise keep the reserve set by the admin.
3234 static int reserve_mem_notifier(struct notifier_block *nb,
3235 unsigned long action, void *data)
3237 unsigned long tmp, free_kbytes;
3241 /* Default max is 128MB. Leave alone if modified by operator. */
3242 tmp = sysctl_user_reserve_kbytes;
3243 if (0 < tmp && tmp < (1UL << 17))
3244 init_user_reserve();
3246 /* Default max is 8MB. Leave alone if modified by operator. */
3247 tmp = sysctl_admin_reserve_kbytes;
3248 if (0 < tmp && tmp < (1UL << 13))
3249 init_admin_reserve();
3253 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3255 if (sysctl_user_reserve_kbytes > free_kbytes) {
3256 init_user_reserve();
3257 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3258 sysctl_user_reserve_kbytes);
3261 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3262 init_admin_reserve();
3263 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3264 sysctl_admin_reserve_kbytes);
3273 static struct notifier_block reserve_mem_nb = {
3274 .notifier_call = reserve_mem_notifier,
3277 static int __meminit init_reserve_notifier(void)
3279 if (register_hotmemory_notifier(&reserve_mem_nb))
3280 printk("Failed registering memory add/remove notifier for admin reserve");
3284 module_init(init_reserve_notifier)