6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/backing-dev.h>
15 #include <linux/vmacache.h>
16 #include <linux/shm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/syscalls.h>
21 #include <linux/capability.h>
22 #include <linux/init.h>
23 #include <linux/file.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/hugetlb.h>
28 #include <linux/profile.h>
29 #include <linux/export.h>
30 #include <linux/mount.h>
31 #include <linux/mempolicy.h>
32 #include <linux/rmap.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/mmdebug.h>
35 #include <linux/perf_event.h>
36 #include <linux/audit.h>
37 #include <linux/khugepaged.h>
38 #include <linux/uprobes.h>
39 #include <linux/rbtree_augmented.h>
40 #include <linux/sched/sysctl.h>
41 #include <linux/notifier.h>
42 #include <linux/memory.h>
43 #include <linux/printk.h>
45 #include <asm/uaccess.h>
46 #include <asm/cacheflush.h>
48 #include <asm/mmu_context.h>
52 #ifndef arch_mmap_check
53 #define arch_mmap_check(addr, len, flags) (0)
56 #ifndef arch_rebalance_pgtables
57 #define arch_rebalance_pgtables(addr, len) (addr)
60 static void unmap_region(struct mm_struct *mm,
61 struct vm_area_struct *vma, struct vm_area_struct *prev,
62 unsigned long start, unsigned long end);
64 /* description of effects of mapping type and prot in current implementation.
65 * this is due to the limited x86 page protection hardware. The expected
66 * behavior is in parens:
69 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
70 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (yes) yes w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
74 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
75 * w: (no) no w: (no) no w: (copy) copy w: (no) no
76 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
79 pgprot_t protection_map[16] = {
80 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
81 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
84 pgprot_t vm_get_page_prot(unsigned long vm_flags)
86 return __pgprot(pgprot_val(protection_map[vm_flags &
87 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
88 pgprot_val(arch_vm_get_page_prot(vm_flags)));
90 EXPORT_SYMBOL(vm_get_page_prot);
92 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
93 int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
94 unsigned long sysctl_overcommit_kbytes __read_mostly;
95 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
96 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
97 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
99 * Make sure vm_committed_as in one cacheline and not cacheline shared with
100 * other variables. It can be updated by several CPUs frequently.
102 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
105 * The global memory commitment made in the system can be a metric
106 * that can be used to drive ballooning decisions when Linux is hosted
107 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
108 * balancing memory across competing virtual machines that are hosted.
109 * Several metrics drive this policy engine including the guest reported
112 unsigned long vm_memory_committed(void)
114 return percpu_counter_read_positive(&vm_committed_as);
116 EXPORT_SYMBOL_GPL(vm_memory_committed);
119 * Check that a process has enough memory to allocate a new virtual
120 * mapping. 0 means there is enough memory for the allocation to
121 * succeed and -ENOMEM implies there is not.
123 * We currently support three overcommit policies, which are set via the
124 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
126 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
127 * Additional code 2002 Jul 20 by Robert Love.
129 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
131 * Note this is a helper function intended to be used by LSMs which
132 * wish to use this logic.
134 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
136 unsigned long free, allowed, reserve;
138 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
139 -(s64)vm_committed_as_batch * num_online_cpus(),
140 "memory commitment underflow");
142 vm_acct_memory(pages);
145 * Sometimes we want to use more memory than we have
147 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
150 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
151 free = global_page_state(NR_FREE_PAGES);
152 free += global_page_state(NR_FILE_PAGES);
155 * shmem pages shouldn't be counted as free in this
156 * case, they can't be purged, only swapped out, and
157 * that won't affect the overall amount of available
158 * memory in the system.
160 free -= global_page_state(NR_SHMEM);
162 free += get_nr_swap_pages();
165 * Any slabs which are created with the
166 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
167 * which are reclaimable, under pressure. The dentry
168 * cache and most inode caches should fall into this
170 free += global_page_state(NR_SLAB_RECLAIMABLE);
173 * Leave reserved pages. The pages are not for anonymous pages.
175 if (free <= totalreserve_pages)
178 free -= totalreserve_pages;
181 * Reserve some for root
184 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
192 allowed = vm_commit_limit();
194 * Reserve some for root
197 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
200 * Don't let a single process grow so big a user can't recover
203 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
204 allowed -= min(mm->total_vm / 32, reserve);
207 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
210 vm_unacct_memory(pages);
216 * Requires inode->i_mapping->i_mmap_mutex
218 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
219 struct file *file, struct address_space *mapping)
221 if (vma->vm_flags & VM_DENYWRITE)
222 atomic_inc(&file_inode(file)->i_writecount);
223 if (vma->vm_flags & VM_SHARED)
224 mapping_unmap_writable(mapping);
226 flush_dcache_mmap_lock(mapping);
227 if (unlikely(vma->vm_flags & VM_NONLINEAR))
228 list_del_init(&vma->shared.nonlinear);
230 vma_interval_tree_remove(vma, &mapping->i_mmap);
231 flush_dcache_mmap_unlock(mapping);
235 * Unlink a file-based vm structure from its interval tree, to hide
236 * vma from rmap and vmtruncate before freeing its page tables.
238 void unlink_file_vma(struct vm_area_struct *vma)
240 struct file *file = vma->vm_file;
243 struct address_space *mapping = file->f_mapping;
244 mutex_lock(&mapping->i_mmap_mutex);
245 __remove_shared_vm_struct(vma, file, mapping);
246 mutex_unlock(&mapping->i_mmap_mutex);
251 * Close a vm structure and free it, returning the next.
253 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
255 struct vm_area_struct *next = vma->vm_next;
258 if (vma->vm_ops && vma->vm_ops->close)
259 vma->vm_ops->close(vma);
262 mpol_put(vma_policy(vma));
263 kmem_cache_free(vm_area_cachep, vma);
267 static unsigned long do_brk(unsigned long addr, unsigned long len);
269 SYSCALL_DEFINE1(brk, unsigned long, brk)
271 unsigned long retval;
272 unsigned long newbrk, oldbrk;
273 struct mm_struct *mm = current->mm;
274 unsigned long min_brk;
277 down_write(&mm->mmap_sem);
279 #ifdef CONFIG_COMPAT_BRK
281 * CONFIG_COMPAT_BRK can still be overridden by setting
282 * randomize_va_space to 2, which will still cause mm->start_brk
283 * to be arbitrarily shifted
285 if (current->brk_randomized)
286 min_brk = mm->start_brk;
288 min_brk = mm->end_data;
290 min_brk = mm->start_brk;
296 * Check against rlimit here. If this check is done later after the test
297 * of oldbrk with newbrk then it can escape the test and let the data
298 * segment grow beyond its set limit the in case where the limit is
299 * not page aligned -Ram Gupta
301 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
302 mm->end_data, mm->start_data))
305 newbrk = PAGE_ALIGN(brk);
306 oldbrk = PAGE_ALIGN(mm->brk);
307 if (oldbrk == newbrk)
310 /* Always allow shrinking brk. */
311 if (brk <= mm->brk) {
312 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
317 /* Check against existing mmap mappings. */
318 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
321 /* Ok, looks good - let it rip. */
322 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
327 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
328 up_write(&mm->mmap_sem);
330 mm_populate(oldbrk, newbrk - oldbrk);
335 up_write(&mm->mmap_sem);
339 static long vma_compute_subtree_gap(struct vm_area_struct *vma)
341 unsigned long max, subtree_gap;
344 max -= vma->vm_prev->vm_end;
345 if (vma->vm_rb.rb_left) {
346 subtree_gap = rb_entry(vma->vm_rb.rb_left,
347 struct vm_area_struct, vm_rb)->rb_subtree_gap;
348 if (subtree_gap > max)
351 if (vma->vm_rb.rb_right) {
352 subtree_gap = rb_entry(vma->vm_rb.rb_right,
353 struct vm_area_struct, vm_rb)->rb_subtree_gap;
354 if (subtree_gap > max)
360 #ifdef CONFIG_DEBUG_VM_RB
361 static int browse_rb(struct rb_root *root)
363 int i = 0, j, bug = 0;
364 struct rb_node *nd, *pn = NULL;
365 unsigned long prev = 0, pend = 0;
367 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
368 struct vm_area_struct *vma;
369 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
370 if (vma->vm_start < prev) {
371 pr_emerg("vm_start %lx prev %lx\n", vma->vm_start, prev);
374 if (vma->vm_start < pend) {
375 pr_emerg("vm_start %lx pend %lx\n", vma->vm_start, pend);
378 if (vma->vm_start > vma->vm_end) {
379 pr_emerg("vm_end %lx < vm_start %lx\n",
380 vma->vm_end, vma->vm_start);
383 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
384 pr_emerg("free gap %lx, correct %lx\n",
386 vma_compute_subtree_gap(vma));
391 prev = vma->vm_start;
395 for (nd = pn; nd; nd = rb_prev(nd))
398 pr_emerg("backwards %d, forwards %d\n", j, i);
404 static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
408 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
409 struct vm_area_struct *vma;
410 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
411 BUG_ON(vma != ignore &&
412 vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
416 static void validate_mm(struct mm_struct *mm)
420 unsigned long highest_address = 0;
421 struct vm_area_struct *vma = mm->mmap;
423 struct anon_vma_chain *avc;
424 vma_lock_anon_vma(vma);
425 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
426 anon_vma_interval_tree_verify(avc);
427 vma_unlock_anon_vma(vma);
428 highest_address = vma->vm_end;
432 if (i != mm->map_count) {
433 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
436 if (highest_address != mm->highest_vm_end) {
437 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
438 mm->highest_vm_end, highest_address);
441 i = browse_rb(&mm->mm_rb);
442 if (i != mm->map_count) {
443 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
449 #define validate_mm_rb(root, ignore) do { } while (0)
450 #define validate_mm(mm) do { } while (0)
453 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
454 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
457 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
458 * vma->vm_prev->vm_end values changed, without modifying the vma's position
461 static void vma_gap_update(struct vm_area_struct *vma)
464 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
465 * function that does exacltly what we want.
467 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
470 static inline void vma_rb_insert(struct vm_area_struct *vma,
471 struct rb_root *root)
473 /* All rb_subtree_gap values must be consistent prior to insertion */
474 validate_mm_rb(root, NULL);
476 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
479 static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
482 * All rb_subtree_gap values must be consistent prior to erase,
483 * with the possible exception of the vma being erased.
485 validate_mm_rb(root, vma);
488 * Note rb_erase_augmented is a fairly large inline function,
489 * so make sure we instantiate it only once with our desired
490 * augmented rbtree callbacks.
492 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
496 * vma has some anon_vma assigned, and is already inserted on that
497 * anon_vma's interval trees.
499 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
500 * vma must be removed from the anon_vma's interval trees using
501 * anon_vma_interval_tree_pre_update_vma().
503 * After the update, the vma will be reinserted using
504 * anon_vma_interval_tree_post_update_vma().
506 * The entire update must be protected by exclusive mmap_sem and by
507 * the root anon_vma's mutex.
510 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
512 struct anon_vma_chain *avc;
514 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
515 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
519 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
521 struct anon_vma_chain *avc;
523 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
524 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
527 static int find_vma_links(struct mm_struct *mm, unsigned long addr,
528 unsigned long end, struct vm_area_struct **pprev,
529 struct rb_node ***rb_link, struct rb_node **rb_parent)
531 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
533 __rb_link = &mm->mm_rb.rb_node;
534 rb_prev = __rb_parent = NULL;
537 struct vm_area_struct *vma_tmp;
539 __rb_parent = *__rb_link;
540 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
542 if (vma_tmp->vm_end > addr) {
543 /* Fail if an existing vma overlaps the area */
544 if (vma_tmp->vm_start < end)
546 __rb_link = &__rb_parent->rb_left;
548 rb_prev = __rb_parent;
549 __rb_link = &__rb_parent->rb_right;
555 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
556 *rb_link = __rb_link;
557 *rb_parent = __rb_parent;
561 static unsigned long count_vma_pages_range(struct mm_struct *mm,
562 unsigned long addr, unsigned long end)
564 unsigned long nr_pages = 0;
565 struct vm_area_struct *vma;
567 /* Find first overlaping mapping */
568 vma = find_vma_intersection(mm, addr, end);
572 nr_pages = (min(end, vma->vm_end) -
573 max(addr, vma->vm_start)) >> PAGE_SHIFT;
575 /* Iterate over the rest of the overlaps */
576 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
577 unsigned long overlap_len;
579 if (vma->vm_start > end)
582 overlap_len = min(end, vma->vm_end) - vma->vm_start;
583 nr_pages += overlap_len >> PAGE_SHIFT;
589 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
590 struct rb_node **rb_link, struct rb_node *rb_parent)
592 /* Update tracking information for the gap following the new vma. */
594 vma_gap_update(vma->vm_next);
596 mm->highest_vm_end = vma->vm_end;
599 * vma->vm_prev wasn't known when we followed the rbtree to find the
600 * correct insertion point for that vma. As a result, we could not
601 * update the vma vm_rb parents rb_subtree_gap values on the way down.
602 * So, we first insert the vma with a zero rb_subtree_gap value
603 * (to be consistent with what we did on the way down), and then
604 * immediately update the gap to the correct value. Finally we
605 * rebalance the rbtree after all augmented values have been set.
607 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
608 vma->rb_subtree_gap = 0;
610 vma_rb_insert(vma, &mm->mm_rb);
613 static void __vma_link_file(struct vm_area_struct *vma)
619 struct address_space *mapping = file->f_mapping;
621 if (vma->vm_flags & VM_DENYWRITE)
622 atomic_dec(&file_inode(file)->i_writecount);
623 if (vma->vm_flags & VM_SHARED)
624 atomic_inc(&mapping->i_mmap_writable);
626 flush_dcache_mmap_lock(mapping);
627 if (unlikely(vma->vm_flags & VM_NONLINEAR))
628 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
630 vma_interval_tree_insert(vma, &mapping->i_mmap);
631 flush_dcache_mmap_unlock(mapping);
636 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
637 struct vm_area_struct *prev, struct rb_node **rb_link,
638 struct rb_node *rb_parent)
640 __vma_link_list(mm, vma, prev, rb_parent);
641 __vma_link_rb(mm, vma, rb_link, rb_parent);
644 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
645 struct vm_area_struct *prev, struct rb_node **rb_link,
646 struct rb_node *rb_parent)
648 struct address_space *mapping = NULL;
651 mapping = vma->vm_file->f_mapping;
652 mutex_lock(&mapping->i_mmap_mutex);
655 __vma_link(mm, vma, prev, rb_link, rb_parent);
656 __vma_link_file(vma);
659 mutex_unlock(&mapping->i_mmap_mutex);
666 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
667 * mm's list and rbtree. It has already been inserted into the interval tree.
669 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
671 struct vm_area_struct *prev;
672 struct rb_node **rb_link, *rb_parent;
674 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
675 &prev, &rb_link, &rb_parent))
677 __vma_link(mm, vma, prev, rb_link, rb_parent);
682 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
683 struct vm_area_struct *prev)
685 struct vm_area_struct *next;
687 vma_rb_erase(vma, &mm->mm_rb);
688 prev->vm_next = next = vma->vm_next;
690 next->vm_prev = prev;
693 vmacache_invalidate(mm);
697 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
698 * is already present in an i_mmap tree without adjusting the tree.
699 * The following helper function should be used when such adjustments
700 * are necessary. The "insert" vma (if any) is to be inserted
701 * before we drop the necessary locks.
703 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
704 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
706 struct mm_struct *mm = vma->vm_mm;
707 struct vm_area_struct *next = vma->vm_next;
708 struct vm_area_struct *importer = NULL;
709 struct address_space *mapping = NULL;
710 struct rb_root *root = NULL;
711 struct anon_vma *anon_vma = NULL;
712 struct file *file = vma->vm_file;
713 bool start_changed = false, end_changed = false;
714 long adjust_next = 0;
717 if (next && !insert) {
718 struct vm_area_struct *exporter = NULL;
720 if (end >= next->vm_end) {
722 * vma expands, overlapping all the next, and
723 * perhaps the one after too (mprotect case 6).
725 again: remove_next = 1 + (end > next->vm_end);
729 } else if (end > next->vm_start) {
731 * vma expands, overlapping part of the next:
732 * mprotect case 5 shifting the boundary up.
734 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
737 } else if (end < vma->vm_end) {
739 * vma shrinks, and !insert tells it's not
740 * split_vma inserting another: so it must be
741 * mprotect case 4 shifting the boundary down.
743 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
749 * Easily overlooked: when mprotect shifts the boundary,
750 * make sure the expanding vma has anon_vma set if the
751 * shrinking vma had, to cover any anon pages imported.
753 if (exporter && exporter->anon_vma && !importer->anon_vma) {
754 if (anon_vma_clone(importer, exporter))
756 importer->anon_vma = exporter->anon_vma;
761 mapping = file->f_mapping;
762 if (!(vma->vm_flags & VM_NONLINEAR)) {
763 root = &mapping->i_mmap;
764 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
767 uprobe_munmap(next, next->vm_start,
771 mutex_lock(&mapping->i_mmap_mutex);
774 * Put into interval tree now, so instantiated pages
775 * are visible to arm/parisc __flush_dcache_page
776 * throughout; but we cannot insert into address
777 * space until vma start or end is updated.
779 __vma_link_file(insert);
783 vma_adjust_trans_huge(vma, start, end, adjust_next);
785 anon_vma = vma->anon_vma;
786 if (!anon_vma && adjust_next)
787 anon_vma = next->anon_vma;
789 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
790 anon_vma != next->anon_vma, next);
791 anon_vma_lock_write(anon_vma);
792 anon_vma_interval_tree_pre_update_vma(vma);
794 anon_vma_interval_tree_pre_update_vma(next);
798 flush_dcache_mmap_lock(mapping);
799 vma_interval_tree_remove(vma, root);
801 vma_interval_tree_remove(next, root);
804 if (start != vma->vm_start) {
805 vma->vm_start = start;
806 start_changed = true;
808 if (end != vma->vm_end) {
812 vma->vm_pgoff = pgoff;
814 next->vm_start += adjust_next << PAGE_SHIFT;
815 next->vm_pgoff += adjust_next;
820 vma_interval_tree_insert(next, root);
821 vma_interval_tree_insert(vma, root);
822 flush_dcache_mmap_unlock(mapping);
827 * vma_merge has merged next into vma, and needs
828 * us to remove next before dropping the locks.
830 __vma_unlink(mm, next, vma);
832 __remove_shared_vm_struct(next, file, mapping);
835 * split_vma has split insert from vma, and needs
836 * us to insert it before dropping the locks
837 * (it may either follow vma or precede it).
839 __insert_vm_struct(mm, insert);
845 mm->highest_vm_end = end;
846 else if (!adjust_next)
847 vma_gap_update(next);
852 anon_vma_interval_tree_post_update_vma(vma);
854 anon_vma_interval_tree_post_update_vma(next);
855 anon_vma_unlock_write(anon_vma);
858 mutex_unlock(&mapping->i_mmap_mutex);
869 uprobe_munmap(next, next->vm_start, next->vm_end);
873 anon_vma_merge(vma, next);
875 mpol_put(vma_policy(next));
876 kmem_cache_free(vm_area_cachep, next);
878 * In mprotect's case 6 (see comments on vma_merge),
879 * we must remove another next too. It would clutter
880 * up the code too much to do both in one go.
883 if (remove_next == 2)
886 vma_gap_update(next);
888 mm->highest_vm_end = end;
899 * If the vma has a ->close operation then the driver probably needs to release
900 * per-vma resources, so we don't attempt to merge those.
902 static inline int is_mergeable_vma(struct vm_area_struct *vma,
903 struct file *file, unsigned long vm_flags)
906 * VM_SOFTDIRTY should not prevent from VMA merging, if we
907 * match the flags but dirty bit -- the caller should mark
908 * merged VMA as dirty. If dirty bit won't be excluded from
909 * comparison, we increase pressue on the memory system forcing
910 * the kernel to generate new VMAs when old one could be
913 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
915 if (vma->vm_file != file)
917 if (vma->vm_ops && vma->vm_ops->close)
922 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
923 struct anon_vma *anon_vma2,
924 struct vm_area_struct *vma)
927 * The list_is_singular() test is to avoid merging VMA cloned from
928 * parents. This can improve scalability caused by anon_vma lock.
930 if ((!anon_vma1 || !anon_vma2) && (!vma ||
931 list_is_singular(&vma->anon_vma_chain)))
933 return anon_vma1 == anon_vma2;
937 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
938 * in front of (at a lower virtual address and file offset than) the vma.
940 * We cannot merge two vmas if they have differently assigned (non-NULL)
941 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
943 * We don't check here for the merged mmap wrapping around the end of pagecache
944 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
945 * wrap, nor mmaps which cover the final page at index -1UL.
948 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
949 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
951 if (is_mergeable_vma(vma, file, vm_flags) &&
952 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
953 if (vma->vm_pgoff == vm_pgoff)
960 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
961 * beyond (at a higher virtual address and file offset than) the vma.
963 * We cannot merge two vmas if they have differently assigned (non-NULL)
964 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
967 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
968 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
970 if (is_mergeable_vma(vma, file, vm_flags) &&
971 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
973 vm_pglen = vma_pages(vma);
974 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
981 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
982 * whether that can be merged with its predecessor or its successor.
983 * Or both (it neatly fills a hole).
985 * In most cases - when called for mmap, brk or mremap - [addr,end) is
986 * certain not to be mapped by the time vma_merge is called; but when
987 * called for mprotect, it is certain to be already mapped (either at
988 * an offset within prev, or at the start of next), and the flags of
989 * this area are about to be changed to vm_flags - and the no-change
990 * case has already been eliminated.
992 * The following mprotect cases have to be considered, where AAAA is
993 * the area passed down from mprotect_fixup, never extending beyond one
994 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
996 * AAAA AAAA AAAA AAAA
997 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
998 * cannot merge might become might become might become
999 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1000 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1001 * mremap move: PPPPNNNNNNNN 8
1003 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1004 * might become case 1 below case 2 below case 3 below
1006 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1007 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1009 struct vm_area_struct *vma_merge(struct mm_struct *mm,
1010 struct vm_area_struct *prev, unsigned long addr,
1011 unsigned long end, unsigned long vm_flags,
1012 struct anon_vma *anon_vma, struct file *file,
1013 pgoff_t pgoff, struct mempolicy *policy)
1015 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1016 struct vm_area_struct *area, *next;
1020 * We later require that vma->vm_flags == vm_flags,
1021 * so this tests vma->vm_flags & VM_SPECIAL, too.
1023 if (vm_flags & VM_SPECIAL)
1027 next = prev->vm_next;
1031 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1032 next = next->vm_next;
1035 * Can it merge with the predecessor?
1037 if (prev && prev->vm_end == addr &&
1038 mpol_equal(vma_policy(prev), policy) &&
1039 can_vma_merge_after(prev, vm_flags,
1040 anon_vma, file, pgoff)) {
1042 * OK, it can. Can we now merge in the successor as well?
1044 if (next && end == next->vm_start &&
1045 mpol_equal(policy, vma_policy(next)) &&
1046 can_vma_merge_before(next, vm_flags,
1047 anon_vma, file, pgoff+pglen) &&
1048 is_mergeable_anon_vma(prev->anon_vma,
1049 next->anon_vma, NULL)) {
1051 err = vma_adjust(prev, prev->vm_start,
1052 next->vm_end, prev->vm_pgoff, NULL);
1053 } else /* cases 2, 5, 7 */
1054 err = vma_adjust(prev, prev->vm_start,
1055 end, prev->vm_pgoff, NULL);
1058 khugepaged_enter_vma_merge(prev);
1063 * Can this new request be merged in front of next?
1065 if (next && end == next->vm_start &&
1066 mpol_equal(policy, vma_policy(next)) &&
1067 can_vma_merge_before(next, vm_flags,
1068 anon_vma, file, pgoff+pglen)) {
1069 if (prev && addr < prev->vm_end) /* case 4 */
1070 err = vma_adjust(prev, prev->vm_start,
1071 addr, prev->vm_pgoff, NULL);
1072 else /* cases 3, 8 */
1073 err = vma_adjust(area, addr, next->vm_end,
1074 next->vm_pgoff - pglen, NULL);
1077 khugepaged_enter_vma_merge(area);
1085 * Rough compatbility check to quickly see if it's even worth looking
1086 * at sharing an anon_vma.
1088 * They need to have the same vm_file, and the flags can only differ
1089 * in things that mprotect may change.
1091 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1092 * we can merge the two vma's. For example, we refuse to merge a vma if
1093 * there is a vm_ops->close() function, because that indicates that the
1094 * driver is doing some kind of reference counting. But that doesn't
1095 * really matter for the anon_vma sharing case.
1097 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1099 return a->vm_end == b->vm_start &&
1100 mpol_equal(vma_policy(a), vma_policy(b)) &&
1101 a->vm_file == b->vm_file &&
1102 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
1103 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1107 * Do some basic sanity checking to see if we can re-use the anon_vma
1108 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1109 * the same as 'old', the other will be the new one that is trying
1110 * to share the anon_vma.
1112 * NOTE! This runs with mm_sem held for reading, so it is possible that
1113 * the anon_vma of 'old' is concurrently in the process of being set up
1114 * by another page fault trying to merge _that_. But that's ok: if it
1115 * is being set up, that automatically means that it will be a singleton
1116 * acceptable for merging, so we can do all of this optimistically. But
1117 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1119 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1120 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1121 * is to return an anon_vma that is "complex" due to having gone through
1124 * We also make sure that the two vma's are compatible (adjacent,
1125 * and with the same memory policies). That's all stable, even with just
1126 * a read lock on the mm_sem.
1128 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1130 if (anon_vma_compatible(a, b)) {
1131 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1133 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1140 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1141 * neighbouring vmas for a suitable anon_vma, before it goes off
1142 * to allocate a new anon_vma. It checks because a repetitive
1143 * sequence of mprotects and faults may otherwise lead to distinct
1144 * anon_vmas being allocated, preventing vma merge in subsequent
1147 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1149 struct anon_vma *anon_vma;
1150 struct vm_area_struct *near;
1152 near = vma->vm_next;
1156 anon_vma = reusable_anon_vma(near, vma, near);
1160 near = vma->vm_prev;
1164 anon_vma = reusable_anon_vma(near, near, vma);
1169 * There's no absolute need to look only at touching neighbours:
1170 * we could search further afield for "compatible" anon_vmas.
1171 * But it would probably just be a waste of time searching,
1172 * or lead to too many vmas hanging off the same anon_vma.
1173 * We're trying to allow mprotect remerging later on,
1174 * not trying to minimize memory used for anon_vmas.
1179 #ifdef CONFIG_PROC_FS
1180 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
1181 struct file *file, long pages)
1183 const unsigned long stack_flags
1184 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1186 mm->total_vm += pages;
1189 mm->shared_vm += pages;
1190 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1191 mm->exec_vm += pages;
1192 } else if (flags & stack_flags)
1193 mm->stack_vm += pages;
1195 #endif /* CONFIG_PROC_FS */
1198 * If a hint addr is less than mmap_min_addr change hint to be as
1199 * low as possible but still greater than mmap_min_addr
1201 static inline unsigned long round_hint_to_min(unsigned long hint)
1204 if (((void *)hint != NULL) &&
1205 (hint < mmap_min_addr))
1206 return PAGE_ALIGN(mmap_min_addr);
1210 static inline int mlock_future_check(struct mm_struct *mm,
1211 unsigned long flags,
1214 unsigned long locked, lock_limit;
1216 /* mlock MCL_FUTURE? */
1217 if (flags & VM_LOCKED) {
1218 locked = len >> PAGE_SHIFT;
1219 locked += mm->locked_vm;
1220 lock_limit = rlimit(RLIMIT_MEMLOCK);
1221 lock_limit >>= PAGE_SHIFT;
1222 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1229 * The caller must hold down_write(¤t->mm->mmap_sem).
1232 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
1233 unsigned long len, unsigned long prot,
1234 unsigned long flags, unsigned long pgoff,
1235 unsigned long *populate)
1237 struct mm_struct *mm = current->mm;
1238 vm_flags_t vm_flags;
1243 * Does the application expect PROT_READ to imply PROT_EXEC?
1245 * (the exception is when the underlying filesystem is noexec
1246 * mounted, in which case we dont add PROT_EXEC.)
1248 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1249 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
1255 if (!(flags & MAP_FIXED))
1256 addr = round_hint_to_min(addr);
1258 /* Careful about overflows.. */
1259 len = PAGE_ALIGN(len);
1263 /* offset overflow? */
1264 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1267 /* Too many mappings? */
1268 if (mm->map_count > sysctl_max_map_count)
1271 /* Obtain the address to map to. we verify (or select) it and ensure
1272 * that it represents a valid section of the address space.
1274 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1275 if (addr & ~PAGE_MASK)
1278 /* Do simple checking here so the lower-level routines won't have
1279 * to. we assume access permissions have been handled by the open
1280 * of the memory object, so we don't do any here.
1282 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1283 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1285 if (flags & MAP_LOCKED)
1286 if (!can_do_mlock())
1289 if (mlock_future_check(mm, vm_flags, len))
1293 struct inode *inode = file_inode(file);
1295 switch (flags & MAP_TYPE) {
1297 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1301 * Make sure we don't allow writing to an append-only
1304 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1308 * Make sure there are no mandatory locks on the file.
1310 if (locks_verify_locked(file))
1313 vm_flags |= VM_SHARED | VM_MAYSHARE;
1314 if (!(file->f_mode & FMODE_WRITE))
1315 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1319 if (!(file->f_mode & FMODE_READ))
1321 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1322 if (vm_flags & VM_EXEC)
1324 vm_flags &= ~VM_MAYEXEC;
1327 if (!file->f_op->mmap)
1329 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1337 switch (flags & MAP_TYPE) {
1339 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1345 vm_flags |= VM_SHARED | VM_MAYSHARE;
1349 * Set pgoff according to addr for anon_vma.
1351 pgoff = addr >> PAGE_SHIFT;
1359 * Set 'VM_NORESERVE' if we should not account for the
1360 * memory use of this mapping.
1362 if (flags & MAP_NORESERVE) {
1363 /* We honor MAP_NORESERVE if allowed to overcommit */
1364 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1365 vm_flags |= VM_NORESERVE;
1367 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1368 if (file && is_file_hugepages(file))
1369 vm_flags |= VM_NORESERVE;
1372 addr = mmap_region(file, addr, len, vm_flags, pgoff);
1373 if (!IS_ERR_VALUE(addr) &&
1374 ((vm_flags & VM_LOCKED) ||
1375 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1380 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1381 unsigned long, prot, unsigned long, flags,
1382 unsigned long, fd, unsigned long, pgoff)
1384 struct file *file = NULL;
1385 unsigned long retval = -EBADF;
1387 if (!(flags & MAP_ANONYMOUS)) {
1388 audit_mmap_fd(fd, flags);
1392 if (is_file_hugepages(file))
1393 len = ALIGN(len, huge_page_size(hstate_file(file)));
1395 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1397 } else if (flags & MAP_HUGETLB) {
1398 struct user_struct *user = NULL;
1401 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
1405 len = ALIGN(len, huge_page_size(hs));
1407 * VM_NORESERVE is used because the reservations will be
1408 * taken when vm_ops->mmap() is called
1409 * A dummy user value is used because we are not locking
1410 * memory so no accounting is necessary
1412 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1414 &user, HUGETLB_ANONHUGE_INODE,
1415 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1417 return PTR_ERR(file);
1420 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1422 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1430 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1431 struct mmap_arg_struct {
1435 unsigned long flags;
1437 unsigned long offset;
1440 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1442 struct mmap_arg_struct a;
1444 if (copy_from_user(&a, arg, sizeof(a)))
1446 if (a.offset & ~PAGE_MASK)
1449 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1450 a.offset >> PAGE_SHIFT);
1452 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1455 * Some shared mappigns will want the pages marked read-only
1456 * to track write events. If so, we'll downgrade vm_page_prot
1457 * to the private version (using protection_map[] without the
1460 int vma_wants_writenotify(struct vm_area_struct *vma)
1462 vm_flags_t vm_flags = vma->vm_flags;
1464 /* If it was private or non-writable, the write bit is already clear */
1465 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1468 /* The backer wishes to know when pages are first written to? */
1469 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1472 /* The open routine did something to the protections already? */
1473 if (pgprot_val(vma->vm_page_prot) !=
1474 pgprot_val(vm_get_page_prot(vm_flags)))
1477 /* Specialty mapping? */
1478 if (vm_flags & VM_PFNMAP)
1481 /* Can the mapping track the dirty pages? */
1482 return vma->vm_file && vma->vm_file->f_mapping &&
1483 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1487 * We account for memory if it's a private writeable mapping,
1488 * not hugepages and VM_NORESERVE wasn't set.
1490 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1493 * hugetlb has its own accounting separate from the core VM
1494 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1496 if (file && is_file_hugepages(file))
1499 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1502 unsigned long mmap_region(struct file *file, unsigned long addr,
1503 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
1505 struct mm_struct *mm = current->mm;
1506 struct vm_area_struct *vma, *prev;
1508 struct rb_node **rb_link, *rb_parent;
1509 unsigned long charged = 0;
1511 /* Check against address space limit. */
1512 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1513 unsigned long nr_pages;
1516 * MAP_FIXED may remove pages of mappings that intersects with
1517 * requested mapping. Account for the pages it would unmap.
1519 if (!(vm_flags & MAP_FIXED))
1522 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1524 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1528 /* Clear old maps */
1531 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
1532 if (do_munmap(mm, addr, len))
1538 * Private writable mapping: check memory availability
1540 if (accountable_mapping(file, vm_flags)) {
1541 charged = len >> PAGE_SHIFT;
1542 if (security_vm_enough_memory_mm(mm, charged))
1544 vm_flags |= VM_ACCOUNT;
1548 * Can we just expand an old mapping?
1550 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1555 * Determine the object being mapped and call the appropriate
1556 * specific mapper. the address has already been validated, but
1557 * not unmapped, but the maps are removed from the list.
1559 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1566 vma->vm_start = addr;
1567 vma->vm_end = addr + len;
1568 vma->vm_flags = vm_flags;
1569 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1570 vma->vm_pgoff = pgoff;
1571 INIT_LIST_HEAD(&vma->anon_vma_chain);
1574 if (vm_flags & VM_DENYWRITE) {
1575 error = deny_write_access(file);
1579 if (vm_flags & VM_SHARED) {
1580 error = mapping_map_writable(file->f_mapping);
1582 goto allow_write_and_free_vma;
1585 /* ->mmap() can change vma->vm_file, but must guarantee that
1586 * vma_link() below can deny write-access if VM_DENYWRITE is set
1587 * and map writably if VM_SHARED is set. This usually means the
1588 * new file must not have been exposed to user-space, yet.
1590 vma->vm_file = get_file(file);
1591 error = file->f_op->mmap(file, vma);
1593 goto unmap_and_free_vma;
1595 /* Can addr have changed??
1597 * Answer: Yes, several device drivers can do it in their
1598 * f_op->mmap method. -DaveM
1599 * Bug: If addr is changed, prev, rb_link, rb_parent should
1600 * be updated for vma_link()
1602 WARN_ON_ONCE(addr != vma->vm_start);
1604 addr = vma->vm_start;
1605 vm_flags = vma->vm_flags;
1606 } else if (vm_flags & VM_SHARED) {
1607 error = shmem_zero_setup(vma);
1612 if (vma_wants_writenotify(vma)) {
1613 pgprot_t pprot = vma->vm_page_prot;
1615 /* Can vma->vm_page_prot have changed??
1617 * Answer: Yes, drivers may have changed it in their
1618 * f_op->mmap method.
1620 * Ensures that vmas marked as uncached stay that way.
1622 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1623 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1624 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1627 vma_link(mm, vma, prev, rb_link, rb_parent);
1628 /* Once vma denies write, undo our temporary denial count */
1630 if (vm_flags & VM_SHARED)
1631 mapping_unmap_writable(file->f_mapping);
1632 if (vm_flags & VM_DENYWRITE)
1633 allow_write_access(file);
1635 file = vma->vm_file;
1637 perf_event_mmap(vma);
1639 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1640 if (vm_flags & VM_LOCKED) {
1641 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1642 vma == get_gate_vma(current->mm)))
1643 mm->locked_vm += (len >> PAGE_SHIFT);
1645 vma->vm_flags &= ~VM_LOCKED;
1652 * New (or expanded) vma always get soft dirty status.
1653 * Otherwise user-space soft-dirty page tracker won't
1654 * be able to distinguish situation when vma area unmapped,
1655 * then new mapped in-place (which must be aimed as
1656 * a completely new data area).
1658 vma->vm_flags |= VM_SOFTDIRTY;
1663 vma->vm_file = NULL;
1666 /* Undo any partial mapping done by a device driver. */
1667 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1669 if (vm_flags & VM_SHARED)
1670 mapping_unmap_writable(file->f_mapping);
1671 allow_write_and_free_vma:
1672 if (vm_flags & VM_DENYWRITE)
1673 allow_write_access(file);
1675 kmem_cache_free(vm_area_cachep, vma);
1678 vm_unacct_memory(charged);
1682 unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1685 * We implement the search by looking for an rbtree node that
1686 * immediately follows a suitable gap. That is,
1687 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1688 * - gap_end = vma->vm_start >= info->low_limit + length;
1689 * - gap_end - gap_start >= length
1692 struct mm_struct *mm = current->mm;
1693 struct vm_area_struct *vma;
1694 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1696 /* Adjust search length to account for worst case alignment overhead */
1697 length = info->length + info->align_mask;
1698 if (length < info->length)
1701 /* Adjust search limits by the desired length */
1702 if (info->high_limit < length)
1704 high_limit = info->high_limit - length;
1706 if (info->low_limit > high_limit)
1708 low_limit = info->low_limit + length;
1710 /* Check if rbtree root looks promising */
1711 if (RB_EMPTY_ROOT(&mm->mm_rb))
1713 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1714 if (vma->rb_subtree_gap < length)
1718 /* Visit left subtree if it looks promising */
1719 gap_end = vma->vm_start;
1720 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1721 struct vm_area_struct *left =
1722 rb_entry(vma->vm_rb.rb_left,
1723 struct vm_area_struct, vm_rb);
1724 if (left->rb_subtree_gap >= length) {
1730 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1732 /* Check if current node has a suitable gap */
1733 if (gap_start > high_limit)
1735 if (gap_end >= low_limit && gap_end - gap_start >= length)
1738 /* Visit right subtree if it looks promising */
1739 if (vma->vm_rb.rb_right) {
1740 struct vm_area_struct *right =
1741 rb_entry(vma->vm_rb.rb_right,
1742 struct vm_area_struct, vm_rb);
1743 if (right->rb_subtree_gap >= length) {
1749 /* Go back up the rbtree to find next candidate node */
1751 struct rb_node *prev = &vma->vm_rb;
1752 if (!rb_parent(prev))
1754 vma = rb_entry(rb_parent(prev),
1755 struct vm_area_struct, vm_rb);
1756 if (prev == vma->vm_rb.rb_left) {
1757 gap_start = vma->vm_prev->vm_end;
1758 gap_end = vma->vm_start;
1765 /* Check highest gap, which does not precede any rbtree node */
1766 gap_start = mm->highest_vm_end;
1767 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1768 if (gap_start > high_limit)
1772 /* We found a suitable gap. Clip it with the original low_limit. */
1773 if (gap_start < info->low_limit)
1774 gap_start = info->low_limit;
1776 /* Adjust gap address to the desired alignment */
1777 gap_start += (info->align_offset - gap_start) & info->align_mask;
1779 VM_BUG_ON(gap_start + info->length > info->high_limit);
1780 VM_BUG_ON(gap_start + info->length > gap_end);
1784 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1786 struct mm_struct *mm = current->mm;
1787 struct vm_area_struct *vma;
1788 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1790 /* Adjust search length to account for worst case alignment overhead */
1791 length = info->length + info->align_mask;
1792 if (length < info->length)
1796 * Adjust search limits by the desired length.
1797 * See implementation comment at top of unmapped_area().
1799 gap_end = info->high_limit;
1800 if (gap_end < length)
1802 high_limit = gap_end - length;
1804 if (info->low_limit > high_limit)
1806 low_limit = info->low_limit + length;
1808 /* Check highest gap, which does not precede any rbtree node */
1809 gap_start = mm->highest_vm_end;
1810 if (gap_start <= high_limit)
1813 /* Check if rbtree root looks promising */
1814 if (RB_EMPTY_ROOT(&mm->mm_rb))
1816 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1817 if (vma->rb_subtree_gap < length)
1821 /* Visit right subtree if it looks promising */
1822 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1823 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1824 struct vm_area_struct *right =
1825 rb_entry(vma->vm_rb.rb_right,
1826 struct vm_area_struct, vm_rb);
1827 if (right->rb_subtree_gap >= length) {
1834 /* Check if current node has a suitable gap */
1835 gap_end = vma->vm_start;
1836 if (gap_end < low_limit)
1838 if (gap_start <= high_limit && gap_end - gap_start >= length)
1841 /* Visit left subtree if it looks promising */
1842 if (vma->vm_rb.rb_left) {
1843 struct vm_area_struct *left =
1844 rb_entry(vma->vm_rb.rb_left,
1845 struct vm_area_struct, vm_rb);
1846 if (left->rb_subtree_gap >= length) {
1852 /* Go back up the rbtree to find next candidate node */
1854 struct rb_node *prev = &vma->vm_rb;
1855 if (!rb_parent(prev))
1857 vma = rb_entry(rb_parent(prev),
1858 struct vm_area_struct, vm_rb);
1859 if (prev == vma->vm_rb.rb_right) {
1860 gap_start = vma->vm_prev ?
1861 vma->vm_prev->vm_end : 0;
1868 /* We found a suitable gap. Clip it with the original high_limit. */
1869 if (gap_end > info->high_limit)
1870 gap_end = info->high_limit;
1873 /* Compute highest gap address at the desired alignment */
1874 gap_end -= info->length;
1875 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1877 VM_BUG_ON(gap_end < info->low_limit);
1878 VM_BUG_ON(gap_end < gap_start);
1882 /* Get an address range which is currently unmapped.
1883 * For shmat() with addr=0.
1885 * Ugly calling convention alert:
1886 * Return value with the low bits set means error value,
1888 * if (ret & ~PAGE_MASK)
1891 * This function "knows" that -ENOMEM has the bits set.
1893 #ifndef HAVE_ARCH_UNMAPPED_AREA
1895 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1896 unsigned long len, unsigned long pgoff, unsigned long flags)
1898 struct mm_struct *mm = current->mm;
1899 struct vm_area_struct *vma;
1900 struct vm_unmapped_area_info info;
1902 if (len > TASK_SIZE - mmap_min_addr)
1905 if (flags & MAP_FIXED)
1909 addr = PAGE_ALIGN(addr);
1910 vma = find_vma(mm, addr);
1911 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1912 (!vma || addr + len <= vma->vm_start))
1918 info.low_limit = mm->mmap_base;
1919 info.high_limit = TASK_SIZE;
1920 info.align_mask = 0;
1921 return vm_unmapped_area(&info);
1926 * This mmap-allocator allocates new areas top-down from below the
1927 * stack's low limit (the base):
1929 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1931 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1932 const unsigned long len, const unsigned long pgoff,
1933 const unsigned long flags)
1935 struct vm_area_struct *vma;
1936 struct mm_struct *mm = current->mm;
1937 unsigned long addr = addr0;
1938 struct vm_unmapped_area_info info;
1940 /* requested length too big for entire address space */
1941 if (len > TASK_SIZE - mmap_min_addr)
1944 if (flags & MAP_FIXED)
1947 /* requesting a specific address */
1949 addr = PAGE_ALIGN(addr);
1950 vma = find_vma(mm, addr);
1951 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1952 (!vma || addr + len <= vma->vm_start))
1956 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1958 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
1959 info.high_limit = mm->mmap_base;
1960 info.align_mask = 0;
1961 addr = vm_unmapped_area(&info);
1964 * A failed mmap() very likely causes application failure,
1965 * so fall back to the bottom-up function here. This scenario
1966 * can happen with large stack limits and large mmap()
1969 if (addr & ~PAGE_MASK) {
1970 VM_BUG_ON(addr != -ENOMEM);
1972 info.low_limit = TASK_UNMAPPED_BASE;
1973 info.high_limit = TASK_SIZE;
1974 addr = vm_unmapped_area(&info);
1982 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1983 unsigned long pgoff, unsigned long flags)
1985 unsigned long (*get_area)(struct file *, unsigned long,
1986 unsigned long, unsigned long, unsigned long);
1988 unsigned long error = arch_mmap_check(addr, len, flags);
1992 /* Careful about overflows.. */
1993 if (len > TASK_SIZE)
1996 get_area = current->mm->get_unmapped_area;
1997 if (file && file->f_op->get_unmapped_area)
1998 get_area = file->f_op->get_unmapped_area;
1999 addr = get_area(file, addr, len, pgoff, flags);
2000 if (IS_ERR_VALUE(addr))
2003 if (addr > TASK_SIZE - len)
2005 if (addr & ~PAGE_MASK)
2008 addr = arch_rebalance_pgtables(addr, len);
2009 error = security_mmap_addr(addr);
2010 return error ? error : addr;
2013 EXPORT_SYMBOL(get_unmapped_area);
2015 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2016 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
2018 struct rb_node *rb_node;
2019 struct vm_area_struct *vma;
2021 /* Check the cache first. */
2022 vma = vmacache_find(mm, addr);
2026 rb_node = mm->mm_rb.rb_node;
2030 struct vm_area_struct *tmp;
2032 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2034 if (tmp->vm_end > addr) {
2036 if (tmp->vm_start <= addr)
2038 rb_node = rb_node->rb_left;
2040 rb_node = rb_node->rb_right;
2044 vmacache_update(addr, vma);
2048 EXPORT_SYMBOL(find_vma);
2051 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2053 struct vm_area_struct *
2054 find_vma_prev(struct mm_struct *mm, unsigned long addr,
2055 struct vm_area_struct **pprev)
2057 struct vm_area_struct *vma;
2059 vma = find_vma(mm, addr);
2061 *pprev = vma->vm_prev;
2063 struct rb_node *rb_node = mm->mm_rb.rb_node;
2066 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2067 rb_node = rb_node->rb_right;
2074 * Verify that the stack growth is acceptable and
2075 * update accounting. This is shared with both the
2076 * grow-up and grow-down cases.
2078 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
2080 struct mm_struct *mm = vma->vm_mm;
2081 struct rlimit *rlim = current->signal->rlim;
2082 unsigned long new_start;
2084 /* address space limit tests */
2085 if (!may_expand_vm(mm, grow))
2088 /* Stack limit test */
2089 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
2092 /* mlock limit tests */
2093 if (vma->vm_flags & VM_LOCKED) {
2094 unsigned long locked;
2095 unsigned long limit;
2096 locked = mm->locked_vm + grow;
2097 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2098 limit >>= PAGE_SHIFT;
2099 if (locked > limit && !capable(CAP_IPC_LOCK))
2103 /* Check to ensure the stack will not grow into a hugetlb-only region */
2104 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2106 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2110 * Overcommit.. This must be the final test, as it will
2111 * update security statistics.
2113 if (security_vm_enough_memory_mm(mm, grow))
2116 /* Ok, everything looks good - let it rip */
2117 if (vma->vm_flags & VM_LOCKED)
2118 mm->locked_vm += grow;
2119 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
2123 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2125 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2126 * vma is the last one with address > vma->vm_end. Have to extend vma.
2128 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2132 if (!(vma->vm_flags & VM_GROWSUP))
2136 * We must make sure the anon_vma is allocated
2137 * so that the anon_vma locking is not a noop.
2139 if (unlikely(anon_vma_prepare(vma)))
2141 vma_lock_anon_vma(vma);
2144 * vma->vm_start/vm_end cannot change under us because the caller
2145 * is required to hold the mmap_sem in read mode. We need the
2146 * anon_vma lock to serialize against concurrent expand_stacks.
2147 * Also guard against wrapping around to address 0.
2149 if (address < PAGE_ALIGN(address+4))
2150 address = PAGE_ALIGN(address+4);
2152 vma_unlock_anon_vma(vma);
2157 /* Somebody else might have raced and expanded it already */
2158 if (address > vma->vm_end) {
2159 unsigned long size, grow;
2161 size = address - vma->vm_start;
2162 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2165 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2166 error = acct_stack_growth(vma, size, grow);
2169 * vma_gap_update() doesn't support concurrent
2170 * updates, but we only hold a shared mmap_sem
2171 * lock here, so we need to protect against
2172 * concurrent vma expansions.
2173 * vma_lock_anon_vma() doesn't help here, as
2174 * we don't guarantee that all growable vmas
2175 * in a mm share the same root anon vma.
2176 * So, we reuse mm->page_table_lock to guard
2177 * against concurrent vma expansions.
2179 spin_lock(&vma->vm_mm->page_table_lock);
2180 anon_vma_interval_tree_pre_update_vma(vma);
2181 vma->vm_end = address;
2182 anon_vma_interval_tree_post_update_vma(vma);
2184 vma_gap_update(vma->vm_next);
2186 vma->vm_mm->highest_vm_end = address;
2187 spin_unlock(&vma->vm_mm->page_table_lock);
2189 perf_event_mmap(vma);
2193 vma_unlock_anon_vma(vma);
2194 khugepaged_enter_vma_merge(vma);
2195 validate_mm(vma->vm_mm);
2198 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2201 * vma is the first one with address < vma->vm_start. Have to extend vma.
2203 int expand_downwards(struct vm_area_struct *vma,
2204 unsigned long address)
2209 * We must make sure the anon_vma is allocated
2210 * so that the anon_vma locking is not a noop.
2212 if (unlikely(anon_vma_prepare(vma)))
2215 address &= PAGE_MASK;
2216 error = security_mmap_addr(address);
2220 vma_lock_anon_vma(vma);
2223 * vma->vm_start/vm_end cannot change under us because the caller
2224 * is required to hold the mmap_sem in read mode. We need the
2225 * anon_vma lock to serialize against concurrent expand_stacks.
2228 /* Somebody else might have raced and expanded it already */
2229 if (address < vma->vm_start) {
2230 unsigned long size, grow;
2232 size = vma->vm_end - address;
2233 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2236 if (grow <= vma->vm_pgoff) {
2237 error = acct_stack_growth(vma, size, grow);
2240 * vma_gap_update() doesn't support concurrent
2241 * updates, but we only hold a shared mmap_sem
2242 * lock here, so we need to protect against
2243 * concurrent vma expansions.
2244 * vma_lock_anon_vma() doesn't help here, as
2245 * we don't guarantee that all growable vmas
2246 * in a mm share the same root anon vma.
2247 * So, we reuse mm->page_table_lock to guard
2248 * against concurrent vma expansions.
2250 spin_lock(&vma->vm_mm->page_table_lock);
2251 anon_vma_interval_tree_pre_update_vma(vma);
2252 vma->vm_start = address;
2253 vma->vm_pgoff -= grow;
2254 anon_vma_interval_tree_post_update_vma(vma);
2255 vma_gap_update(vma);
2256 spin_unlock(&vma->vm_mm->page_table_lock);
2258 perf_event_mmap(vma);
2262 vma_unlock_anon_vma(vma);
2263 khugepaged_enter_vma_merge(vma);
2264 validate_mm(vma->vm_mm);
2269 * Note how expand_stack() refuses to expand the stack all the way to
2270 * abut the next virtual mapping, *unless* that mapping itself is also
2271 * a stack mapping. We want to leave room for a guard page, after all
2272 * (the guard page itself is not added here, that is done by the
2273 * actual page faulting logic)
2275 * This matches the behavior of the guard page logic (see mm/memory.c:
2276 * check_stack_guard_page()), which only allows the guard page to be
2277 * removed under these circumstances.
2279 #ifdef CONFIG_STACK_GROWSUP
2280 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2282 struct vm_area_struct *next;
2284 address &= PAGE_MASK;
2285 next = vma->vm_next;
2286 if (next && next->vm_start == address + PAGE_SIZE) {
2287 if (!(next->vm_flags & VM_GROWSUP))
2290 return expand_upwards(vma, address);
2293 struct vm_area_struct *
2294 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2296 struct vm_area_struct *vma, *prev;
2299 vma = find_vma_prev(mm, addr, &prev);
2300 if (vma && (vma->vm_start <= addr))
2302 if (!prev || expand_stack(prev, addr))
2304 if (prev->vm_flags & VM_LOCKED)
2305 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
2309 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2311 struct vm_area_struct *prev;
2313 address &= PAGE_MASK;
2314 prev = vma->vm_prev;
2315 if (prev && prev->vm_end == address) {
2316 if (!(prev->vm_flags & VM_GROWSDOWN))
2319 return expand_downwards(vma, address);
2322 struct vm_area_struct *
2323 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2325 struct vm_area_struct *vma;
2326 unsigned long start;
2329 vma = find_vma(mm, addr);
2332 if (vma->vm_start <= addr)
2334 if (!(vma->vm_flags & VM_GROWSDOWN))
2336 start = vma->vm_start;
2337 if (expand_stack(vma, addr))
2339 if (vma->vm_flags & VM_LOCKED)
2340 __mlock_vma_pages_range(vma, addr, start, NULL);
2346 * Ok - we have the memory areas we should free on the vma list,
2347 * so release them, and do the vma updates.
2349 * Called with the mm semaphore held.
2351 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2353 unsigned long nr_accounted = 0;
2355 /* Update high watermark before we lower total_vm */
2356 update_hiwater_vm(mm);
2358 long nrpages = vma_pages(vma);
2360 if (vma->vm_flags & VM_ACCOUNT)
2361 nr_accounted += nrpages;
2362 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
2363 vma = remove_vma(vma);
2365 vm_unacct_memory(nr_accounted);
2370 * Get rid of page table information in the indicated region.
2372 * Called with the mm semaphore held.
2374 static void unmap_region(struct mm_struct *mm,
2375 struct vm_area_struct *vma, struct vm_area_struct *prev,
2376 unsigned long start, unsigned long end)
2378 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
2379 struct mmu_gather tlb;
2382 tlb_gather_mmu(&tlb, mm, start, end);
2383 update_hiwater_rss(mm);
2384 unmap_vmas(&tlb, vma, start, end);
2385 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2386 next ? next->vm_start : USER_PGTABLES_CEILING);
2387 tlb_finish_mmu(&tlb, start, end);
2391 * Create a list of vma's touched by the unmap, removing them from the mm's
2392 * vma list as we go..
2395 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2396 struct vm_area_struct *prev, unsigned long end)
2398 struct vm_area_struct **insertion_point;
2399 struct vm_area_struct *tail_vma = NULL;
2401 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2402 vma->vm_prev = NULL;
2404 vma_rb_erase(vma, &mm->mm_rb);
2408 } while (vma && vma->vm_start < end);
2409 *insertion_point = vma;
2411 vma->vm_prev = prev;
2412 vma_gap_update(vma);
2414 mm->highest_vm_end = prev ? prev->vm_end : 0;
2415 tail_vma->vm_next = NULL;
2417 /* Kill the cache */
2418 vmacache_invalidate(mm);
2422 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2423 * munmap path where it doesn't make sense to fail.
2425 static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2426 unsigned long addr, int new_below)
2428 struct vm_area_struct *new;
2431 if (is_vm_hugetlb_page(vma) && (addr &
2432 ~(huge_page_mask(hstate_vma(vma)))))
2435 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2439 /* most fields are the same, copy all, and then fixup */
2442 INIT_LIST_HEAD(&new->anon_vma_chain);
2447 new->vm_start = addr;
2448 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2451 err = vma_dup_policy(vma, new);
2455 if (anon_vma_clone(new, vma))
2459 get_file(new->vm_file);
2461 if (new->vm_ops && new->vm_ops->open)
2462 new->vm_ops->open(new);
2465 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2466 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2468 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2474 /* Clean everything up if vma_adjust failed. */
2475 if (new->vm_ops && new->vm_ops->close)
2476 new->vm_ops->close(new);
2479 unlink_anon_vmas(new);
2481 mpol_put(vma_policy(new));
2483 kmem_cache_free(vm_area_cachep, new);
2489 * Split a vma into two pieces at address 'addr', a new vma is allocated
2490 * either for the first part or the tail.
2492 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2493 unsigned long addr, int new_below)
2495 if (mm->map_count >= sysctl_max_map_count)
2498 return __split_vma(mm, vma, addr, new_below);
2501 /* Munmap is split into 2 main parts -- this part which finds
2502 * what needs doing, and the areas themselves, which do the
2503 * work. This now handles partial unmappings.
2504 * Jeremy Fitzhardinge <jeremy@goop.org>
2506 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2509 struct vm_area_struct *vma, *prev, *last;
2511 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2514 len = PAGE_ALIGN(len);
2518 /* Find the first overlapping VMA */
2519 vma = find_vma(mm, start);
2522 prev = vma->vm_prev;
2523 /* we have start < vma->vm_end */
2525 /* if it doesn't overlap, we have nothing.. */
2527 if (vma->vm_start >= end)
2531 * If we need to split any vma, do it now to save pain later.
2533 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2534 * unmapped vm_area_struct will remain in use: so lower split_vma
2535 * places tmp vma above, and higher split_vma places tmp vma below.
2537 if (start > vma->vm_start) {
2541 * Make sure that map_count on return from munmap() will
2542 * not exceed its limit; but let map_count go just above
2543 * its limit temporarily, to help free resources as expected.
2545 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2548 error = __split_vma(mm, vma, start, 0);
2554 /* Does it split the last one? */
2555 last = find_vma(mm, end);
2556 if (last && end > last->vm_start) {
2557 int error = __split_vma(mm, last, end, 1);
2561 vma = prev ? prev->vm_next : mm->mmap;
2564 * unlock any mlock()ed ranges before detaching vmas
2566 if (mm->locked_vm) {
2567 struct vm_area_struct *tmp = vma;
2568 while (tmp && tmp->vm_start < end) {
2569 if (tmp->vm_flags & VM_LOCKED) {
2570 mm->locked_vm -= vma_pages(tmp);
2571 munlock_vma_pages_all(tmp);
2578 * Remove the vma's, and unmap the actual pages
2580 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2581 unmap_region(mm, vma, prev, start, end);
2583 /* Fix up all other VM information */
2584 remove_vma_list(mm, vma);
2589 int vm_munmap(unsigned long start, size_t len)
2592 struct mm_struct *mm = current->mm;
2594 down_write(&mm->mmap_sem);
2595 ret = do_munmap(mm, start, len);
2596 up_write(&mm->mmap_sem);
2599 EXPORT_SYMBOL(vm_munmap);
2601 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2603 profile_munmap(addr);
2604 return vm_munmap(addr, len);
2607 static inline void verify_mm_writelocked(struct mm_struct *mm)
2609 #ifdef CONFIG_DEBUG_VM
2610 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2612 up_read(&mm->mmap_sem);
2618 * this is really a simplified "do_mmap". it only handles
2619 * anonymous maps. eventually we may be able to do some
2620 * brk-specific accounting here.
2622 static unsigned long do_brk(unsigned long addr, unsigned long len)
2624 struct mm_struct *mm = current->mm;
2625 struct vm_area_struct *vma, *prev;
2626 unsigned long flags;
2627 struct rb_node **rb_link, *rb_parent;
2628 pgoff_t pgoff = addr >> PAGE_SHIFT;
2631 len = PAGE_ALIGN(len);
2635 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2637 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2638 if (error & ~PAGE_MASK)
2641 error = mlock_future_check(mm, mm->def_flags, len);
2646 * mm->mmap_sem is required to protect against another thread
2647 * changing the mappings in case we sleep.
2649 verify_mm_writelocked(mm);
2652 * Clear old maps. this also does some error checking for us
2655 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
2656 if (do_munmap(mm, addr, len))
2661 /* Check against address space limits *after* clearing old maps... */
2662 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2665 if (mm->map_count > sysctl_max_map_count)
2668 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2671 /* Can we just expand an old private anonymous mapping? */
2672 vma = vma_merge(mm, prev, addr, addr + len, flags,
2673 NULL, NULL, pgoff, NULL);
2678 * create a vma struct for an anonymous mapping
2680 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2682 vm_unacct_memory(len >> PAGE_SHIFT);
2686 INIT_LIST_HEAD(&vma->anon_vma_chain);
2688 vma->vm_start = addr;
2689 vma->vm_end = addr + len;
2690 vma->vm_pgoff = pgoff;
2691 vma->vm_flags = flags;
2692 vma->vm_page_prot = vm_get_page_prot(flags);
2693 vma_link(mm, vma, prev, rb_link, rb_parent);
2695 perf_event_mmap(vma);
2696 mm->total_vm += len >> PAGE_SHIFT;
2697 if (flags & VM_LOCKED)
2698 mm->locked_vm += (len >> PAGE_SHIFT);
2699 vma->vm_flags |= VM_SOFTDIRTY;
2703 unsigned long vm_brk(unsigned long addr, unsigned long len)
2705 struct mm_struct *mm = current->mm;
2709 down_write(&mm->mmap_sem);
2710 ret = do_brk(addr, len);
2711 populate = ((mm->def_flags & VM_LOCKED) != 0);
2712 up_write(&mm->mmap_sem);
2714 mm_populate(addr, len);
2717 EXPORT_SYMBOL(vm_brk);
2719 /* Release all mmaps. */
2720 void exit_mmap(struct mm_struct *mm)
2722 struct mmu_gather tlb;
2723 struct vm_area_struct *vma;
2724 unsigned long nr_accounted = 0;
2726 /* mm's last user has gone, and its about to be pulled down */
2727 mmu_notifier_release(mm);
2729 if (mm->locked_vm) {
2732 if (vma->vm_flags & VM_LOCKED)
2733 munlock_vma_pages_all(vma);
2741 if (!vma) /* Can happen if dup_mmap() received an OOM */
2746 tlb_gather_mmu(&tlb, mm, 0, -1);
2747 /* update_hiwater_rss(mm) here? but nobody should be looking */
2748 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2749 unmap_vmas(&tlb, vma, 0, -1);
2751 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
2752 tlb_finish_mmu(&tlb, 0, -1);
2755 * Walk the list again, actually closing and freeing it,
2756 * with preemption enabled, without holding any MM locks.
2759 if (vma->vm_flags & VM_ACCOUNT)
2760 nr_accounted += vma_pages(vma);
2761 vma = remove_vma(vma);
2763 vm_unacct_memory(nr_accounted);
2765 WARN_ON(atomic_long_read(&mm->nr_ptes) >
2766 (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2769 /* Insert vm structure into process list sorted by address
2770 * and into the inode's i_mmap tree. If vm_file is non-NULL
2771 * then i_mmap_mutex is taken here.
2773 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
2775 struct vm_area_struct *prev;
2776 struct rb_node **rb_link, *rb_parent;
2779 * The vm_pgoff of a purely anonymous vma should be irrelevant
2780 * until its first write fault, when page's anon_vma and index
2781 * are set. But now set the vm_pgoff it will almost certainly
2782 * end up with (unless mremap moves it elsewhere before that
2783 * first wfault), so /proc/pid/maps tells a consistent story.
2785 * By setting it to reflect the virtual start address of the
2786 * vma, merges and splits can happen in a seamless way, just
2787 * using the existing file pgoff checks and manipulations.
2788 * Similarly in do_mmap_pgoff and in do_brk.
2790 if (!vma->vm_file) {
2791 BUG_ON(vma->anon_vma);
2792 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2794 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2795 &prev, &rb_link, &rb_parent))
2797 if ((vma->vm_flags & VM_ACCOUNT) &&
2798 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2801 vma_link(mm, vma, prev, rb_link, rb_parent);
2806 * Copy the vma structure to a new location in the same mm,
2807 * prior to moving page table entries, to effect an mremap move.
2809 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2810 unsigned long addr, unsigned long len, pgoff_t pgoff,
2811 bool *need_rmap_locks)
2813 struct vm_area_struct *vma = *vmap;
2814 unsigned long vma_start = vma->vm_start;
2815 struct mm_struct *mm = vma->vm_mm;
2816 struct vm_area_struct *new_vma, *prev;
2817 struct rb_node **rb_link, *rb_parent;
2818 bool faulted_in_anon_vma = true;
2821 * If anonymous vma has not yet been faulted, update new pgoff
2822 * to match new location, to increase its chance of merging.
2824 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2825 pgoff = addr >> PAGE_SHIFT;
2826 faulted_in_anon_vma = false;
2829 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2830 return NULL; /* should never get here */
2831 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2832 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2835 * Source vma may have been merged into new_vma
2837 if (unlikely(vma_start >= new_vma->vm_start &&
2838 vma_start < new_vma->vm_end)) {
2840 * The only way we can get a vma_merge with
2841 * self during an mremap is if the vma hasn't
2842 * been faulted in yet and we were allowed to
2843 * reset the dst vma->vm_pgoff to the
2844 * destination address of the mremap to allow
2845 * the merge to happen. mremap must change the
2846 * vm_pgoff linearity between src and dst vmas
2847 * (in turn preventing a vma_merge) to be
2848 * safe. It is only safe to keep the vm_pgoff
2849 * linear if there are no pages mapped yet.
2851 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
2852 *vmap = vma = new_vma;
2854 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
2856 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2859 new_vma->vm_start = addr;
2860 new_vma->vm_end = addr + len;
2861 new_vma->vm_pgoff = pgoff;
2862 if (vma_dup_policy(vma, new_vma))
2864 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2865 if (anon_vma_clone(new_vma, vma))
2866 goto out_free_mempol;
2867 if (new_vma->vm_file)
2868 get_file(new_vma->vm_file);
2869 if (new_vma->vm_ops && new_vma->vm_ops->open)
2870 new_vma->vm_ops->open(new_vma);
2871 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2872 *need_rmap_locks = false;
2878 mpol_put(vma_policy(new_vma));
2880 kmem_cache_free(vm_area_cachep, new_vma);
2885 * Return true if the calling process may expand its vm space by the passed
2888 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2890 unsigned long cur = mm->total_vm; /* pages */
2893 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2895 if (cur + npages > lim)
2900 static int special_mapping_fault(struct vm_area_struct *vma,
2901 struct vm_fault *vmf);
2904 * Having a close hook prevents vma merging regardless of flags.
2906 static void special_mapping_close(struct vm_area_struct *vma)
2910 static const char *special_mapping_name(struct vm_area_struct *vma)
2912 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
2915 static const struct vm_operations_struct special_mapping_vmops = {
2916 .close = special_mapping_close,
2917 .fault = special_mapping_fault,
2918 .name = special_mapping_name,
2921 static const struct vm_operations_struct legacy_special_mapping_vmops = {
2922 .close = special_mapping_close,
2923 .fault = special_mapping_fault,
2926 static int special_mapping_fault(struct vm_area_struct *vma,
2927 struct vm_fault *vmf)
2930 struct page **pages;
2933 * special mappings have no vm_file, and in that case, the mm
2934 * uses vm_pgoff internally. So we have to subtract it from here.
2935 * We are allowed to do this because we are the mm; do not copy
2936 * this code into drivers!
2938 pgoff = vmf->pgoff - vma->vm_pgoff;
2940 if (vma->vm_ops == &legacy_special_mapping_vmops)
2941 pages = vma->vm_private_data;
2943 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
2946 for (; pgoff && *pages; ++pages)
2950 struct page *page = *pages;
2956 return VM_FAULT_SIGBUS;
2959 static struct vm_area_struct *__install_special_mapping(
2960 struct mm_struct *mm,
2961 unsigned long addr, unsigned long len,
2962 unsigned long vm_flags, const struct vm_operations_struct *ops,
2966 struct vm_area_struct *vma;
2968 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2969 if (unlikely(vma == NULL))
2970 return ERR_PTR(-ENOMEM);
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 | VM_SOFTDIRTY;
2978 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2981 vma->vm_private_data = priv;
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);
2995 return ERR_PTR(ret);
2999 * Called with mm->mmap_sem held for writing.
3000 * Insert a new vma covering the given region, with the given flags.
3001 * Its pages are supplied by the given array of struct page *.
3002 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3003 * The region past the last page supplied will always produce SIGBUS.
3004 * The array pointer and the pages it points to are assumed to stay alive
3005 * for as long as this mapping might exist.
3007 struct vm_area_struct *_install_special_mapping(
3008 struct mm_struct *mm,
3009 unsigned long addr, unsigned long len,
3010 unsigned long vm_flags, const struct vm_special_mapping *spec)
3012 return __install_special_mapping(mm, addr, len, vm_flags,
3013 &special_mapping_vmops, (void *)spec);
3016 int install_special_mapping(struct mm_struct *mm,
3017 unsigned long addr, unsigned long len,
3018 unsigned long vm_flags, struct page **pages)
3020 struct vm_area_struct *vma = __install_special_mapping(
3021 mm, addr, len, vm_flags, &legacy_special_mapping_vmops,
3024 return PTR_ERR_OR_ZERO(vma);
3027 static DEFINE_MUTEX(mm_all_locks_mutex);
3029 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3031 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3033 * The LSB of head.next can't change from under us
3034 * because we hold the mm_all_locks_mutex.
3036 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
3038 * We can safely modify head.next after taking the
3039 * anon_vma->root->rwsem. If some other vma in this mm shares
3040 * the same anon_vma we won't take it again.
3042 * No need of atomic instructions here, head.next
3043 * can't change from under us thanks to the
3044 * anon_vma->root->rwsem.
3046 if (__test_and_set_bit(0, (unsigned long *)
3047 &anon_vma->root->rb_root.rb_node))
3052 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3054 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3056 * AS_MM_ALL_LOCKS can't change from under us because
3057 * we hold the mm_all_locks_mutex.
3059 * Operations on ->flags have to be atomic because
3060 * even if AS_MM_ALL_LOCKS is stable thanks to the
3061 * mm_all_locks_mutex, there may be other cpus
3062 * changing other bitflags in parallel to us.
3064 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3066 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
3071 * This operation locks against the VM for all pte/vma/mm related
3072 * operations that could ever happen on a certain mm. This includes
3073 * vmtruncate, try_to_unmap, and all page faults.
3075 * The caller must take the mmap_sem in write mode before calling
3076 * mm_take_all_locks(). The caller isn't allowed to release the
3077 * mmap_sem until mm_drop_all_locks() returns.
3079 * mmap_sem in write mode is required in order to block all operations
3080 * that could modify pagetables and free pages without need of
3081 * altering the vma layout (for example populate_range() with
3082 * nonlinear vmas). It's also needed in write mode to avoid new
3083 * anon_vmas to be associated with existing vmas.
3085 * A single task can't take more than one mm_take_all_locks() in a row
3086 * or it would deadlock.
3088 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3089 * mapping->flags avoid to take the same lock twice, if more than one
3090 * vma in this mm is backed by the same anon_vma or address_space.
3092 * We can take all the locks in random order because the VM code
3093 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
3094 * takes more than one of them in a row. Secondly we're protected
3095 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3097 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3098 * that may have to take thousand of locks.
3100 * mm_take_all_locks() can fail if it's interrupted by signals.
3102 int mm_take_all_locks(struct mm_struct *mm)
3104 struct vm_area_struct *vma;
3105 struct anon_vma_chain *avc;
3107 BUG_ON(down_read_trylock(&mm->mmap_sem));
3109 mutex_lock(&mm_all_locks_mutex);
3111 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3112 if (signal_pending(current))
3114 if (vma->vm_file && vma->vm_file->f_mapping)
3115 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3118 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3119 if (signal_pending(current))
3122 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3123 vm_lock_anon_vma(mm, avc->anon_vma);
3129 mm_drop_all_locks(mm);
3133 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3135 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3137 * The LSB of head.next can't change to 0 from under
3138 * us because we hold the mm_all_locks_mutex.
3140 * We must however clear the bitflag before unlocking
3141 * the vma so the users using the anon_vma->rb_root will
3142 * never see our bitflag.
3144 * No need of atomic instructions here, head.next
3145 * can't change from under us until we release the
3146 * anon_vma->root->rwsem.
3148 if (!__test_and_clear_bit(0, (unsigned long *)
3149 &anon_vma->root->rb_root.rb_node))
3151 anon_vma_unlock_write(anon_vma);
3155 static void vm_unlock_mapping(struct address_space *mapping)
3157 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3159 * AS_MM_ALL_LOCKS can't change to 0 from under us
3160 * because we hold the mm_all_locks_mutex.
3162 mutex_unlock(&mapping->i_mmap_mutex);
3163 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3170 * The mmap_sem cannot be released by the caller until
3171 * mm_drop_all_locks() returns.
3173 void mm_drop_all_locks(struct mm_struct *mm)
3175 struct vm_area_struct *vma;
3176 struct anon_vma_chain *avc;
3178 BUG_ON(down_read_trylock(&mm->mmap_sem));
3179 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3181 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3183 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3184 vm_unlock_anon_vma(avc->anon_vma);
3185 if (vma->vm_file && vma->vm_file->f_mapping)
3186 vm_unlock_mapping(vma->vm_file->f_mapping);
3189 mutex_unlock(&mm_all_locks_mutex);
3193 * initialise the VMA slab
3195 void __init mmap_init(void)
3199 ret = percpu_counter_init(&vm_committed_as, 0);
3204 * Initialise sysctl_user_reserve_kbytes.
3206 * This is intended to prevent a user from starting a single memory hogging
3207 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3210 * The default value is min(3% of free memory, 128MB)
3211 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3213 static int init_user_reserve(void)
3215 unsigned long free_kbytes;
3217 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3219 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3222 subsys_initcall(init_user_reserve);
3225 * Initialise sysctl_admin_reserve_kbytes.
3227 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3228 * to log in and kill a memory hogging process.
3230 * Systems with more than 256MB will reserve 8MB, enough to recover
3231 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3232 * only reserve 3% of free pages by default.
3234 static int init_admin_reserve(void)
3236 unsigned long free_kbytes;
3238 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3240 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3243 subsys_initcall(init_admin_reserve);
3246 * Reinititalise user and admin reserves if memory is added or removed.
3248 * The default user reserve max is 128MB, and the default max for the
3249 * admin reserve is 8MB. These are usually, but not always, enough to
3250 * enable recovery from a memory hogging process using login/sshd, a shell,
3251 * and tools like top. It may make sense to increase or even disable the
3252 * reserve depending on the existence of swap or variations in the recovery
3253 * tools. So, the admin may have changed them.
3255 * If memory is added and the reserves have been eliminated or increased above
3256 * the default max, then we'll trust the admin.
3258 * If memory is removed and there isn't enough free memory, then we
3259 * need to reset the reserves.
3261 * Otherwise keep the reserve set by the admin.
3263 static int reserve_mem_notifier(struct notifier_block *nb,
3264 unsigned long action, void *data)
3266 unsigned long tmp, free_kbytes;
3270 /* Default max is 128MB. Leave alone if modified by operator. */
3271 tmp = sysctl_user_reserve_kbytes;
3272 if (0 < tmp && tmp < (1UL << 17))
3273 init_user_reserve();
3275 /* Default max is 8MB. Leave alone if modified by operator. */
3276 tmp = sysctl_admin_reserve_kbytes;
3277 if (0 < tmp && tmp < (1UL << 13))
3278 init_admin_reserve();
3282 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3284 if (sysctl_user_reserve_kbytes > free_kbytes) {
3285 init_user_reserve();
3286 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3287 sysctl_user_reserve_kbytes);
3290 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3291 init_admin_reserve();
3292 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3293 sysctl_admin_reserve_kbytes);
3302 static struct notifier_block reserve_mem_nb = {
3303 .notifier_call = reserve_mem_notifier,
3306 static int __meminit init_reserve_notifier(void)
3308 if (register_hotmemory_notifier(&reserve_mem_nb))
3309 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3313 subsys_initcall(init_reserve_notifier);