1 // SPDX-License-Identifier: GPL-2.0-only
7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/backing-dev.h>
16 #include <linux/vmacache.h>
17 #include <linux/shm.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/syscalls.h>
22 #include <linux/capability.h>
23 #include <linux/init.h>
24 #include <linux/file.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/hugetlb.h>
29 #include <linux/shmem_fs.h>
30 #include <linux/profile.h>
31 #include <linux/export.h>
32 #include <linux/mount.h>
33 #include <linux/mempolicy.h>
34 #include <linux/rmap.h>
35 #include <linux/mmu_notifier.h>
36 #include <linux/mmdebug.h>
37 #include <linux/perf_event.h>
38 #include <linux/audit.h>
39 #include <linux/khugepaged.h>
40 #include <linux/uprobes.h>
41 #include <linux/rbtree_augmented.h>
42 #include <linux/notifier.h>
43 #include <linux/memory.h>
44 #include <linux/printk.h>
45 #include <linux/userfaultfd_k.h>
46 #include <linux/moduleparam.h>
47 #include <linux/pkeys.h>
48 #include <linux/oom.h>
49 #include <linux/sched/mm.h>
51 #include <linux/uaccess.h>
52 #include <asm/cacheflush.h>
54 #include <asm/mmu_context.h>
58 #ifndef arch_mmap_check
59 #define arch_mmap_check(addr, len, flags) (0)
62 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
63 const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
64 const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
65 int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
67 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
68 const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
69 const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
70 int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
73 static bool ignore_rlimit_data;
74 core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
76 static void unmap_region(struct mm_struct *mm,
77 struct vm_area_struct *vma, struct vm_area_struct *prev,
78 unsigned long start, unsigned long end);
80 /* description of effects of mapping type and prot in current implementation.
81 * this is due to the limited x86 page protection hardware. The expected
82 * behavior is in parens:
85 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
86 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 * w: (no) no w: (no) no w: (yes) yes w: (no) no
88 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
90 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
91 * w: (no) no w: (no) no w: (copy) copy w: (no) no
92 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
94 pgprot_t protection_map[16] __ro_after_init = {
95 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
96 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
99 #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT
100 static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
106 pgprot_t vm_get_page_prot(unsigned long vm_flags)
108 pgprot_t ret = __pgprot(pgprot_val(protection_map[vm_flags &
109 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
110 pgprot_val(arch_vm_get_page_prot(vm_flags)));
112 return arch_filter_pgprot(ret);
114 EXPORT_SYMBOL(vm_get_page_prot);
116 static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
118 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
121 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
122 void vma_set_page_prot(struct vm_area_struct *vma)
124 unsigned long vm_flags = vma->vm_flags;
125 pgprot_t vm_page_prot;
127 vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
128 if (vma_wants_writenotify(vma, vm_page_prot)) {
129 vm_flags &= ~VM_SHARED;
130 vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
132 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
133 WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
137 * Requires inode->i_mapping->i_mmap_rwsem
139 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
140 struct file *file, struct address_space *mapping)
142 if (vma->vm_flags & VM_DENYWRITE)
143 atomic_inc(&file_inode(file)->i_writecount);
144 if (vma->vm_flags & VM_SHARED)
145 mapping_unmap_writable(mapping);
147 flush_dcache_mmap_lock(mapping);
148 vma_interval_tree_remove(vma, &mapping->i_mmap);
149 flush_dcache_mmap_unlock(mapping);
153 * Unlink a file-based vm structure from its interval tree, to hide
154 * vma from rmap and vmtruncate before freeing its page tables.
156 void unlink_file_vma(struct vm_area_struct *vma)
158 struct file *file = vma->vm_file;
161 struct address_space *mapping = file->f_mapping;
162 i_mmap_lock_write(mapping);
163 __remove_shared_vm_struct(vma, file, mapping);
164 i_mmap_unlock_write(mapping);
169 * Close a vm structure and free it, returning the next.
171 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
173 struct vm_area_struct *next = vma->vm_next;
176 if (vma->vm_ops && vma->vm_ops->close)
177 vma->vm_ops->close(vma);
180 mpol_put(vma_policy(vma));
185 static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags,
186 struct list_head *uf);
187 SYSCALL_DEFINE1(brk, unsigned long, brk)
189 unsigned long retval;
190 unsigned long newbrk, oldbrk, origbrk;
191 struct mm_struct *mm = current->mm;
192 struct vm_area_struct *next;
193 unsigned long min_brk;
195 bool downgraded = false;
198 if (down_write_killable(&mm->mmap_sem))
203 #ifdef CONFIG_COMPAT_BRK
205 * CONFIG_COMPAT_BRK can still be overridden by setting
206 * randomize_va_space to 2, which will still cause mm->start_brk
207 * to be arbitrarily shifted
209 if (current->brk_randomized)
210 min_brk = mm->start_brk;
212 min_brk = mm->end_data;
214 min_brk = mm->start_brk;
220 * Check against rlimit here. If this check is done later after the test
221 * of oldbrk with newbrk then it can escape the test and let the data
222 * segment grow beyond its set limit the in case where the limit is
223 * not page aligned -Ram Gupta
225 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
226 mm->end_data, mm->start_data))
229 newbrk = PAGE_ALIGN(brk);
230 oldbrk = PAGE_ALIGN(mm->brk);
231 if (oldbrk == newbrk) {
237 * Always allow shrinking brk.
238 * __do_munmap() may downgrade mmap_sem to read.
240 if (brk <= mm->brk) {
244 * mm->brk must to be protected by write mmap_sem so update it
245 * before downgrading mmap_sem. When __do_munmap() fails,
246 * mm->brk will be restored from origbrk.
249 ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true);
253 } else if (ret == 1) {
259 /* Check against existing mmap mappings. */
260 next = find_vma(mm, oldbrk);
261 if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
264 /* Ok, looks good - let it rip. */
265 if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0)
270 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
272 up_read(&mm->mmap_sem);
274 up_write(&mm->mmap_sem);
275 userfaultfd_unmap_complete(mm, &uf);
277 mm_populate(oldbrk, newbrk - oldbrk);
282 up_write(&mm->mmap_sem);
286 static inline unsigned long vma_compute_gap(struct vm_area_struct *vma)
288 unsigned long gap, prev_end;
291 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
292 * allow two stack_guard_gaps between them here, and when choosing
293 * an unmapped area; whereas when expanding we only require one.
294 * That's a little inconsistent, but keeps the code here simpler.
296 gap = vm_start_gap(vma);
298 prev_end = vm_end_gap(vma->vm_prev);
307 #ifdef CONFIG_DEBUG_VM_RB
308 static unsigned long vma_compute_subtree_gap(struct vm_area_struct *vma)
310 unsigned long max = vma_compute_gap(vma), subtree_gap;
311 if (vma->vm_rb.rb_left) {
312 subtree_gap = rb_entry(vma->vm_rb.rb_left,
313 struct vm_area_struct, vm_rb)->rb_subtree_gap;
314 if (subtree_gap > max)
317 if (vma->vm_rb.rb_right) {
318 subtree_gap = rb_entry(vma->vm_rb.rb_right,
319 struct vm_area_struct, vm_rb)->rb_subtree_gap;
320 if (subtree_gap > max)
326 static int browse_rb(struct mm_struct *mm)
328 struct rb_root *root = &mm->mm_rb;
329 int i = 0, j, bug = 0;
330 struct rb_node *nd, *pn = NULL;
331 unsigned long prev = 0, pend = 0;
333 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
334 struct vm_area_struct *vma;
335 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
336 if (vma->vm_start < prev) {
337 pr_emerg("vm_start %lx < prev %lx\n",
338 vma->vm_start, prev);
341 if (vma->vm_start < pend) {
342 pr_emerg("vm_start %lx < pend %lx\n",
343 vma->vm_start, pend);
346 if (vma->vm_start > vma->vm_end) {
347 pr_emerg("vm_start %lx > vm_end %lx\n",
348 vma->vm_start, vma->vm_end);
351 spin_lock(&mm->page_table_lock);
352 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
353 pr_emerg("free gap %lx, correct %lx\n",
355 vma_compute_subtree_gap(vma));
358 spin_unlock(&mm->page_table_lock);
361 prev = vma->vm_start;
365 for (nd = pn; nd; nd = rb_prev(nd))
368 pr_emerg("backwards %d, forwards %d\n", j, i);
374 static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
378 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
379 struct vm_area_struct *vma;
380 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
381 VM_BUG_ON_VMA(vma != ignore &&
382 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
387 static void validate_mm(struct mm_struct *mm)
391 unsigned long highest_address = 0;
392 struct vm_area_struct *vma = mm->mmap;
395 struct anon_vma *anon_vma = vma->anon_vma;
396 struct anon_vma_chain *avc;
399 anon_vma_lock_read(anon_vma);
400 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
401 anon_vma_interval_tree_verify(avc);
402 anon_vma_unlock_read(anon_vma);
405 highest_address = vm_end_gap(vma);
409 if (i != mm->map_count) {
410 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
413 if (highest_address != mm->highest_vm_end) {
414 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
415 mm->highest_vm_end, highest_address);
419 if (i != mm->map_count) {
421 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
424 VM_BUG_ON_MM(bug, mm);
427 #define validate_mm_rb(root, ignore) do { } while (0)
428 #define validate_mm(mm) do { } while (0)
431 RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks,
432 struct vm_area_struct, vm_rb,
433 unsigned long, rb_subtree_gap, vma_compute_gap)
436 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
437 * vma->vm_prev->vm_end values changed, without modifying the vma's position
440 static void vma_gap_update(struct vm_area_struct *vma)
443 * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created
444 * a callback function that does exactly what we want.
446 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
449 static inline void vma_rb_insert(struct vm_area_struct *vma,
450 struct rb_root *root)
452 /* All rb_subtree_gap values must be consistent prior to insertion */
453 validate_mm_rb(root, NULL);
455 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
458 static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
461 * Note rb_erase_augmented is a fairly large inline function,
462 * so make sure we instantiate it only once with our desired
463 * augmented rbtree callbacks.
465 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
468 static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
469 struct rb_root *root,
470 struct vm_area_struct *ignore)
473 * All rb_subtree_gap values must be consistent prior to erase,
474 * with the possible exception of the "next" vma being erased if
475 * next->vm_start was reduced.
477 validate_mm_rb(root, ignore);
479 __vma_rb_erase(vma, root);
482 static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
483 struct rb_root *root)
486 * All rb_subtree_gap values must be consistent prior to erase,
487 * with the possible exception of the vma being erased.
489 validate_mm_rb(root, vma);
491 __vma_rb_erase(vma, root);
495 * vma has some anon_vma assigned, and is already inserted on that
496 * anon_vma's interval trees.
498 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
499 * vma must be removed from the anon_vma's interval trees using
500 * anon_vma_interval_tree_pre_update_vma().
502 * After the update, the vma will be reinserted using
503 * anon_vma_interval_tree_post_update_vma().
505 * The entire update must be protected by exclusive mmap_sem and by
506 * the root anon_vma's mutex.
509 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
511 struct anon_vma_chain *avc;
513 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
514 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
518 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
520 struct anon_vma_chain *avc;
522 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
523 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
526 static int find_vma_links(struct mm_struct *mm, unsigned long addr,
527 unsigned long end, struct vm_area_struct **pprev,
528 struct rb_node ***rb_link, struct rb_node **rb_parent)
530 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
532 __rb_link = &mm->mm_rb.rb_node;
533 rb_prev = __rb_parent = NULL;
536 struct vm_area_struct *vma_tmp;
538 __rb_parent = *__rb_link;
539 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
541 if (vma_tmp->vm_end > addr) {
542 /* Fail if an existing vma overlaps the area */
543 if (vma_tmp->vm_start < end)
545 __rb_link = &__rb_parent->rb_left;
547 rb_prev = __rb_parent;
548 __rb_link = &__rb_parent->rb_right;
554 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
555 *rb_link = __rb_link;
556 *rb_parent = __rb_parent;
560 static unsigned long count_vma_pages_range(struct mm_struct *mm,
561 unsigned long addr, unsigned long end)
563 unsigned long nr_pages = 0;
564 struct vm_area_struct *vma;
566 /* Find first overlaping mapping */
567 vma = find_vma_intersection(mm, addr, end);
571 nr_pages = (min(end, vma->vm_end) -
572 max(addr, vma->vm_start)) >> PAGE_SHIFT;
574 /* Iterate over the rest of the overlaps */
575 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
576 unsigned long overlap_len;
578 if (vma->vm_start > end)
581 overlap_len = min(end, vma->vm_end) - vma->vm_start;
582 nr_pages += overlap_len >> PAGE_SHIFT;
588 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
589 struct rb_node **rb_link, struct rb_node *rb_parent)
591 /* Update tracking information for the gap following the new vma. */
593 vma_gap_update(vma->vm_next);
595 mm->highest_vm_end = vm_end_gap(vma);
598 * vma->vm_prev wasn't known when we followed the rbtree to find the
599 * correct insertion point for that vma. As a result, we could not
600 * update the vma vm_rb parents rb_subtree_gap values on the way down.
601 * So, we first insert the vma with a zero rb_subtree_gap value
602 * (to be consistent with what we did on the way down), and then
603 * immediately update the gap to the correct value. Finally we
604 * rebalance the rbtree after all augmented values have been set.
606 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
607 vma->rb_subtree_gap = 0;
609 vma_rb_insert(vma, &mm->mm_rb);
612 static void __vma_link_file(struct vm_area_struct *vma)
618 struct address_space *mapping = file->f_mapping;
620 if (vma->vm_flags & VM_DENYWRITE)
621 atomic_dec(&file_inode(file)->i_writecount);
622 if (vma->vm_flags & VM_SHARED)
623 atomic_inc(&mapping->i_mmap_writable);
625 flush_dcache_mmap_lock(mapping);
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;
648 i_mmap_lock_write(mapping);
651 __vma_link(mm, vma, prev, rb_link, rb_parent);
652 __vma_link_file(vma);
655 i_mmap_unlock_write(mapping);
662 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
663 * mm's list and rbtree. It has already been inserted into the interval tree.
665 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
667 struct vm_area_struct *prev;
668 struct rb_node **rb_link, *rb_parent;
670 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
671 &prev, &rb_link, &rb_parent))
673 __vma_link(mm, vma, prev, rb_link, rb_parent);
677 static __always_inline void __vma_unlink_common(struct mm_struct *mm,
678 struct vm_area_struct *vma,
679 struct vm_area_struct *prev,
681 struct vm_area_struct *ignore)
683 struct vm_area_struct *next;
685 vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
688 prev->vm_next = next;
692 prev->vm_next = next;
697 next->vm_prev = prev;
700 vmacache_invalidate(mm);
703 static inline void __vma_unlink_prev(struct mm_struct *mm,
704 struct vm_area_struct *vma,
705 struct vm_area_struct *prev)
707 __vma_unlink_common(mm, vma, prev, true, vma);
711 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
712 * is already present in an i_mmap tree without adjusting the tree.
713 * The following helper function should be used when such adjustments
714 * are necessary. The "insert" vma (if any) is to be inserted
715 * before we drop the necessary locks.
717 int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
718 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
719 struct vm_area_struct *expand)
721 struct mm_struct *mm = vma->vm_mm;
722 struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
723 struct address_space *mapping = NULL;
724 struct rb_root_cached *root = NULL;
725 struct anon_vma *anon_vma = NULL;
726 struct file *file = vma->vm_file;
727 bool start_changed = false, end_changed = false;
728 long adjust_next = 0;
731 if (next && !insert) {
732 struct vm_area_struct *exporter = NULL, *importer = NULL;
734 if (end >= next->vm_end) {
736 * vma expands, overlapping all the next, and
737 * perhaps the one after too (mprotect case 6).
738 * The only other cases that gets here are
739 * case 1, case 7 and case 8.
741 if (next == expand) {
743 * The only case where we don't expand "vma"
744 * and we expand "next" instead is case 8.
746 VM_WARN_ON(end != next->vm_end);
748 * remove_next == 3 means we're
749 * removing "vma" and that to do so we
750 * swapped "vma" and "next".
753 VM_WARN_ON(file != next->vm_file);
756 VM_WARN_ON(expand != vma);
758 * case 1, 6, 7, remove_next == 2 is case 6,
759 * remove_next == 1 is case 1 or 7.
761 remove_next = 1 + (end > next->vm_end);
762 VM_WARN_ON(remove_next == 2 &&
763 end != next->vm_next->vm_end);
764 VM_WARN_ON(remove_next == 1 &&
765 end != next->vm_end);
766 /* trim end to next, for case 6 first pass */
774 * If next doesn't have anon_vma, import from vma after
775 * next, if the vma overlaps with it.
777 if (remove_next == 2 && !next->anon_vma)
778 exporter = next->vm_next;
780 } else if (end > next->vm_start) {
782 * vma expands, overlapping part of the next:
783 * mprotect case 5 shifting the boundary up.
785 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
788 VM_WARN_ON(expand != importer);
789 } else if (end < vma->vm_end) {
791 * vma shrinks, and !insert tells it's not
792 * split_vma inserting another: so it must be
793 * mprotect case 4 shifting the boundary down.
795 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
798 VM_WARN_ON(expand != importer);
802 * Easily overlooked: when mprotect shifts the boundary,
803 * make sure the expanding vma has anon_vma set if the
804 * shrinking vma had, to cover any anon pages imported.
806 if (exporter && exporter->anon_vma && !importer->anon_vma) {
809 importer->anon_vma = exporter->anon_vma;
810 error = anon_vma_clone(importer, exporter);
816 vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
819 mapping = file->f_mapping;
820 root = &mapping->i_mmap;
821 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
824 uprobe_munmap(next, next->vm_start, next->vm_end);
826 i_mmap_lock_write(mapping);
829 * Put into interval tree now, so instantiated pages
830 * are visible to arm/parisc __flush_dcache_page
831 * throughout; but we cannot insert into address
832 * space until vma start or end is updated.
834 __vma_link_file(insert);
838 anon_vma = vma->anon_vma;
839 if (!anon_vma && adjust_next)
840 anon_vma = next->anon_vma;
842 VM_WARN_ON(adjust_next && next->anon_vma &&
843 anon_vma != next->anon_vma);
844 anon_vma_lock_write(anon_vma);
845 anon_vma_interval_tree_pre_update_vma(vma);
847 anon_vma_interval_tree_pre_update_vma(next);
851 flush_dcache_mmap_lock(mapping);
852 vma_interval_tree_remove(vma, root);
854 vma_interval_tree_remove(next, root);
857 if (start != vma->vm_start) {
858 vma->vm_start = start;
859 start_changed = true;
861 if (end != vma->vm_end) {
865 vma->vm_pgoff = pgoff;
867 next->vm_start += adjust_next << PAGE_SHIFT;
868 next->vm_pgoff += adjust_next;
873 vma_interval_tree_insert(next, root);
874 vma_interval_tree_insert(vma, root);
875 flush_dcache_mmap_unlock(mapping);
880 * vma_merge has merged next into vma, and needs
881 * us to remove next before dropping the locks.
883 if (remove_next != 3)
884 __vma_unlink_prev(mm, next, vma);
887 * vma is not before next if they've been
890 * pre-swap() next->vm_start was reduced so
891 * tell validate_mm_rb to ignore pre-swap()
892 * "next" (which is stored in post-swap()
895 __vma_unlink_common(mm, next, NULL, false, vma);
897 __remove_shared_vm_struct(next, file, mapping);
900 * split_vma has split insert from vma, and needs
901 * us to insert it before dropping the locks
902 * (it may either follow vma or precede it).
904 __insert_vm_struct(mm, insert);
910 mm->highest_vm_end = vm_end_gap(vma);
911 else if (!adjust_next)
912 vma_gap_update(next);
917 anon_vma_interval_tree_post_update_vma(vma);
919 anon_vma_interval_tree_post_update_vma(next);
920 anon_vma_unlock_write(anon_vma);
923 i_mmap_unlock_write(mapping);
934 uprobe_munmap(next, next->vm_start, next->vm_end);
938 anon_vma_merge(vma, next);
940 mpol_put(vma_policy(next));
943 * In mprotect's case 6 (see comments on vma_merge),
944 * we must remove another next too. It would clutter
945 * up the code too much to do both in one go.
947 if (remove_next != 3) {
949 * If "next" was removed and vma->vm_end was
950 * expanded (up) over it, in turn
951 * "next->vm_prev->vm_end" changed and the
952 * "vma->vm_next" gap must be updated.
957 * For the scope of the comment "next" and
958 * "vma" considered pre-swap(): if "vma" was
959 * removed, next->vm_start was expanded (down)
960 * over it and the "next" gap must be updated.
961 * Because of the swap() the post-swap() "vma"
962 * actually points to pre-swap() "next"
963 * (post-swap() "next" as opposed is now a
968 if (remove_next == 2) {
974 vma_gap_update(next);
977 * If remove_next == 2 we obviously can't
980 * If remove_next == 3 we can't reach this
981 * path because pre-swap() next is always not
982 * NULL. pre-swap() "next" is not being
983 * removed and its next->vm_end is not altered
984 * (and furthermore "end" already matches
985 * next->vm_end in remove_next == 3).
987 * We reach this only in the remove_next == 1
988 * case if the "next" vma that was removed was
989 * the highest vma of the mm. However in such
990 * case next->vm_end == "end" and the extended
991 * "vma" has vma->vm_end == next->vm_end so
992 * mm->highest_vm_end doesn't need any update
993 * in remove_next == 1 case.
995 VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
1007 * If the vma has a ->close operation then the driver probably needs to release
1008 * per-vma resources, so we don't attempt to merge those.
1010 static inline int is_mergeable_vma(struct vm_area_struct *vma,
1011 struct file *file, unsigned long vm_flags,
1012 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1015 * VM_SOFTDIRTY should not prevent from VMA merging, if we
1016 * match the flags but dirty bit -- the caller should mark
1017 * merged VMA as dirty. If dirty bit won't be excluded from
1018 * comparison, we increase pressure on the memory system forcing
1019 * the kernel to generate new VMAs when old one could be
1022 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
1024 if (vma->vm_file != file)
1026 if (vma->vm_ops && vma->vm_ops->close)
1028 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
1033 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
1034 struct anon_vma *anon_vma2,
1035 struct vm_area_struct *vma)
1038 * The list_is_singular() test is to avoid merging VMA cloned from
1039 * parents. This can improve scalability caused by anon_vma lock.
1041 if ((!anon_vma1 || !anon_vma2) && (!vma ||
1042 list_is_singular(&vma->anon_vma_chain)))
1044 return anon_vma1 == anon_vma2;
1048 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1049 * in front of (at a lower virtual address and file offset than) the vma.
1051 * We cannot merge two vmas if they have differently assigned (non-NULL)
1052 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1054 * We don't check here for the merged mmap wrapping around the end of pagecache
1055 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1056 * wrap, nor mmaps which cover the final page at index -1UL.
1059 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
1060 struct anon_vma *anon_vma, struct file *file,
1062 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1064 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
1065 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1066 if (vma->vm_pgoff == vm_pgoff)
1073 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1074 * beyond (at a higher virtual address and file offset than) the vma.
1076 * We cannot merge two vmas if they have differently assigned (non-NULL)
1077 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1080 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
1081 struct anon_vma *anon_vma, struct file *file,
1083 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1085 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
1086 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1088 vm_pglen = vma_pages(vma);
1089 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1096 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1097 * whether that can be merged with its predecessor or its successor.
1098 * Or both (it neatly fills a hole).
1100 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1101 * certain not to be mapped by the time vma_merge is called; but when
1102 * called for mprotect, it is certain to be already mapped (either at
1103 * an offset within prev, or at the start of next), and the flags of
1104 * this area are about to be changed to vm_flags - and the no-change
1105 * case has already been eliminated.
1107 * The following mprotect cases have to be considered, where AAAA is
1108 * the area passed down from mprotect_fixup, never extending beyond one
1109 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1111 * AAAA AAAA AAAA AAAA
1112 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1113 * cannot merge might become might become might become
1114 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1115 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1116 * mremap move: PPPPXXXXXXXX 8
1118 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1119 * might become case 1 below case 2 below case 3 below
1121 * It is important for case 8 that the vma NNNN overlapping the
1122 * region AAAA is never going to extended over XXXX. Instead XXXX must
1123 * be extended in region AAAA and NNNN must be removed. This way in
1124 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1125 * rmap_locks, the properties of the merged vma will be already
1126 * correct for the whole merged range. Some of those properties like
1127 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1128 * be correct for the whole merged range immediately after the
1129 * rmap_locks are released. Otherwise if XXXX would be removed and
1130 * NNNN would be extended over the XXXX range, remove_migration_ptes
1131 * or other rmap walkers (if working on addresses beyond the "end"
1132 * parameter) may establish ptes with the wrong permissions of NNNN
1133 * instead of the right permissions of XXXX.
1135 struct vm_area_struct *vma_merge(struct mm_struct *mm,
1136 struct vm_area_struct *prev, unsigned long addr,
1137 unsigned long end, unsigned long vm_flags,
1138 struct anon_vma *anon_vma, struct file *file,
1139 pgoff_t pgoff, struct mempolicy *policy,
1140 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1142 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1143 struct vm_area_struct *area, *next;
1147 * We later require that vma->vm_flags == vm_flags,
1148 * so this tests vma->vm_flags & VM_SPECIAL, too.
1150 if (vm_flags & VM_SPECIAL)
1154 next = prev->vm_next;
1158 if (area && area->vm_end == end) /* cases 6, 7, 8 */
1159 next = next->vm_next;
1161 /* verify some invariant that must be enforced by the caller */
1162 VM_WARN_ON(prev && addr <= prev->vm_start);
1163 VM_WARN_ON(area && end > area->vm_end);
1164 VM_WARN_ON(addr >= end);
1167 * Can it merge with the predecessor?
1169 if (prev && prev->vm_end == addr &&
1170 mpol_equal(vma_policy(prev), policy) &&
1171 can_vma_merge_after(prev, vm_flags,
1172 anon_vma, file, pgoff,
1173 vm_userfaultfd_ctx)) {
1175 * OK, it can. Can we now merge in the successor as well?
1177 if (next && end == next->vm_start &&
1178 mpol_equal(policy, vma_policy(next)) &&
1179 can_vma_merge_before(next, vm_flags,
1182 vm_userfaultfd_ctx) &&
1183 is_mergeable_anon_vma(prev->anon_vma,
1184 next->anon_vma, NULL)) {
1186 err = __vma_adjust(prev, prev->vm_start,
1187 next->vm_end, prev->vm_pgoff, NULL,
1189 } else /* cases 2, 5, 7 */
1190 err = __vma_adjust(prev, prev->vm_start,
1191 end, prev->vm_pgoff, NULL, prev);
1194 khugepaged_enter_vma_merge(prev, vm_flags);
1199 * Can this new request be merged in front of next?
1201 if (next && end == next->vm_start &&
1202 mpol_equal(policy, vma_policy(next)) &&
1203 can_vma_merge_before(next, vm_flags,
1204 anon_vma, file, pgoff+pglen,
1205 vm_userfaultfd_ctx)) {
1206 if (prev && addr < prev->vm_end) /* case 4 */
1207 err = __vma_adjust(prev, prev->vm_start,
1208 addr, prev->vm_pgoff, NULL, next);
1209 else { /* cases 3, 8 */
1210 err = __vma_adjust(area, addr, next->vm_end,
1211 next->vm_pgoff - pglen, NULL, next);
1213 * In case 3 area is already equal to next and
1214 * this is a noop, but in case 8 "area" has
1215 * been removed and next was expanded over it.
1221 khugepaged_enter_vma_merge(area, vm_flags);
1229 * Rough compatbility check to quickly see if it's even worth looking
1230 * at sharing an anon_vma.
1232 * They need to have the same vm_file, and the flags can only differ
1233 * in things that mprotect may change.
1235 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1236 * we can merge the two vma's. For example, we refuse to merge a vma if
1237 * there is a vm_ops->close() function, because that indicates that the
1238 * driver is doing some kind of reference counting. But that doesn't
1239 * really matter for the anon_vma sharing case.
1241 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1243 return a->vm_end == b->vm_start &&
1244 mpol_equal(vma_policy(a), vma_policy(b)) &&
1245 a->vm_file == b->vm_file &&
1246 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
1247 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1251 * Do some basic sanity checking to see if we can re-use the anon_vma
1252 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1253 * the same as 'old', the other will be the new one that is trying
1254 * to share the anon_vma.
1256 * NOTE! This runs with mm_sem held for reading, so it is possible that
1257 * the anon_vma of 'old' is concurrently in the process of being set up
1258 * by another page fault trying to merge _that_. But that's ok: if it
1259 * is being set up, that automatically means that it will be a singleton
1260 * acceptable for merging, so we can do all of this optimistically. But
1261 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1263 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1264 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1265 * is to return an anon_vma that is "complex" due to having gone through
1268 * We also make sure that the two vma's are compatible (adjacent,
1269 * and with the same memory policies). That's all stable, even with just
1270 * a read lock on the mm_sem.
1272 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1274 if (anon_vma_compatible(a, b)) {
1275 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1277 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1284 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1285 * neighbouring vmas for a suitable anon_vma, before it goes off
1286 * to allocate a new anon_vma. It checks because a repetitive
1287 * sequence of mprotects and faults may otherwise lead to distinct
1288 * anon_vmas being allocated, preventing vma merge in subsequent
1291 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1293 struct anon_vma *anon_vma;
1294 struct vm_area_struct *near;
1296 near = vma->vm_next;
1300 anon_vma = reusable_anon_vma(near, vma, near);
1304 near = vma->vm_prev;
1308 anon_vma = reusable_anon_vma(near, near, vma);
1313 * There's no absolute need to look only at touching neighbours:
1314 * we could search further afield for "compatible" anon_vmas.
1315 * But it would probably just be a waste of time searching,
1316 * or lead to too many vmas hanging off the same anon_vma.
1317 * We're trying to allow mprotect remerging later on,
1318 * not trying to minimize memory used for anon_vmas.
1324 * If a hint addr is less than mmap_min_addr change hint to be as
1325 * low as possible but still greater than mmap_min_addr
1327 static inline unsigned long round_hint_to_min(unsigned long hint)
1330 if (((void *)hint != NULL) &&
1331 (hint < mmap_min_addr))
1332 return PAGE_ALIGN(mmap_min_addr);
1336 static inline int mlock_future_check(struct mm_struct *mm,
1337 unsigned long flags,
1340 unsigned long locked, lock_limit;
1342 /* mlock MCL_FUTURE? */
1343 if (flags & VM_LOCKED) {
1344 locked = len >> PAGE_SHIFT;
1345 locked += mm->locked_vm;
1346 lock_limit = rlimit(RLIMIT_MEMLOCK);
1347 lock_limit >>= PAGE_SHIFT;
1348 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1354 static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1356 if (S_ISREG(inode->i_mode))
1357 return MAX_LFS_FILESIZE;
1359 if (S_ISBLK(inode->i_mode))
1360 return MAX_LFS_FILESIZE;
1362 if (S_ISSOCK(inode->i_mode))
1363 return MAX_LFS_FILESIZE;
1365 /* Special "we do even unsigned file positions" case */
1366 if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1369 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1373 static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1374 unsigned long pgoff, unsigned long len)
1376 u64 maxsize = file_mmap_size_max(file, inode);
1378 if (maxsize && len > maxsize)
1381 if (pgoff > maxsize >> PAGE_SHIFT)
1387 * The caller must hold down_write(¤t->mm->mmap_sem).
1389 unsigned long do_mmap(struct file *file, unsigned long addr,
1390 unsigned long len, unsigned long prot,
1391 unsigned long flags, vm_flags_t vm_flags,
1392 unsigned long pgoff, unsigned long *populate,
1393 struct list_head *uf)
1395 struct mm_struct *mm = current->mm;
1404 * Does the application expect PROT_READ to imply PROT_EXEC?
1406 * (the exception is when the underlying filesystem is noexec
1407 * mounted, in which case we dont add PROT_EXEC.)
1409 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1410 if (!(file && path_noexec(&file->f_path)))
1413 /* force arch specific MAP_FIXED handling in get_unmapped_area */
1414 if (flags & MAP_FIXED_NOREPLACE)
1417 if (!(flags & MAP_FIXED))
1418 addr = round_hint_to_min(addr);
1420 /* Careful about overflows.. */
1421 len = PAGE_ALIGN(len);
1425 /* offset overflow? */
1426 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1429 /* Too many mappings? */
1430 if (mm->map_count > sysctl_max_map_count)
1433 /* Obtain the address to map to. we verify (or select) it and ensure
1434 * that it represents a valid section of the address space.
1436 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1437 if (offset_in_page(addr))
1440 if (flags & MAP_FIXED_NOREPLACE) {
1441 struct vm_area_struct *vma = find_vma(mm, addr);
1443 if (vma && vma->vm_start < addr + len)
1447 if (prot == PROT_EXEC) {
1448 pkey = execute_only_pkey(mm);
1453 /* Do simple checking here so the lower-level routines won't have
1454 * to. we assume access permissions have been handled by the open
1455 * of the memory object, so we don't do any here.
1457 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1458 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1460 if (flags & MAP_LOCKED)
1461 if (!can_do_mlock())
1464 if (mlock_future_check(mm, vm_flags, len))
1468 struct inode *inode = file_inode(file);
1469 unsigned long flags_mask;
1471 if (!file_mmap_ok(file, inode, pgoff, len))
1474 flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
1476 switch (flags & MAP_TYPE) {
1479 * Force use of MAP_SHARED_VALIDATE with non-legacy
1480 * flags. E.g. MAP_SYNC is dangerous to use with
1481 * MAP_SHARED as you don't know which consistency model
1482 * you will get. We silently ignore unsupported flags
1483 * with MAP_SHARED to preserve backward compatibility.
1485 flags &= LEGACY_MAP_MASK;
1487 case MAP_SHARED_VALIDATE:
1488 if (flags & ~flags_mask)
1490 if (prot & PROT_WRITE) {
1491 if (!(file->f_mode & FMODE_WRITE))
1493 if (IS_SWAPFILE(file->f_mapping->host))
1498 * Make sure we don't allow writing to an append-only
1501 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1505 * Make sure there are no mandatory locks on the file.
1507 if (locks_verify_locked(file))
1510 vm_flags |= VM_SHARED | VM_MAYSHARE;
1511 if (!(file->f_mode & FMODE_WRITE))
1512 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1516 if (!(file->f_mode & FMODE_READ))
1518 if (path_noexec(&file->f_path)) {
1519 if (vm_flags & VM_EXEC)
1521 vm_flags &= ~VM_MAYEXEC;
1524 if (!file->f_op->mmap)
1526 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1534 switch (flags & MAP_TYPE) {
1536 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1542 vm_flags |= VM_SHARED | VM_MAYSHARE;
1546 * Set pgoff according to addr for anon_vma.
1548 pgoff = addr >> PAGE_SHIFT;
1556 * Set 'VM_NORESERVE' if we should not account for the
1557 * memory use of this mapping.
1559 if (flags & MAP_NORESERVE) {
1560 /* We honor MAP_NORESERVE if allowed to overcommit */
1561 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1562 vm_flags |= VM_NORESERVE;
1564 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1565 if (file && is_file_hugepages(file))
1566 vm_flags |= VM_NORESERVE;
1569 addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
1570 if (!IS_ERR_VALUE(addr) &&
1571 ((vm_flags & VM_LOCKED) ||
1572 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1577 unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1578 unsigned long prot, unsigned long flags,
1579 unsigned long fd, unsigned long pgoff)
1581 struct file *file = NULL;
1582 unsigned long retval;
1584 if (!(flags & MAP_ANONYMOUS)) {
1585 audit_mmap_fd(fd, flags);
1589 if (is_file_hugepages(file))
1590 len = ALIGN(len, huge_page_size(hstate_file(file)));
1592 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1594 } else if (flags & MAP_HUGETLB) {
1595 struct user_struct *user = NULL;
1598 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1602 len = ALIGN(len, huge_page_size(hs));
1604 * VM_NORESERVE is used because the reservations will be
1605 * taken when vm_ops->mmap() is called
1606 * A dummy user value is used because we are not locking
1607 * memory so no accounting is necessary
1609 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1611 &user, HUGETLB_ANONHUGE_INODE,
1612 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1614 return PTR_ERR(file);
1617 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1619 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1626 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1627 unsigned long, prot, unsigned long, flags,
1628 unsigned long, fd, unsigned long, pgoff)
1630 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1633 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1634 struct mmap_arg_struct {
1638 unsigned long flags;
1640 unsigned long offset;
1643 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1645 struct mmap_arg_struct a;
1647 if (copy_from_user(&a, arg, sizeof(a)))
1649 if (offset_in_page(a.offset))
1652 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1653 a.offset >> PAGE_SHIFT);
1655 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1658 * Some shared mappings will want the pages marked read-only
1659 * to track write events. If so, we'll downgrade vm_page_prot
1660 * to the private version (using protection_map[] without the
1663 int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
1665 vm_flags_t vm_flags = vma->vm_flags;
1666 const struct vm_operations_struct *vm_ops = vma->vm_ops;
1668 /* If it was private or non-writable, the write bit is already clear */
1669 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1672 /* The backer wishes to know when pages are first written to? */
1673 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
1676 /* The open routine did something to the protections that pgprot_modify
1677 * won't preserve? */
1678 if (pgprot_val(vm_page_prot) !=
1679 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
1683 * Do we need to track softdirty? hugetlb does not support softdirty
1686 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY) &&
1687 !is_vm_hugetlb_page(vma))
1690 /* Specialty mapping? */
1691 if (vm_flags & VM_PFNMAP)
1694 /* Can the mapping track the dirty pages? */
1695 return vma->vm_file && vma->vm_file->f_mapping &&
1696 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1700 * We account for memory if it's a private writeable mapping,
1701 * not hugepages and VM_NORESERVE wasn't set.
1703 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1706 * hugetlb has its own accounting separate from the core VM
1707 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1709 if (file && is_file_hugepages(file))
1712 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1715 unsigned long mmap_region(struct file *file, unsigned long addr,
1716 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
1717 struct list_head *uf)
1719 struct mm_struct *mm = current->mm;
1720 struct vm_area_struct *vma, *prev;
1722 struct rb_node **rb_link, *rb_parent;
1723 unsigned long charged = 0;
1725 /* Check against address space limit. */
1726 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
1727 unsigned long nr_pages;
1730 * MAP_FIXED may remove pages of mappings that intersects with
1731 * requested mapping. Account for the pages it would unmap.
1733 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1735 if (!may_expand_vm(mm, vm_flags,
1736 (len >> PAGE_SHIFT) - nr_pages))
1740 /* Clear old maps */
1741 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1743 if (do_munmap(mm, addr, len, uf))
1748 * Private writable mapping: check memory availability
1750 if (accountable_mapping(file, vm_flags)) {
1751 charged = len >> PAGE_SHIFT;
1752 if (security_vm_enough_memory_mm(mm, charged))
1754 vm_flags |= VM_ACCOUNT;
1758 * Can we just expand an old mapping?
1760 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1761 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
1766 * Determine the object being mapped and call the appropriate
1767 * specific mapper. the address has already been validated, but
1768 * not unmapped, but the maps are removed from the list.
1770 vma = vm_area_alloc(mm);
1776 vma->vm_start = addr;
1777 vma->vm_end = addr + len;
1778 vma->vm_flags = vm_flags;
1779 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1780 vma->vm_pgoff = pgoff;
1783 if (vm_flags & VM_DENYWRITE) {
1784 error = deny_write_access(file);
1788 if (vm_flags & VM_SHARED) {
1789 error = mapping_map_writable(file->f_mapping);
1791 goto allow_write_and_free_vma;
1794 /* ->mmap() can change vma->vm_file, but must guarantee that
1795 * vma_link() below can deny write-access if VM_DENYWRITE is set
1796 * and map writably if VM_SHARED is set. This usually means the
1797 * new file must not have been exposed to user-space, yet.
1799 vma->vm_file = get_file(file);
1800 error = call_mmap(file, vma);
1802 goto unmap_and_free_vma;
1804 /* Can addr have changed??
1806 * Answer: Yes, several device drivers can do it in their
1807 * f_op->mmap method. -DaveM
1808 * Bug: If addr is changed, prev, rb_link, rb_parent should
1809 * be updated for vma_link()
1811 WARN_ON_ONCE(addr != vma->vm_start);
1813 addr = vma->vm_start;
1814 vm_flags = vma->vm_flags;
1815 } else if (vm_flags & VM_SHARED) {
1816 error = shmem_zero_setup(vma);
1820 vma_set_anonymous(vma);
1823 vma_link(mm, vma, prev, rb_link, rb_parent);
1824 /* Once vma denies write, undo our temporary denial count */
1826 if (vm_flags & VM_SHARED)
1827 mapping_unmap_writable(file->f_mapping);
1828 if (vm_flags & VM_DENYWRITE)
1829 allow_write_access(file);
1831 file = vma->vm_file;
1833 perf_event_mmap(vma);
1835 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
1836 if (vm_flags & VM_LOCKED) {
1837 if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) ||
1838 is_vm_hugetlb_page(vma) ||
1839 vma == get_gate_vma(current->mm))
1840 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
1842 mm->locked_vm += (len >> PAGE_SHIFT);
1849 * New (or expanded) vma always get soft dirty status.
1850 * Otherwise user-space soft-dirty page tracker won't
1851 * be able to distinguish situation when vma area unmapped,
1852 * then new mapped in-place (which must be aimed as
1853 * a completely new data area).
1855 vma->vm_flags |= VM_SOFTDIRTY;
1857 vma_set_page_prot(vma);
1862 vma->vm_file = NULL;
1865 /* Undo any partial mapping done by a device driver. */
1866 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1867 if (vm_flags & VM_SHARED)
1868 mapping_unmap_writable(file->f_mapping);
1869 allow_write_and_free_vma:
1870 if (vm_flags & VM_DENYWRITE)
1871 allow_write_access(file);
1876 vm_unacct_memory(charged);
1880 unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1883 * We implement the search by looking for an rbtree node that
1884 * immediately follows a suitable gap. That is,
1885 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1886 * - gap_end = vma->vm_start >= info->low_limit + length;
1887 * - gap_end - gap_start >= length
1890 struct mm_struct *mm = current->mm;
1891 struct vm_area_struct *vma;
1892 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1894 /* Adjust search length to account for worst case alignment overhead */
1895 length = info->length + info->align_mask;
1896 if (length < info->length)
1899 /* Adjust search limits by the desired length */
1900 if (info->high_limit < length)
1902 high_limit = info->high_limit - length;
1904 if (info->low_limit > high_limit)
1906 low_limit = info->low_limit + length;
1908 /* Check if rbtree root looks promising */
1909 if (RB_EMPTY_ROOT(&mm->mm_rb))
1911 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1912 if (vma->rb_subtree_gap < length)
1916 /* Visit left subtree if it looks promising */
1917 gap_end = vm_start_gap(vma);
1918 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1919 struct vm_area_struct *left =
1920 rb_entry(vma->vm_rb.rb_left,
1921 struct vm_area_struct, vm_rb);
1922 if (left->rb_subtree_gap >= length) {
1928 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
1930 /* Check if current node has a suitable gap */
1931 if (gap_start > high_limit)
1933 if (gap_end >= low_limit &&
1934 gap_end > gap_start && gap_end - gap_start >= length)
1937 /* Visit right subtree if it looks promising */
1938 if (vma->vm_rb.rb_right) {
1939 struct vm_area_struct *right =
1940 rb_entry(vma->vm_rb.rb_right,
1941 struct vm_area_struct, vm_rb);
1942 if (right->rb_subtree_gap >= length) {
1948 /* Go back up the rbtree to find next candidate node */
1950 struct rb_node *prev = &vma->vm_rb;
1951 if (!rb_parent(prev))
1953 vma = rb_entry(rb_parent(prev),
1954 struct vm_area_struct, vm_rb);
1955 if (prev == vma->vm_rb.rb_left) {
1956 gap_start = vm_end_gap(vma->vm_prev);
1957 gap_end = vm_start_gap(vma);
1964 /* Check highest gap, which does not precede any rbtree node */
1965 gap_start = mm->highest_vm_end;
1966 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1967 if (gap_start > high_limit)
1971 /* We found a suitable gap. Clip it with the original low_limit. */
1972 if (gap_start < info->low_limit)
1973 gap_start = info->low_limit;
1975 /* Adjust gap address to the desired alignment */
1976 gap_start += (info->align_offset - gap_start) & info->align_mask;
1978 VM_BUG_ON(gap_start + info->length > info->high_limit);
1979 VM_BUG_ON(gap_start + info->length > gap_end);
1983 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1985 struct mm_struct *mm = current->mm;
1986 struct vm_area_struct *vma;
1987 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1989 /* Adjust search length to account for worst case alignment overhead */
1990 length = info->length + info->align_mask;
1991 if (length < info->length)
1995 * Adjust search limits by the desired length.
1996 * See implementation comment at top of unmapped_area().
1998 gap_end = info->high_limit;
1999 if (gap_end < length)
2001 high_limit = gap_end - length;
2003 if (info->low_limit > high_limit)
2005 low_limit = info->low_limit + length;
2007 /* Check highest gap, which does not precede any rbtree node */
2008 gap_start = mm->highest_vm_end;
2009 if (gap_start <= high_limit)
2012 /* Check if rbtree root looks promising */
2013 if (RB_EMPTY_ROOT(&mm->mm_rb))
2015 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
2016 if (vma->rb_subtree_gap < length)
2020 /* Visit right subtree if it looks promising */
2021 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
2022 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
2023 struct vm_area_struct *right =
2024 rb_entry(vma->vm_rb.rb_right,
2025 struct vm_area_struct, vm_rb);
2026 if (right->rb_subtree_gap >= length) {
2033 /* Check if current node has a suitable gap */
2034 gap_end = vm_start_gap(vma);
2035 if (gap_end < low_limit)
2037 if (gap_start <= high_limit &&
2038 gap_end > gap_start && gap_end - gap_start >= length)
2041 /* Visit left subtree if it looks promising */
2042 if (vma->vm_rb.rb_left) {
2043 struct vm_area_struct *left =
2044 rb_entry(vma->vm_rb.rb_left,
2045 struct vm_area_struct, vm_rb);
2046 if (left->rb_subtree_gap >= length) {
2052 /* Go back up the rbtree to find next candidate node */
2054 struct rb_node *prev = &vma->vm_rb;
2055 if (!rb_parent(prev))
2057 vma = rb_entry(rb_parent(prev),
2058 struct vm_area_struct, vm_rb);
2059 if (prev == vma->vm_rb.rb_right) {
2060 gap_start = vma->vm_prev ?
2061 vm_end_gap(vma->vm_prev) : 0;
2068 /* We found a suitable gap. Clip it with the original high_limit. */
2069 if (gap_end > info->high_limit)
2070 gap_end = info->high_limit;
2073 /* Compute highest gap address at the desired alignment */
2074 gap_end -= info->length;
2075 gap_end -= (gap_end - info->align_offset) & info->align_mask;
2077 VM_BUG_ON(gap_end < info->low_limit);
2078 VM_BUG_ON(gap_end < gap_start);
2083 /* Get an address range which is currently unmapped.
2084 * For shmat() with addr=0.
2086 * Ugly calling convention alert:
2087 * Return value with the low bits set means error value,
2089 * if (ret & ~PAGE_MASK)
2092 * This function "knows" that -ENOMEM has the bits set.
2094 #ifndef HAVE_ARCH_UNMAPPED_AREA
2096 arch_get_unmapped_area(struct file *filp, unsigned long addr,
2097 unsigned long len, unsigned long pgoff, unsigned long flags)
2099 struct mm_struct *mm = current->mm;
2100 struct vm_area_struct *vma, *prev;
2101 struct vm_unmapped_area_info info;
2102 const unsigned long mmap_end = arch_get_mmap_end(addr);
2104 if (len > mmap_end - mmap_min_addr)
2107 if (flags & MAP_FIXED)
2111 addr = PAGE_ALIGN(addr);
2112 vma = find_vma_prev(mm, addr, &prev);
2113 if (mmap_end - len >= addr && addr >= mmap_min_addr &&
2114 (!vma || addr + len <= vm_start_gap(vma)) &&
2115 (!prev || addr >= vm_end_gap(prev)))
2121 info.low_limit = mm->mmap_base;
2122 info.high_limit = mmap_end;
2123 info.align_mask = 0;
2124 info.align_offset = 0;
2125 return vm_unmapped_area(&info);
2130 * This mmap-allocator allocates new areas top-down from below the
2131 * stack's low limit (the base):
2133 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2135 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
2136 unsigned long len, unsigned long pgoff,
2137 unsigned long flags)
2139 struct vm_area_struct *vma, *prev;
2140 struct mm_struct *mm = current->mm;
2141 struct vm_unmapped_area_info info;
2142 const unsigned long mmap_end = arch_get_mmap_end(addr);
2144 /* requested length too big for entire address space */
2145 if (len > mmap_end - mmap_min_addr)
2148 if (flags & MAP_FIXED)
2151 /* requesting a specific address */
2153 addr = PAGE_ALIGN(addr);
2154 vma = find_vma_prev(mm, addr, &prev);
2155 if (mmap_end - len >= addr && addr >= mmap_min_addr &&
2156 (!vma || addr + len <= vm_start_gap(vma)) &&
2157 (!prev || addr >= vm_end_gap(prev)))
2161 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2163 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
2164 info.high_limit = arch_get_mmap_base(addr, mm->mmap_base);
2165 info.align_mask = 0;
2166 info.align_offset = 0;
2167 addr = vm_unmapped_area(&info);
2170 * A failed mmap() very likely causes application failure,
2171 * so fall back to the bottom-up function here. This scenario
2172 * can happen with large stack limits and large mmap()
2175 if (offset_in_page(addr)) {
2176 VM_BUG_ON(addr != -ENOMEM);
2178 info.low_limit = TASK_UNMAPPED_BASE;
2179 info.high_limit = mmap_end;
2180 addr = vm_unmapped_area(&info);
2188 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2189 unsigned long pgoff, unsigned long flags)
2191 unsigned long (*get_area)(struct file *, unsigned long,
2192 unsigned long, unsigned long, unsigned long);
2194 unsigned long error = arch_mmap_check(addr, len, flags);
2198 /* Careful about overflows.. */
2199 if (len > TASK_SIZE)
2202 get_area = current->mm->get_unmapped_area;
2204 if (file->f_op->get_unmapped_area)
2205 get_area = file->f_op->get_unmapped_area;
2206 } else if (flags & MAP_SHARED) {
2208 * mmap_region() will call shmem_zero_setup() to create a file,
2209 * so use shmem's get_unmapped_area in case it can be huge.
2210 * do_mmap_pgoff() will clear pgoff, so match alignment.
2213 get_area = shmem_get_unmapped_area;
2216 addr = get_area(file, addr, len, pgoff, flags);
2217 if (IS_ERR_VALUE(addr))
2220 if (addr > TASK_SIZE - len)
2222 if (offset_in_page(addr))
2225 error = security_mmap_addr(addr);
2226 return error ? error : addr;
2229 EXPORT_SYMBOL(get_unmapped_area);
2231 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2232 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
2234 struct rb_node *rb_node;
2235 struct vm_area_struct *vma;
2237 /* Check the cache first. */
2238 vma = vmacache_find(mm, addr);
2242 rb_node = mm->mm_rb.rb_node;
2245 struct vm_area_struct *tmp;
2247 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2249 if (tmp->vm_end > addr) {
2251 if (tmp->vm_start <= addr)
2253 rb_node = rb_node->rb_left;
2255 rb_node = rb_node->rb_right;
2259 vmacache_update(addr, vma);
2263 EXPORT_SYMBOL(find_vma);
2266 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2268 struct vm_area_struct *
2269 find_vma_prev(struct mm_struct *mm, unsigned long addr,
2270 struct vm_area_struct **pprev)
2272 struct vm_area_struct *vma;
2274 vma = find_vma(mm, addr);
2276 *pprev = vma->vm_prev;
2278 struct rb_node *rb_node = rb_last(&mm->mm_rb);
2280 *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL;
2286 * Verify that the stack growth is acceptable and
2287 * update accounting. This is shared with both the
2288 * grow-up and grow-down cases.
2290 static int acct_stack_growth(struct vm_area_struct *vma,
2291 unsigned long size, unsigned long grow)
2293 struct mm_struct *mm = vma->vm_mm;
2294 unsigned long new_start;
2296 /* address space limit tests */
2297 if (!may_expand_vm(mm, vma->vm_flags, grow))
2300 /* Stack limit test */
2301 if (size > rlimit(RLIMIT_STACK))
2304 /* mlock limit tests */
2305 if (vma->vm_flags & VM_LOCKED) {
2306 unsigned long locked;
2307 unsigned long limit;
2308 locked = mm->locked_vm + grow;
2309 limit = rlimit(RLIMIT_MEMLOCK);
2310 limit >>= PAGE_SHIFT;
2311 if (locked > limit && !capable(CAP_IPC_LOCK))
2315 /* Check to ensure the stack will not grow into a hugetlb-only region */
2316 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2318 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2322 * Overcommit.. This must be the final test, as it will
2323 * update security statistics.
2325 if (security_vm_enough_memory_mm(mm, grow))
2331 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2333 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2334 * vma is the last one with address > vma->vm_end. Have to extend vma.
2336 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2338 struct mm_struct *mm = vma->vm_mm;
2339 struct vm_area_struct *next;
2340 unsigned long gap_addr;
2343 if (!(vma->vm_flags & VM_GROWSUP))
2346 /* Guard against exceeding limits of the address space. */
2347 address &= PAGE_MASK;
2348 if (address >= (TASK_SIZE & PAGE_MASK))
2350 address += PAGE_SIZE;
2352 /* Enforce stack_guard_gap */
2353 gap_addr = address + stack_guard_gap;
2355 /* Guard against overflow */
2356 if (gap_addr < address || gap_addr > TASK_SIZE)
2357 gap_addr = TASK_SIZE;
2359 next = vma->vm_next;
2360 if (next && next->vm_start < gap_addr &&
2361 (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
2362 if (!(next->vm_flags & VM_GROWSUP))
2364 /* Check that both stack segments have the same anon_vma? */
2367 /* We must make sure the anon_vma is allocated. */
2368 if (unlikely(anon_vma_prepare(vma)))
2372 * vma->vm_start/vm_end cannot change under us because the caller
2373 * is required to hold the mmap_sem in read mode. We need the
2374 * anon_vma lock to serialize against concurrent expand_stacks.
2376 anon_vma_lock_write(vma->anon_vma);
2378 /* Somebody else might have raced and expanded it already */
2379 if (address > vma->vm_end) {
2380 unsigned long size, grow;
2382 size = address - vma->vm_start;
2383 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2386 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2387 error = acct_stack_growth(vma, size, grow);
2390 * vma_gap_update() doesn't support concurrent
2391 * updates, but we only hold a shared mmap_sem
2392 * lock here, so we need to protect against
2393 * concurrent vma expansions.
2394 * anon_vma_lock_write() doesn't help here, as
2395 * we don't guarantee that all growable vmas
2396 * in a mm share the same root anon vma.
2397 * So, we reuse mm->page_table_lock to guard
2398 * against concurrent vma expansions.
2400 spin_lock(&mm->page_table_lock);
2401 if (vma->vm_flags & VM_LOCKED)
2402 mm->locked_vm += grow;
2403 vm_stat_account(mm, vma->vm_flags, grow);
2404 anon_vma_interval_tree_pre_update_vma(vma);
2405 vma->vm_end = address;
2406 anon_vma_interval_tree_post_update_vma(vma);
2408 vma_gap_update(vma->vm_next);
2410 mm->highest_vm_end = vm_end_gap(vma);
2411 spin_unlock(&mm->page_table_lock);
2413 perf_event_mmap(vma);
2417 anon_vma_unlock_write(vma->anon_vma);
2418 khugepaged_enter_vma_merge(vma, vma->vm_flags);
2422 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2425 * vma is the first one with address < vma->vm_start. Have to extend vma.
2427 int expand_downwards(struct vm_area_struct *vma,
2428 unsigned long address)
2430 struct mm_struct *mm = vma->vm_mm;
2431 struct vm_area_struct *prev;
2434 address &= PAGE_MASK;
2435 if (address < mmap_min_addr)
2438 /* Enforce stack_guard_gap */
2439 prev = vma->vm_prev;
2440 /* Check that both stack segments have the same anon_vma? */
2441 if (prev && !(prev->vm_flags & VM_GROWSDOWN) &&
2442 (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
2443 if (address - prev->vm_end < stack_guard_gap)
2447 /* We must make sure the anon_vma is allocated. */
2448 if (unlikely(anon_vma_prepare(vma)))
2452 * vma->vm_start/vm_end cannot change under us because the caller
2453 * is required to hold the mmap_sem in read mode. We need the
2454 * anon_vma lock to serialize against concurrent expand_stacks.
2456 anon_vma_lock_write(vma->anon_vma);
2458 /* Somebody else might have raced and expanded it already */
2459 if (address < vma->vm_start) {
2460 unsigned long size, grow;
2462 size = vma->vm_end - address;
2463 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2466 if (grow <= vma->vm_pgoff) {
2467 error = acct_stack_growth(vma, size, grow);
2470 * vma_gap_update() doesn't support concurrent
2471 * updates, but we only hold a shared mmap_sem
2472 * lock here, so we need to protect against
2473 * concurrent vma expansions.
2474 * anon_vma_lock_write() doesn't help here, as
2475 * we don't guarantee that all growable vmas
2476 * in a mm share the same root anon vma.
2477 * So, we reuse mm->page_table_lock to guard
2478 * against concurrent vma expansions.
2480 spin_lock(&mm->page_table_lock);
2481 if (vma->vm_flags & VM_LOCKED)
2482 mm->locked_vm += grow;
2483 vm_stat_account(mm, vma->vm_flags, grow);
2484 anon_vma_interval_tree_pre_update_vma(vma);
2485 vma->vm_start = address;
2486 vma->vm_pgoff -= grow;
2487 anon_vma_interval_tree_post_update_vma(vma);
2488 vma_gap_update(vma);
2489 spin_unlock(&mm->page_table_lock);
2491 perf_event_mmap(vma);
2495 anon_vma_unlock_write(vma->anon_vma);
2496 khugepaged_enter_vma_merge(vma, vma->vm_flags);
2501 /* enforced gap between the expanding stack and other mappings. */
2502 unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2504 static int __init cmdline_parse_stack_guard_gap(char *p)
2509 val = simple_strtoul(p, &endptr, 10);
2511 stack_guard_gap = val << PAGE_SHIFT;
2515 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2517 #ifdef CONFIG_STACK_GROWSUP
2518 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2520 return expand_upwards(vma, address);
2523 struct vm_area_struct *
2524 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2526 struct vm_area_struct *vma, *prev;
2529 vma = find_vma_prev(mm, addr, &prev);
2530 if (vma && (vma->vm_start <= addr))
2532 /* don't alter vm_end if the coredump is running */
2533 if (!prev || !mmget_still_valid(mm) || expand_stack(prev, addr))
2535 if (prev->vm_flags & VM_LOCKED)
2536 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
2540 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2542 return expand_downwards(vma, address);
2545 struct vm_area_struct *
2546 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2548 struct vm_area_struct *vma;
2549 unsigned long start;
2552 vma = find_vma(mm, addr);
2555 if (vma->vm_start <= addr)
2557 if (!(vma->vm_flags & VM_GROWSDOWN))
2559 /* don't alter vm_start if the coredump is running */
2560 if (!mmget_still_valid(mm))
2562 start = vma->vm_start;
2563 if (expand_stack(vma, addr))
2565 if (vma->vm_flags & VM_LOCKED)
2566 populate_vma_page_range(vma, addr, start, NULL);
2571 EXPORT_SYMBOL_GPL(find_extend_vma);
2574 * Ok - we have the memory areas we should free on the vma list,
2575 * so release them, and do the vma updates.
2577 * Called with the mm semaphore held.
2579 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2581 unsigned long nr_accounted = 0;
2583 /* Update high watermark before we lower total_vm */
2584 update_hiwater_vm(mm);
2586 long nrpages = vma_pages(vma);
2588 if (vma->vm_flags & VM_ACCOUNT)
2589 nr_accounted += nrpages;
2590 vm_stat_account(mm, vma->vm_flags, -nrpages);
2591 vma = remove_vma(vma);
2593 vm_unacct_memory(nr_accounted);
2598 * Get rid of page table information in the indicated region.
2600 * Called with the mm semaphore held.
2602 static void unmap_region(struct mm_struct *mm,
2603 struct vm_area_struct *vma, struct vm_area_struct *prev,
2604 unsigned long start, unsigned long end)
2606 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
2607 struct mmu_gather tlb;
2608 struct vm_area_struct *cur_vma;
2611 tlb_gather_mmu(&tlb, mm, start, end);
2612 update_hiwater_rss(mm);
2613 unmap_vmas(&tlb, vma, start, end);
2616 * Ensure we have no stale TLB entries by the time this mapping is
2617 * removed from the rmap.
2618 * Note that we don't have to worry about nested flushes here because
2619 * we're holding the mm semaphore for removing the mapping - so any
2620 * concurrent flush in this region has to be coming through the rmap,
2621 * and we synchronize against that using the rmap lock.
2623 for (cur_vma = vma; cur_vma; cur_vma = cur_vma->vm_next) {
2624 if ((cur_vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) != 0) {
2625 tlb_flush_mmu(&tlb);
2630 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2631 next ? next->vm_start : USER_PGTABLES_CEILING);
2632 tlb_finish_mmu(&tlb, start, end);
2636 * Create a list of vma's touched by the unmap, removing them from the mm's
2637 * vma list as we go..
2640 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2641 struct vm_area_struct *prev, unsigned long end)
2643 struct vm_area_struct **insertion_point;
2644 struct vm_area_struct *tail_vma = NULL;
2646 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2647 vma->vm_prev = NULL;
2649 vma_rb_erase(vma, &mm->mm_rb);
2653 } while (vma && vma->vm_start < end);
2654 *insertion_point = vma;
2656 vma->vm_prev = prev;
2657 vma_gap_update(vma);
2659 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
2660 tail_vma->vm_next = NULL;
2662 /* Kill the cache */
2663 vmacache_invalidate(mm);
2666 * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or
2667 * VM_GROWSUP VMA. Such VMAs can change their size under
2668 * down_read(mmap_lock) and collide with the VMA we are about to unmap.
2670 if (vma && (vma->vm_flags & VM_GROWSDOWN))
2672 if (prev && (prev->vm_flags & VM_GROWSUP))
2678 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2679 * has already been checked or doesn't make sense to fail.
2681 int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2682 unsigned long addr, int new_below)
2684 struct vm_area_struct *new;
2687 if (vma->vm_ops && vma->vm_ops->split) {
2688 err = vma->vm_ops->split(vma, addr);
2693 new = vm_area_dup(vma);
2700 new->vm_start = addr;
2701 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2704 err = vma_dup_policy(vma, new);
2708 err = anon_vma_clone(new, vma);
2713 get_file(new->vm_file);
2715 if (new->vm_ops && new->vm_ops->open)
2716 new->vm_ops->open(new);
2719 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2720 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2722 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2728 /* Clean everything up if vma_adjust failed. */
2729 if (new->vm_ops && new->vm_ops->close)
2730 new->vm_ops->close(new);
2733 unlink_anon_vmas(new);
2735 mpol_put(vma_policy(new));
2742 * Split a vma into two pieces at address 'addr', a new vma is allocated
2743 * either for the first part or the tail.
2745 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2746 unsigned long addr, int new_below)
2748 if (mm->map_count >= sysctl_max_map_count)
2751 return __split_vma(mm, vma, addr, new_below);
2754 /* Munmap is split into 2 main parts -- this part which finds
2755 * what needs doing, and the areas themselves, which do the
2756 * work. This now handles partial unmappings.
2757 * Jeremy Fitzhardinge <jeremy@goop.org>
2759 int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2760 struct list_head *uf, bool downgrade)
2763 struct vm_area_struct *vma, *prev, *last;
2765 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2768 len = PAGE_ALIGN(len);
2774 * arch_unmap() might do unmaps itself. It must be called
2775 * and finish any rbtree manipulation before this code
2776 * runs and also starts to manipulate the rbtree.
2778 arch_unmap(mm, start, end);
2780 /* Find the first overlapping VMA */
2781 vma = find_vma(mm, start);
2784 prev = vma->vm_prev;
2785 /* we have start < vma->vm_end */
2787 /* if it doesn't overlap, we have nothing.. */
2788 if (vma->vm_start >= end)
2792 * If we need to split any vma, do it now to save pain later.
2794 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2795 * unmapped vm_area_struct will remain in use: so lower split_vma
2796 * places tmp vma above, and higher split_vma places tmp vma below.
2798 if (start > vma->vm_start) {
2802 * Make sure that map_count on return from munmap() will
2803 * not exceed its limit; but let map_count go just above
2804 * its limit temporarily, to help free resources as expected.
2806 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2809 error = __split_vma(mm, vma, start, 0);
2815 /* Does it split the last one? */
2816 last = find_vma(mm, end);
2817 if (last && end > last->vm_start) {
2818 int error = __split_vma(mm, last, end, 1);
2822 vma = prev ? prev->vm_next : mm->mmap;
2826 * If userfaultfd_unmap_prep returns an error the vmas
2827 * will remain splitted, but userland will get a
2828 * highly unexpected error anyway. This is no
2829 * different than the case where the first of the two
2830 * __split_vma fails, but we don't undo the first
2831 * split, despite we could. This is unlikely enough
2832 * failure that it's not worth optimizing it for.
2834 int error = userfaultfd_unmap_prep(vma, start, end, uf);
2840 * unlock any mlock()ed ranges before detaching vmas
2842 if (mm->locked_vm) {
2843 struct vm_area_struct *tmp = vma;
2844 while (tmp && tmp->vm_start < end) {
2845 if (tmp->vm_flags & VM_LOCKED) {
2846 mm->locked_vm -= vma_pages(tmp);
2847 munlock_vma_pages_all(tmp);
2854 /* Detach vmas from rbtree */
2855 if (!detach_vmas_to_be_unmapped(mm, vma, prev, end))
2859 downgrade_write(&mm->mmap_sem);
2861 unmap_region(mm, vma, prev, start, end);
2863 /* Fix up all other VM information */
2864 remove_vma_list(mm, vma);
2866 return downgrade ? 1 : 0;
2869 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2870 struct list_head *uf)
2872 return __do_munmap(mm, start, len, uf, false);
2875 static int __vm_munmap(unsigned long start, size_t len, bool downgrade)
2878 struct mm_struct *mm = current->mm;
2881 if (down_write_killable(&mm->mmap_sem))
2884 ret = __do_munmap(mm, start, len, &uf, downgrade);
2886 * Returning 1 indicates mmap_sem is downgraded.
2887 * But 1 is not legal return value of vm_munmap() and munmap(), reset
2888 * it to 0 before return.
2891 up_read(&mm->mmap_sem);
2894 up_write(&mm->mmap_sem);
2896 userfaultfd_unmap_complete(mm, &uf);
2900 int vm_munmap(unsigned long start, size_t len)
2902 return __vm_munmap(start, len, false);
2904 EXPORT_SYMBOL(vm_munmap);
2906 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2908 addr = untagged_addr(addr);
2909 profile_munmap(addr);
2910 return __vm_munmap(addr, len, true);
2915 * Emulation of deprecated remap_file_pages() syscall.
2917 SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2918 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2921 struct mm_struct *mm = current->mm;
2922 struct vm_area_struct *vma;
2923 unsigned long populate = 0;
2924 unsigned long ret = -EINVAL;
2927 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n",
2928 current->comm, current->pid);
2932 start = start & PAGE_MASK;
2933 size = size & PAGE_MASK;
2935 if (start + size <= start)
2938 /* Does pgoff wrap? */
2939 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2942 if (down_write_killable(&mm->mmap_sem))
2945 vma = find_vma(mm, start);
2947 if (!vma || !(vma->vm_flags & VM_SHARED))
2950 if (start < vma->vm_start)
2953 if (start + size > vma->vm_end) {
2954 struct vm_area_struct *next;
2956 for (next = vma->vm_next; next; next = next->vm_next) {
2957 /* hole between vmas ? */
2958 if (next->vm_start != next->vm_prev->vm_end)
2961 if (next->vm_file != vma->vm_file)
2964 if (next->vm_flags != vma->vm_flags)
2967 if (start + size <= next->vm_end)
2975 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2976 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2977 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2979 flags &= MAP_NONBLOCK;
2980 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2981 if (vma->vm_flags & VM_LOCKED) {
2982 struct vm_area_struct *tmp;
2983 flags |= MAP_LOCKED;
2985 /* drop PG_Mlocked flag for over-mapped range */
2986 for (tmp = vma; tmp->vm_start >= start + size;
2987 tmp = tmp->vm_next) {
2989 * Split pmd and munlock page on the border
2992 vma_adjust_trans_huge(tmp, start, start + size, 0);
2994 munlock_vma_pages_range(tmp,
2995 max(tmp->vm_start, start),
2996 min(tmp->vm_end, start + size));
3000 file = get_file(vma->vm_file);
3001 ret = do_mmap_pgoff(vma->vm_file, start, size,
3002 prot, flags, pgoff, &populate, NULL);
3005 up_write(&mm->mmap_sem);
3007 mm_populate(ret, populate);
3008 if (!IS_ERR_VALUE(ret))
3014 * this is really a simplified "do_mmap". it only handles
3015 * anonymous maps. eventually we may be able to do some
3016 * brk-specific accounting here.
3018 static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf)
3020 struct mm_struct *mm = current->mm;
3021 struct vm_area_struct *vma, *prev;
3022 struct rb_node **rb_link, *rb_parent;
3023 pgoff_t pgoff = addr >> PAGE_SHIFT;
3026 /* Until we need other flags, refuse anything except VM_EXEC. */
3027 if ((flags & (~VM_EXEC)) != 0)
3029 flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
3031 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
3032 if (offset_in_page(error))
3035 error = mlock_future_check(mm, mm->def_flags, len);
3040 * Clear old maps. this also does some error checking for us
3042 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
3044 if (do_munmap(mm, addr, len, uf))
3048 /* Check against address space limits *after* clearing old maps... */
3049 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
3052 if (mm->map_count > sysctl_max_map_count)
3055 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
3058 /* Can we just expand an old private anonymous mapping? */
3059 vma = vma_merge(mm, prev, addr, addr + len, flags,
3060 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
3065 * create a vma struct for an anonymous mapping
3067 vma = vm_area_alloc(mm);
3069 vm_unacct_memory(len >> PAGE_SHIFT);
3073 vma_set_anonymous(vma);
3074 vma->vm_start = addr;
3075 vma->vm_end = addr + len;
3076 vma->vm_pgoff = pgoff;
3077 vma->vm_flags = flags;
3078 vma->vm_page_prot = vm_get_page_prot(flags);
3079 vma_link(mm, vma, prev, rb_link, rb_parent);
3081 perf_event_mmap(vma);
3082 mm->total_vm += len >> PAGE_SHIFT;
3083 mm->data_vm += len >> PAGE_SHIFT;
3084 if (flags & VM_LOCKED)
3085 mm->locked_vm += (len >> PAGE_SHIFT);
3086 vma->vm_flags |= VM_SOFTDIRTY;
3090 int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
3092 struct mm_struct *mm = current->mm;
3098 len = PAGE_ALIGN(request);
3104 if (down_write_killable(&mm->mmap_sem))
3107 ret = do_brk_flags(addr, len, flags, &uf);
3108 populate = ((mm->def_flags & VM_LOCKED) != 0);
3109 up_write(&mm->mmap_sem);
3110 userfaultfd_unmap_complete(mm, &uf);
3111 if (populate && !ret)
3112 mm_populate(addr, len);
3115 EXPORT_SYMBOL(vm_brk_flags);
3117 int vm_brk(unsigned long addr, unsigned long len)
3119 return vm_brk_flags(addr, len, 0);
3121 EXPORT_SYMBOL(vm_brk);
3123 /* Release all mmaps. */
3124 void exit_mmap(struct mm_struct *mm)
3126 struct mmu_gather tlb;
3127 struct vm_area_struct *vma;
3128 unsigned long nr_accounted = 0;
3130 /* mm's last user has gone, and its about to be pulled down */
3131 mmu_notifier_release(mm);
3133 if (unlikely(mm_is_oom_victim(mm))) {
3135 * Manually reap the mm to free as much memory as possible.
3136 * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3137 * this mm from further consideration. Taking mm->mmap_sem for
3138 * write after setting MMF_OOM_SKIP will guarantee that the oom
3139 * reaper will not run on this mm again after mmap_sem is
3142 * Nothing can be holding mm->mmap_sem here and the above call
3143 * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3144 * __oom_reap_task_mm() will not block.
3146 * This needs to be done before calling munlock_vma_pages_all(),
3147 * which clears VM_LOCKED, otherwise the oom reaper cannot
3150 (void)__oom_reap_task_mm(mm);
3152 set_bit(MMF_OOM_SKIP, &mm->flags);
3153 down_write(&mm->mmap_sem);
3154 up_write(&mm->mmap_sem);
3157 if (mm->locked_vm) {
3160 if (vma->vm_flags & VM_LOCKED)
3161 munlock_vma_pages_all(vma);
3169 if (!vma) /* Can happen if dup_mmap() received an OOM */
3174 tlb_gather_mmu(&tlb, mm, 0, -1);
3175 /* update_hiwater_rss(mm) here? but nobody should be looking */
3176 /* Use -1 here to ensure all VMAs in the mm are unmapped */
3177 unmap_vmas(&tlb, vma, 0, -1);
3178 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
3179 tlb_finish_mmu(&tlb, 0, -1);
3182 * Walk the list again, actually closing and freeing it,
3183 * with preemption enabled, without holding any MM locks.
3186 if (vma->vm_flags & VM_ACCOUNT)
3187 nr_accounted += vma_pages(vma);
3188 vma = remove_vma(vma);
3191 vm_unacct_memory(nr_accounted);
3194 /* Insert vm structure into process list sorted by address
3195 * and into the inode's i_mmap tree. If vm_file is non-NULL
3196 * then i_mmap_rwsem is taken here.
3198 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
3200 struct vm_area_struct *prev;
3201 struct rb_node **rb_link, *rb_parent;
3203 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3204 &prev, &rb_link, &rb_parent))
3206 if ((vma->vm_flags & VM_ACCOUNT) &&
3207 security_vm_enough_memory_mm(mm, vma_pages(vma)))
3211 * The vm_pgoff of a purely anonymous vma should be irrelevant
3212 * until its first write fault, when page's anon_vma and index
3213 * are set. But now set the vm_pgoff it will almost certainly
3214 * end up with (unless mremap moves it elsewhere before that
3215 * first wfault), so /proc/pid/maps tells a consistent story.
3217 * By setting it to reflect the virtual start address of the
3218 * vma, merges and splits can happen in a seamless way, just
3219 * using the existing file pgoff checks and manipulations.
3220 * Similarly in do_mmap_pgoff and in do_brk.
3222 if (vma_is_anonymous(vma)) {
3223 BUG_ON(vma->anon_vma);
3224 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3227 vma_link(mm, vma, prev, rb_link, rb_parent);
3232 * Copy the vma structure to a new location in the same mm,
3233 * prior to moving page table entries, to effect an mremap move.
3235 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
3236 unsigned long addr, unsigned long len, pgoff_t pgoff,
3237 bool *need_rmap_locks)
3239 struct vm_area_struct *vma = *vmap;
3240 unsigned long vma_start = vma->vm_start;
3241 struct mm_struct *mm = vma->vm_mm;
3242 struct vm_area_struct *new_vma, *prev;
3243 struct rb_node **rb_link, *rb_parent;
3244 bool faulted_in_anon_vma = true;
3247 * If anonymous vma has not yet been faulted, update new pgoff
3248 * to match new location, to increase its chance of merging.
3250 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
3251 pgoff = addr >> PAGE_SHIFT;
3252 faulted_in_anon_vma = false;
3255 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3256 return NULL; /* should never get here */
3257 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
3258 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3259 vma->vm_userfaultfd_ctx);
3262 * Source vma may have been merged into new_vma
3264 if (unlikely(vma_start >= new_vma->vm_start &&
3265 vma_start < new_vma->vm_end)) {
3267 * The only way we can get a vma_merge with
3268 * self during an mremap is if the vma hasn't
3269 * been faulted in yet and we were allowed to
3270 * reset the dst vma->vm_pgoff to the
3271 * destination address of the mremap to allow
3272 * the merge to happen. mremap must change the
3273 * vm_pgoff linearity between src and dst vmas
3274 * (in turn preventing a vma_merge) to be
3275 * safe. It is only safe to keep the vm_pgoff
3276 * linear if there are no pages mapped yet.
3278 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
3279 *vmap = vma = new_vma;
3281 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
3283 new_vma = vm_area_dup(vma);
3286 new_vma->vm_start = addr;
3287 new_vma->vm_end = addr + len;
3288 new_vma->vm_pgoff = pgoff;
3289 if (vma_dup_policy(vma, new_vma))
3291 if (anon_vma_clone(new_vma, vma))
3292 goto out_free_mempol;
3293 if (new_vma->vm_file)
3294 get_file(new_vma->vm_file);
3295 if (new_vma->vm_ops && new_vma->vm_ops->open)
3296 new_vma->vm_ops->open(new_vma);
3297 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3298 *need_rmap_locks = false;
3303 mpol_put(vma_policy(new_vma));
3305 vm_area_free(new_vma);
3311 * Return true if the calling process may expand its vm space by the passed
3314 bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
3316 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3319 if (is_data_mapping(flags) &&
3320 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
3321 /* Workaround for Valgrind */
3322 if (rlimit(RLIMIT_DATA) == 0 &&
3323 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3326 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3327 current->comm, current->pid,
3328 (mm->data_vm + npages) << PAGE_SHIFT,
3329 rlimit(RLIMIT_DATA),
3330 ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data");
3332 if (!ignore_rlimit_data)
3339 void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3341 mm->total_vm += npages;
3343 if (is_exec_mapping(flags))
3344 mm->exec_vm += npages;
3345 else if (is_stack_mapping(flags))
3346 mm->stack_vm += npages;
3347 else if (is_data_mapping(flags))
3348 mm->data_vm += npages;
3351 static vm_fault_t special_mapping_fault(struct vm_fault *vmf);
3354 * Having a close hook prevents vma merging regardless of flags.
3356 static void special_mapping_close(struct vm_area_struct *vma)
3360 static const char *special_mapping_name(struct vm_area_struct *vma)
3362 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3365 static int special_mapping_mremap(struct vm_area_struct *new_vma)
3367 struct vm_special_mapping *sm = new_vma->vm_private_data;
3369 if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3373 return sm->mremap(sm, new_vma);
3378 static const struct vm_operations_struct special_mapping_vmops = {
3379 .close = special_mapping_close,
3380 .fault = special_mapping_fault,
3381 .mremap = special_mapping_mremap,
3382 .name = special_mapping_name,
3385 static const struct vm_operations_struct legacy_special_mapping_vmops = {
3386 .close = special_mapping_close,
3387 .fault = special_mapping_fault,
3390 static vm_fault_t special_mapping_fault(struct vm_fault *vmf)
3392 struct vm_area_struct *vma = vmf->vma;
3394 struct page **pages;
3396 if (vma->vm_ops == &legacy_special_mapping_vmops) {
3397 pages = vma->vm_private_data;
3399 struct vm_special_mapping *sm = vma->vm_private_data;
3402 return sm->fault(sm, vmf->vma, vmf);
3407 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
3411 struct page *page = *pages;
3417 return VM_FAULT_SIGBUS;
3420 static struct vm_area_struct *__install_special_mapping(
3421 struct mm_struct *mm,
3422 unsigned long addr, unsigned long len,
3423 unsigned long vm_flags, void *priv,
3424 const struct vm_operations_struct *ops)
3427 struct vm_area_struct *vma;
3429 vma = vm_area_alloc(mm);
3430 if (unlikely(vma == NULL))
3431 return ERR_PTR(-ENOMEM);
3433 vma->vm_start = addr;
3434 vma->vm_end = addr + len;
3436 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
3437 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3440 vma->vm_private_data = priv;
3442 ret = insert_vm_struct(mm, vma);
3446 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
3448 perf_event_mmap(vma);
3454 return ERR_PTR(ret);
3457 bool vma_is_special_mapping(const struct vm_area_struct *vma,
3458 const struct vm_special_mapping *sm)
3460 return vma->vm_private_data == sm &&
3461 (vma->vm_ops == &special_mapping_vmops ||
3462 vma->vm_ops == &legacy_special_mapping_vmops);
3466 * Called with mm->mmap_sem held for writing.
3467 * Insert a new vma covering the given region, with the given flags.
3468 * Its pages are supplied by the given array of struct page *.
3469 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3470 * The region past the last page supplied will always produce SIGBUS.
3471 * The array pointer and the pages it points to are assumed to stay alive
3472 * for as long as this mapping might exist.
3474 struct vm_area_struct *_install_special_mapping(
3475 struct mm_struct *mm,
3476 unsigned long addr, unsigned long len,
3477 unsigned long vm_flags, const struct vm_special_mapping *spec)
3479 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3480 &special_mapping_vmops);
3483 int install_special_mapping(struct mm_struct *mm,
3484 unsigned long addr, unsigned long len,
3485 unsigned long vm_flags, struct page **pages)
3487 struct vm_area_struct *vma = __install_special_mapping(
3488 mm, addr, len, vm_flags, (void *)pages,
3489 &legacy_special_mapping_vmops);
3491 return PTR_ERR_OR_ZERO(vma);
3494 static DEFINE_MUTEX(mm_all_locks_mutex);
3496 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3498 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3500 * The LSB of head.next can't change from under us
3501 * because we hold the mm_all_locks_mutex.
3503 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
3505 * We can safely modify head.next after taking the
3506 * anon_vma->root->rwsem. If some other vma in this mm shares
3507 * the same anon_vma we won't take it again.
3509 * No need of atomic instructions here, head.next
3510 * can't change from under us thanks to the
3511 * anon_vma->root->rwsem.
3513 if (__test_and_set_bit(0, (unsigned long *)
3514 &anon_vma->root->rb_root.rb_root.rb_node))
3519 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3521 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3523 * AS_MM_ALL_LOCKS can't change from under us because
3524 * we hold the mm_all_locks_mutex.
3526 * Operations on ->flags have to be atomic because
3527 * even if AS_MM_ALL_LOCKS is stable thanks to the
3528 * mm_all_locks_mutex, there may be other cpus
3529 * changing other bitflags in parallel to us.
3531 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3533 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
3538 * This operation locks against the VM for all pte/vma/mm related
3539 * operations that could ever happen on a certain mm. This includes
3540 * vmtruncate, try_to_unmap, and all page faults.
3542 * The caller must take the mmap_sem in write mode before calling
3543 * mm_take_all_locks(). The caller isn't allowed to release the
3544 * mmap_sem until mm_drop_all_locks() returns.
3546 * mmap_sem in write mode is required in order to block all operations
3547 * that could modify pagetables and free pages without need of
3548 * altering the vma layout. It's also needed in write mode to avoid new
3549 * anon_vmas to be associated with existing vmas.
3551 * A single task can't take more than one mm_take_all_locks() in a row
3552 * or it would deadlock.
3554 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3555 * mapping->flags avoid to take the same lock twice, if more than one
3556 * vma in this mm is backed by the same anon_vma or address_space.
3558 * We take locks in following order, accordingly to comment at beginning
3560 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3562 * - all i_mmap_rwsem locks;
3563 * - all anon_vma->rwseml
3565 * We can take all locks within these types randomly because the VM code
3566 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3567 * mm_all_locks_mutex.
3569 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3570 * that may have to take thousand of locks.
3572 * mm_take_all_locks() can fail if it's interrupted by signals.
3574 int mm_take_all_locks(struct mm_struct *mm)
3576 struct vm_area_struct *vma;
3577 struct anon_vma_chain *avc;
3579 BUG_ON(down_read_trylock(&mm->mmap_sem));
3581 mutex_lock(&mm_all_locks_mutex);
3583 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3584 if (signal_pending(current))
3586 if (vma->vm_file && vma->vm_file->f_mapping &&
3587 is_vm_hugetlb_page(vma))
3588 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3591 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3592 if (signal_pending(current))
3594 if (vma->vm_file && vma->vm_file->f_mapping &&
3595 !is_vm_hugetlb_page(vma))
3596 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3599 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3600 if (signal_pending(current))
3603 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3604 vm_lock_anon_vma(mm, avc->anon_vma);
3610 mm_drop_all_locks(mm);
3614 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3616 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3618 * The LSB of head.next can't change to 0 from under
3619 * us because we hold the mm_all_locks_mutex.
3621 * We must however clear the bitflag before unlocking
3622 * the vma so the users using the anon_vma->rb_root will
3623 * never see our bitflag.
3625 * No need of atomic instructions here, head.next
3626 * can't change from under us until we release the
3627 * anon_vma->root->rwsem.
3629 if (!__test_and_clear_bit(0, (unsigned long *)
3630 &anon_vma->root->rb_root.rb_root.rb_node))
3632 anon_vma_unlock_write(anon_vma);
3636 static void vm_unlock_mapping(struct address_space *mapping)
3638 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3640 * AS_MM_ALL_LOCKS can't change to 0 from under us
3641 * because we hold the mm_all_locks_mutex.
3643 i_mmap_unlock_write(mapping);
3644 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3651 * The mmap_sem cannot be released by the caller until
3652 * mm_drop_all_locks() returns.
3654 void mm_drop_all_locks(struct mm_struct *mm)
3656 struct vm_area_struct *vma;
3657 struct anon_vma_chain *avc;
3659 BUG_ON(down_read_trylock(&mm->mmap_sem));
3660 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3662 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3664 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3665 vm_unlock_anon_vma(avc->anon_vma);
3666 if (vma->vm_file && vma->vm_file->f_mapping)
3667 vm_unlock_mapping(vma->vm_file->f_mapping);
3670 mutex_unlock(&mm_all_locks_mutex);
3674 * initialise the percpu counter for VM
3676 void __init mmap_init(void)
3680 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3685 * Initialise sysctl_user_reserve_kbytes.
3687 * This is intended to prevent a user from starting a single memory hogging
3688 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3691 * The default value is min(3% of free memory, 128MB)
3692 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3694 static int init_user_reserve(void)
3696 unsigned long free_kbytes;
3698 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3700 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3703 subsys_initcall(init_user_reserve);
3706 * Initialise sysctl_admin_reserve_kbytes.
3708 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3709 * to log in and kill a memory hogging process.
3711 * Systems with more than 256MB will reserve 8MB, enough to recover
3712 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3713 * only reserve 3% of free pages by default.
3715 static int init_admin_reserve(void)
3717 unsigned long free_kbytes;
3719 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3721 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3724 subsys_initcall(init_admin_reserve);
3727 * Reinititalise user and admin reserves if memory is added or removed.
3729 * The default user reserve max is 128MB, and the default max for the
3730 * admin reserve is 8MB. These are usually, but not always, enough to
3731 * enable recovery from a memory hogging process using login/sshd, a shell,
3732 * and tools like top. It may make sense to increase or even disable the
3733 * reserve depending on the existence of swap or variations in the recovery
3734 * tools. So, the admin may have changed them.
3736 * If memory is added and the reserves have been eliminated or increased above
3737 * the default max, then we'll trust the admin.
3739 * If memory is removed and there isn't enough free memory, then we
3740 * need to reset the reserves.
3742 * Otherwise keep the reserve set by the admin.
3744 static int reserve_mem_notifier(struct notifier_block *nb,
3745 unsigned long action, void *data)
3747 unsigned long tmp, free_kbytes;
3751 /* Default max is 128MB. Leave alone if modified by operator. */
3752 tmp = sysctl_user_reserve_kbytes;
3753 if (0 < tmp && tmp < (1UL << 17))
3754 init_user_reserve();
3756 /* Default max is 8MB. Leave alone if modified by operator. */
3757 tmp = sysctl_admin_reserve_kbytes;
3758 if (0 < tmp && tmp < (1UL << 13))
3759 init_admin_reserve();
3763 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3765 if (sysctl_user_reserve_kbytes > free_kbytes) {
3766 init_user_reserve();
3767 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3768 sysctl_user_reserve_kbytes);
3771 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3772 init_admin_reserve();
3773 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3774 sysctl_admin_reserve_kbytes);
3783 static struct notifier_block reserve_mem_nb = {
3784 .notifier_call = reserve_mem_notifier,
3787 static int __meminit init_reserve_notifier(void)
3789 if (register_hotmemory_notifier(&reserve_mem_nb))
3790 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3794 subsys_initcall(init_reserve_notifier);