6 * Address space accounting code <alan@redhat.com>
9 #include <linux/slab.h>
10 #include <linux/shm.h>
11 #include <linux/mman.h>
12 #include <linux/pagemap.h>
13 #include <linux/swap.h>
14 #include <linux/syscalls.h>
15 #include <linux/init.h>
16 #include <linux/file.h>
18 #include <linux/personality.h>
19 #include <linux/security.h>
20 #include <linux/hugetlb.h>
21 #include <linux/profile.h>
22 #include <linux/module.h>
23 #include <linux/mount.h>
24 #include <linux/mempolicy.h>
25 #include <linux/rmap.h>
26 #include <linux/random.h>
28 #include <asm/uaccess.h>
29 #include <asm/cacheflush.h>
33 * WARNING: the debugging will use recursive algorithms so never enable this
34 * unless you know what you are doing.
38 /* description of effects of mapping type and prot in current implementation.
39 * this is due to the limited x86 page protection hardware. The expected
40 * behavior is in parens:
43 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
44 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
45 * w: (no) no w: (no) no w: (yes) yes w: (no) no
46 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
48 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
49 * w: (no) no w: (no) no w: (copy) copy w: (no) no
50 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
53 pgprot_t protection_map[16] = {
54 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
55 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
58 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
59 int sysctl_overcommit_ratio = 50; /* default is 50% */
60 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
61 atomic_t vm_committed_space = ATOMIC_INIT(0);
63 EXPORT_SYMBOL(sysctl_overcommit_memory);
64 EXPORT_SYMBOL(sysctl_overcommit_ratio);
65 EXPORT_SYMBOL(sysctl_max_map_count);
66 EXPORT_SYMBOL(vm_committed_space);
69 * Requires inode->i_mapping->i_mmap_lock
71 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
72 struct file *file, struct address_space *mapping)
74 if (vma->vm_flags & VM_DENYWRITE)
75 atomic_inc(&file->f_dentry->d_inode->i_writecount);
76 if (vma->vm_flags & VM_SHARED)
77 mapping->i_mmap_writable--;
79 flush_dcache_mmap_lock(mapping);
80 if (unlikely(vma->vm_flags & VM_NONLINEAR))
81 list_del_init(&vma->shared.vm_set.list);
83 vma_prio_tree_remove(vma, &mapping->i_mmap);
84 flush_dcache_mmap_unlock(mapping);
88 * Remove one vm structure and free it.
90 static void remove_vm_struct(struct vm_area_struct *vma)
92 struct file *file = vma->vm_file;
96 struct address_space *mapping = file->f_mapping;
97 spin_lock(&mapping->i_mmap_lock);
98 __remove_shared_vm_struct(vma, file, mapping);
99 spin_unlock(&mapping->i_mmap_lock);
101 if (vma->vm_ops && vma->vm_ops->close)
102 vma->vm_ops->close(vma);
105 anon_vma_unlink(vma);
106 mpol_free(vma_policy(vma));
107 kmem_cache_free(vm_area_cachep, vma);
111 * sys_brk() for the most part doesn't need the global kernel
112 * lock, except when an application is doing something nasty
113 * like trying to un-brk an area that has already been mapped
114 * to a regular file. in this case, the unmapping will need
115 * to invoke file system routines that need the global lock.
117 asmlinkage unsigned long sys_brk(unsigned long brk)
119 unsigned long rlim, retval;
120 unsigned long newbrk, oldbrk;
121 struct mm_struct *mm = current->mm;
123 down_write(&mm->mmap_sem);
125 if (brk < mm->end_code)
127 newbrk = PAGE_ALIGN(brk);
128 oldbrk = PAGE_ALIGN(mm->brk);
129 if (oldbrk == newbrk)
132 /* Always allow shrinking brk. */
133 if (brk <= mm->brk) {
134 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
139 /* Check against rlimit.. */
140 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
141 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
144 /* Check against existing mmap mappings. */
145 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
148 /* Ok, looks good - let it rip. */
149 if (__do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
155 up_write(&mm->mmap_sem);
160 static int browse_rb(struct rb_root *root)
163 struct rb_node *nd, *pn = NULL;
164 unsigned long prev = 0, pend = 0;
166 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
167 struct vm_area_struct *vma;
168 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
169 if (vma->vm_start < prev)
170 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
171 if (vma->vm_start < pend)
172 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
173 if (vma->vm_start > vma->vm_end)
174 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
179 for (nd = pn; nd; nd = rb_prev(nd)) {
183 printk("backwards %d, forwards %d\n", j, i), i = 0;
187 void validate_mm(struct mm_struct *mm)
191 struct vm_area_struct *tmp = mm->mmap;
196 if (i != mm->map_count)
197 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
198 i = browse_rb(&mm->mm_rb);
199 if (i != mm->map_count)
200 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
205 #define validate_mm(mm) do { } while (0)
208 static struct vm_area_struct *
209 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
210 struct vm_area_struct **pprev, struct rb_node ***rb_link,
211 struct rb_node ** rb_parent)
213 struct vm_area_struct * vma;
214 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
216 __rb_link = &mm->mm_rb.rb_node;
217 rb_prev = __rb_parent = NULL;
221 struct vm_area_struct *vma_tmp;
223 __rb_parent = *__rb_link;
224 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
226 if (vma_tmp->vm_end > addr) {
228 if (vma_tmp->vm_start <= addr)
230 __rb_link = &__rb_parent->rb_left;
232 rb_prev = __rb_parent;
233 __rb_link = &__rb_parent->rb_right;
239 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
240 *rb_link = __rb_link;
241 *rb_parent = __rb_parent;
246 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
247 struct vm_area_struct *prev, struct rb_node *rb_parent)
249 if (vma->vm_flags & VM_EXEC)
250 arch_add_exec_range(mm, vma->vm_end);
252 vma->vm_next = prev->vm_next;
257 vma->vm_next = rb_entry(rb_parent,
258 struct vm_area_struct, vm_rb);
264 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
265 struct rb_node **rb_link, struct rb_node *rb_parent)
267 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
268 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
271 static inline void __vma_link_file(struct vm_area_struct *vma)
277 struct address_space *mapping = file->f_mapping;
279 if (vma->vm_flags & VM_DENYWRITE)
280 atomic_dec(&file->f_dentry->d_inode->i_writecount);
281 if (vma->vm_flags & VM_SHARED)
282 mapping->i_mmap_writable++;
284 flush_dcache_mmap_lock(mapping);
285 if (unlikely(vma->vm_flags & VM_NONLINEAR))
286 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
288 vma_prio_tree_insert(vma, &mapping->i_mmap);
289 flush_dcache_mmap_unlock(mapping);
294 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
295 struct vm_area_struct *prev, struct rb_node **rb_link,
296 struct rb_node *rb_parent)
298 __vma_link_list(mm, vma, prev, rb_parent);
299 __vma_link_rb(mm, vma, rb_link, rb_parent);
300 __anon_vma_link(vma);
303 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
304 struct vm_area_struct *prev, struct rb_node **rb_link,
305 struct rb_node *rb_parent)
307 struct address_space *mapping = NULL;
310 mapping = vma->vm_file->f_mapping;
313 spin_lock(&mapping->i_mmap_lock);
316 __vma_link(mm, vma, prev, rb_link, rb_parent);
317 __vma_link_file(vma);
319 anon_vma_unlock(vma);
321 spin_unlock(&mapping->i_mmap_lock);
328 * Helper for vma_adjust in the split_vma insert case:
329 * insert vm structure into list and rbtree and anon_vma,
330 * but it has already been inserted into prio_tree earlier.
333 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
335 struct vm_area_struct * __vma, * prev;
336 struct rb_node ** rb_link, * rb_parent;
338 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
339 if (__vma && __vma->vm_start < vma->vm_end)
341 __vma_link(mm, vma, prev, rb_link, rb_parent);
346 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
347 struct vm_area_struct *prev)
349 prev->vm_next = vma->vm_next;
350 rb_erase(&vma->vm_rb, &mm->mm_rb);
351 if (mm->mmap_cache == vma)
352 mm->mmap_cache = prev;
353 if (vma->vm_flags & VM_EXEC)
354 arch_remove_exec_range(mm, vma->vm_end);
358 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
359 * is already present in an i_mmap tree without adjusting the tree.
360 * The following helper function should be used when such adjustments
361 * are necessary. The "insert" vma (if any) is to be inserted
362 * before we drop the necessary locks.
364 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
365 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
367 struct mm_struct *mm = vma->vm_mm;
368 struct vm_area_struct *next = vma->vm_next;
369 struct vm_area_struct *importer = NULL;
370 struct address_space *mapping = NULL;
371 struct prio_tree_root *root = NULL;
372 struct file *file = vma->vm_file;
373 struct anon_vma *anon_vma = NULL;
374 long adjust_next = 0;
377 if (next && !insert) {
378 if (end >= next->vm_end) {
380 * vma expands, overlapping all the next, and
381 * perhaps the one after too (mprotect case 6).
383 again: remove_next = 1 + (end > next->vm_end);
385 anon_vma = next->anon_vma;
386 } else if (end > next->vm_start) {
388 * vma expands, overlapping part of the next:
389 * mprotect case 5 shifting the boundary up.
391 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
392 anon_vma = next->anon_vma;
394 } else if (end < vma->vm_end) {
396 * vma shrinks, and !insert tells it's not
397 * split_vma inserting another: so it must be
398 * mprotect case 4 shifting the boundary down.
400 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
401 anon_vma = next->anon_vma;
407 mapping = file->f_mapping;
408 if (!(vma->vm_flags & VM_NONLINEAR))
409 root = &mapping->i_mmap;
410 spin_lock(&mapping->i_mmap_lock);
413 * Put into prio_tree now, so instantiated pages
414 * are visible to arm/parisc __flush_dcache_page
415 * throughout; but we cannot insert into address
416 * space until vma start or end is updated.
418 __vma_link_file(insert);
423 * When changing only vma->vm_end, we don't really need
424 * anon_vma lock: but is that case worth optimizing out?
427 anon_vma = vma->anon_vma;
429 spin_lock(&anon_vma->lock);
431 * Easily overlooked: when mprotect shifts the boundary,
432 * make sure the expanding vma has anon_vma set if the
433 * shrinking vma had, to cover any anon pages imported.
435 if (importer && !importer->anon_vma) {
436 importer->anon_vma = anon_vma;
437 __anon_vma_link(importer);
442 flush_dcache_mmap_lock(mapping);
443 vma_prio_tree_remove(vma, root);
445 vma_prio_tree_remove(next, root);
448 vma->vm_start = start;
450 vma->vm_pgoff = pgoff;
452 next->vm_start += adjust_next << PAGE_SHIFT;
453 next->vm_pgoff += adjust_next;
458 vma_prio_tree_insert(next, root);
459 vma_prio_tree_insert(vma, root);
460 flush_dcache_mmap_unlock(mapping);
465 * vma_merge has merged next into vma, and needs
466 * us to remove next before dropping the locks.
468 __vma_unlink(mm, next, vma);
470 __remove_shared_vm_struct(next, file, mapping);
472 __anon_vma_merge(vma, next);
475 * split_vma has split insert from vma, and needs
476 * us to insert it before dropping the locks
477 * (it may either follow vma or precede it).
479 __insert_vm_struct(mm, insert);
483 spin_unlock(&anon_vma->lock);
485 spin_unlock(&mapping->i_mmap_lock);
491 mpol_free(vma_policy(next));
492 kmem_cache_free(vm_area_cachep, next);
494 * In mprotect's case 6 (see comments on vma_merge),
495 * we must remove another next too. It would clutter
496 * up the code too much to do both in one go.
498 if (remove_next == 2) {
508 * If the vma has a ->close operation then the driver probably needs to release
509 * per-vma resources, so we don't attempt to merge those.
511 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
513 static inline int is_mergeable_vma(struct vm_area_struct *vma,
514 struct file *file, unsigned long vm_flags)
516 if (vma->vm_flags != vm_flags)
518 if (vma->vm_file != file)
520 if (vma->vm_ops && vma->vm_ops->close)
525 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
526 struct anon_vma *anon_vma2)
528 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
532 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
533 * in front of (at a lower virtual address and file offset than) the vma.
535 * We cannot merge two vmas if they have differently assigned (non-NULL)
536 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
538 * We don't check here for the merged mmap wrapping around the end of pagecache
539 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
540 * wrap, nor mmaps which cover the final page at index -1UL.
543 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
544 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
546 if (is_mergeable_vma(vma, file, vm_flags) &&
547 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
548 if (vma->vm_pgoff == vm_pgoff)
555 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
556 * beyond (at a higher virtual address and file offset than) the vma.
558 * We cannot merge two vmas if they have differently assigned (non-NULL)
559 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
562 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
563 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
565 if (is_mergeable_vma(vma, file, vm_flags) &&
566 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
568 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
569 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
576 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
577 * whether that can be merged with its predecessor or its successor.
578 * Or both (it neatly fills a hole).
580 * In most cases - when called for mmap, brk or mremap - [addr,end) is
581 * certain not to be mapped by the time vma_merge is called; but when
582 * called for mprotect, it is certain to be already mapped (either at
583 * an offset within prev, or at the start of next), and the flags of
584 * this area are about to be changed to vm_flags - and the no-change
585 * case has already been eliminated.
587 * The following mprotect cases have to be considered, where AAAA is
588 * the area passed down from mprotect_fixup, never extending beyond one
589 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
591 * AAAA AAAA AAAA AAAA
592 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
593 * cannot merge might become might become might become
594 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
595 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
596 * mremap move: PPPPNNNNNNNN 8
598 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
599 * might become case 1 below case 2 below case 3 below
601 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
602 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
604 struct vm_area_struct *vma_merge(struct mm_struct *mm,
605 struct vm_area_struct *prev, unsigned long addr,
606 unsigned long end, unsigned long vm_flags,
607 struct anon_vma *anon_vma, struct file *file,
608 pgoff_t pgoff, struct mempolicy *policy)
610 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
611 struct vm_area_struct *area, *next;
614 * We later require that vma->vm_flags == vm_flags,
615 * so this tests vma->vm_flags & VM_SPECIAL, too.
617 if (vm_flags & VM_SPECIAL)
621 next = prev->vm_next;
625 if (next && next->vm_end == end) /* cases 6, 7, 8 */
626 next = next->vm_next;
629 * Can it merge with the predecessor?
631 if (prev && prev->vm_end == addr &&
632 mpol_equal(vma_policy(prev), policy) &&
633 can_vma_merge_after(prev, vm_flags,
634 anon_vma, file, pgoff)) {
636 * OK, it can. Can we now merge in the successor as well?
638 if (next && end == next->vm_start &&
639 mpol_equal(policy, vma_policy(next)) &&
640 can_vma_merge_before(next, vm_flags,
641 anon_vma, file, pgoff+pglen) &&
642 is_mergeable_anon_vma(prev->anon_vma,
645 vma_adjust(prev, prev->vm_start,
646 next->vm_end, prev->vm_pgoff, NULL);
647 } else /* cases 2, 5, 7 */
648 vma_adjust(prev, prev->vm_start,
649 end, prev->vm_pgoff, NULL);
650 if (prev->vm_flags & VM_EXEC)
651 arch_add_exec_range(mm, prev->vm_end);
656 * Can this new request be merged in front of next?
658 if (next && end == next->vm_start &&
659 mpol_equal(policy, vma_policy(next)) &&
660 can_vma_merge_before(next, vm_flags,
661 anon_vma, file, pgoff+pglen)) {
662 if (prev && addr < prev->vm_end) /* case 4 */
663 vma_adjust(prev, prev->vm_start,
664 addr, prev->vm_pgoff, NULL);
665 else /* cases 3, 8 */
666 vma_adjust(area, addr, next->vm_end,
667 next->vm_pgoff - pglen, NULL);
675 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
676 * neighbouring vmas for a suitable anon_vma, before it goes off
677 * to allocate a new anon_vma. It checks because a repetitive
678 * sequence of mprotects and faults may otherwise lead to distinct
679 * anon_vmas being allocated, preventing vma merge in subsequent
682 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
684 struct vm_area_struct *near;
685 unsigned long vm_flags;
692 * Since only mprotect tries to remerge vmas, match flags
693 * which might be mprotected into each other later on.
694 * Neither mlock nor madvise tries to remerge at present,
695 * so leave their flags as obstructing a merge.
697 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
698 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
700 if (near->anon_vma && vma->vm_end == near->vm_start &&
701 mpol_equal(vma_policy(vma), vma_policy(near)) &&
702 can_vma_merge_before(near, vm_flags,
703 NULL, vma->vm_file, vma->vm_pgoff +
704 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
705 return near->anon_vma;
708 * It is potentially slow to have to call find_vma_prev here.
709 * But it's only on the first write fault on the vma, not
710 * every time, and we could devise a way to avoid it later
711 * (e.g. stash info in next's anon_vma_node when assigning
712 * an anon_vma, or when trying vma_merge). Another time.
714 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
719 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
720 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
722 if (near->anon_vma && near->vm_end == vma->vm_start &&
723 mpol_equal(vma_policy(near), vma_policy(vma)) &&
724 can_vma_merge_after(near, vm_flags,
725 NULL, vma->vm_file, vma->vm_pgoff))
726 return near->anon_vma;
729 * There's no absolute need to look only at touching neighbours:
730 * we could search further afield for "compatible" anon_vmas.
731 * But it would probably just be a waste of time searching,
732 * or lead to too many vmas hanging off the same anon_vma.
733 * We're trying to allow mprotect remerging later on,
734 * not trying to minimize memory used for anon_vmas.
739 #ifdef CONFIG_PROC_FS
740 void __vm_stat_account(struct mm_struct *mm, unsigned long flags,
741 struct file *file, long pages)
743 const unsigned long stack_flags
744 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
746 #ifdef CONFIG_HUGETLB
747 if (flags & VM_HUGETLB) {
748 if (!(flags & VM_DONTCOPY))
749 mm->shared_vm += pages;
752 #endif /* CONFIG_HUGETLB */
755 mm->shared_vm += pages;
756 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
757 mm->exec_vm += pages;
758 } else if (flags & stack_flags)
759 mm->stack_vm += pages;
760 if (flags & (VM_RESERVED|VM_IO))
761 mm->reserved_vm += pages;
763 #endif /* CONFIG_PROC_FS */
766 * The caller must hold down_write(current->mm->mmap_sem).
769 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
770 unsigned long len, unsigned long prot,
771 unsigned long flags, unsigned long pgoff)
773 struct mm_struct * mm = current->mm;
774 struct vm_area_struct * vma, * prev;
776 unsigned int vm_flags;
777 int correct_wcount = 0;
779 struct rb_node ** rb_link, * rb_parent;
781 unsigned long charged = 0;
784 if (is_file_hugepages(file))
787 if (!file->f_op || !file->f_op->mmap)
790 if ((prot & PROT_EXEC) &&
791 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
795 * Does the application expect PROT_READ to imply PROT_EXEC?
797 * (the exception is when the underlying filesystem is noexec
798 * mounted, in which case we dont add PROT_EXEC.)
800 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
801 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
807 /* Careful about overflows.. */
808 len = PAGE_ALIGN(len);
809 if (!len || len > TASK_SIZE)
812 /* offset overflow? */
813 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
816 /* Too many mappings? */
817 if (mm->map_count > sysctl_max_map_count)
820 /* Obtain the address to map to. we verify (or select) it and ensure
821 * that it represents a valid section of the address space.
823 addr = get_unmapped_area_prot(file, addr, len, pgoff, flags, prot & PROT_EXEC);
824 if (addr & ~PAGE_MASK)
827 /* Do simple checking here so the lower-level routines won't have
828 * to. we assume access permissions have been handled by the open
829 * of the memory object, so we don't do any here.
831 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
832 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
834 if (flags & MAP_LOCKED) {
837 vm_flags |= VM_LOCKED;
839 /* mlock MCL_FUTURE? */
840 if (vm_flags & VM_LOCKED) {
841 unsigned long locked, lock_limit;
842 locked = mm->locked_vm << PAGE_SHIFT;
843 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
845 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
849 inode = file ? file->f_dentry->d_inode : NULL;
852 switch (flags & MAP_TYPE) {
854 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
858 * Make sure we don't allow writing to an append-only
861 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
865 * Make sure there are no mandatory locks on the file.
867 if (locks_verify_locked(inode))
870 vm_flags |= VM_SHARED | VM_MAYSHARE;
871 if (!(file->f_mode & FMODE_WRITE))
872 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
876 if (!(file->f_mode & FMODE_READ))
884 switch (flags & MAP_TYPE) {
886 vm_flags |= VM_SHARED | VM_MAYSHARE;
890 * Set pgoff according to addr for anon_vma.
892 pgoff = addr >> PAGE_SHIFT;
899 error = security_file_mmap(file, prot, flags);
906 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
907 if (vma && vma->vm_start < addr + len) {
908 if (do_munmap(mm, addr, len))
913 /* Check against address space limit. */
914 if ((mm->total_vm << PAGE_SHIFT) + len
915 > current->signal->rlim[RLIMIT_AS].rlim_cur)
918 /* check context space, maybe only Private writable mapping? */
919 if (!vx_vmpages_avail(mm, len >> PAGE_SHIFT))
922 if (accountable && (!(flags & MAP_NORESERVE) ||
923 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
924 if (vm_flags & VM_SHARED) {
925 /* Check memory availability in shmem_file_setup? */
926 vm_flags |= VM_ACCOUNT;
927 } else if (vm_flags & VM_WRITE) {
929 * Private writable mapping: check memory availability
931 charged = len >> PAGE_SHIFT;
932 if (security_vm_enough_memory(charged))
934 vm_flags |= VM_ACCOUNT;
939 * Can we just expand an old private anonymous mapping?
940 * The VM_SHARED test is necessary because shmem_zero_setup
941 * will create the file object for a shared anonymous map below.
943 if (!file && !(vm_flags & VM_SHARED) &&
944 vma_merge(mm, prev, addr, addr + len, vm_flags,
945 NULL, NULL, pgoff, NULL))
949 * Determine the object being mapped and call the appropriate
950 * specific mapper. the address has already been validated, but
951 * not unmapped, but the maps are removed from the list.
953 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
958 memset(vma, 0, sizeof(*vma));
961 vma->vm_start = addr;
962 vma->vm_end = addr + len;
963 vma->vm_flags = vm_flags;
964 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
965 vma->vm_pgoff = pgoff;
969 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
971 if (vm_flags & VM_DENYWRITE) {
972 error = deny_write_access(file);
979 error = file->f_op->mmap(file, vma);
981 goto unmap_and_free_vma;
982 } else if (vm_flags & VM_SHARED) {
983 error = shmem_zero_setup(vma);
988 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
989 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
990 * that memory reservation must be checked; but that reservation
991 * belongs to shared memory object, not to vma: so now clear it.
993 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
994 vma->vm_flags &= ~VM_ACCOUNT;
996 /* Can addr have changed??
998 * Answer: Yes, several device drivers can do it in their
999 * f_op->mmap method. -DaveM
1001 addr = vma->vm_start;
1002 pgoff = vma->vm_pgoff;
1003 vm_flags = vma->vm_flags;
1005 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1006 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1007 file = vma->vm_file;
1008 vma_link(mm, vma, prev, rb_link, rb_parent);
1010 atomic_inc(&inode->i_writecount);
1014 atomic_inc(&inode->i_writecount);
1017 mpol_free(vma_policy(vma));
1018 kmem_cache_free(vm_area_cachep, vma);
1021 // mm->total_vm += len >> PAGE_SHIFT;
1022 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1023 __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1024 if (vm_flags & VM_LOCKED) {
1025 // mm->locked_vm += len >> PAGE_SHIFT;
1026 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1027 make_pages_present(addr, addr + len);
1029 if (flags & MAP_POPULATE) {
1030 up_write(&mm->mmap_sem);
1031 sys_remap_file_pages(addr, len, 0,
1032 pgoff, flags & MAP_NONBLOCK);
1033 down_write(&mm->mmap_sem);
1035 __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1040 atomic_inc(&inode->i_writecount);
1041 vma->vm_file = NULL;
1044 /* Undo any partial mapping done by a device driver. */
1045 zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1047 kmem_cache_free(vm_area_cachep, vma);
1050 vm_unacct_memory(charged);
1054 EXPORT_SYMBOL(do_mmap_pgoff);
1056 /* Get an address range which is currently unmapped.
1057 * For shmat() with addr=0.
1059 * Ugly calling convention alert:
1060 * Return value with the low bits set means error value,
1062 * if (ret & ~PAGE_MASK)
1065 * This function "knows" that -ENOMEM has the bits set.
1067 #ifndef HAVE_ARCH_UNMAPPED_AREA
1069 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1070 unsigned long len, unsigned long pgoff, unsigned long flags)
1072 struct mm_struct *mm = current->mm;
1073 struct vm_area_struct *vma;
1074 unsigned long start_addr;
1076 if (len > TASK_SIZE)
1080 addr = PAGE_ALIGN(addr);
1081 vma = find_vma(mm, addr);
1082 if (TASK_SIZE - len >= addr &&
1083 (!vma || addr + len <= vma->vm_start))
1086 start_addr = addr = mm->free_area_cache;
1089 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1090 /* At this point: (!vma || addr < vma->vm_end). */
1091 if (TASK_SIZE - len < addr) {
1093 * Start a new search - just in case we missed
1096 if (start_addr != TASK_UNMAPPED_BASE) {
1097 start_addr = addr = TASK_UNMAPPED_BASE;
1102 if (!vma || addr + len <= vma->vm_start) {
1104 * Remember the place where we stopped the search:
1106 mm->free_area_cache = addr + len;
1114 void arch_unmap_area(struct vm_area_struct *area)
1117 * Is this a new hole at the lowest possible address?
1119 if (area->vm_start >= TASK_UNMAPPED_BASE &&
1120 area->vm_start < area->vm_mm->free_area_cache)
1121 area->vm_mm->free_area_cache = area->vm_start;
1125 * This mmap-allocator allocates new areas top-down from below the
1126 * stack's low limit (the base):
1128 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1130 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1131 const unsigned long len, const unsigned long pgoff,
1132 const unsigned long flags)
1134 struct vm_area_struct *vma, *prev_vma;
1135 struct mm_struct *mm = current->mm;
1136 unsigned long base = mm->mmap_base, addr = addr0;
1139 /* requested length too big for entire address space */
1140 if (len > TASK_SIZE)
1143 /* dont allow allocations above current base */
1144 if (mm->free_area_cache > base)
1145 mm->free_area_cache = base;
1147 /* requesting a specific address */
1149 addr = PAGE_ALIGN(addr);
1150 vma = find_vma(mm, addr);
1151 if (TASK_SIZE - len >= addr &&
1152 (!vma || addr + len <= vma->vm_start))
1157 /* make sure it can fit in the remaining address space */
1158 if (mm->free_area_cache < len)
1161 /* either no address requested or cant fit in requested address hole */
1162 addr = (mm->free_area_cache - len) & PAGE_MASK;
1165 * Lookup failure means no vma is above this address,
1166 * i.e. return with success:
1168 if (!(vma = find_vma_prev(mm, addr, &prev_vma)))
1172 * new region fits between prev_vma->vm_end and
1173 * vma->vm_start, use it:
1175 if (addr+len <= vma->vm_start &&
1176 (!prev_vma || (addr >= prev_vma->vm_end)))
1177 /* remember the address as a hint for next time */
1178 return (mm->free_area_cache = addr);
1180 /* pull free_area_cache down to the first hole */
1181 if (mm->free_area_cache == vma->vm_end)
1182 mm->free_area_cache = vma->vm_start;
1184 /* try just below the current vma->vm_start */
1185 addr = vma->vm_start-len;
1186 } while (len <= vma->vm_start);
1190 * if hint left us with no space for the requested
1191 * mapping then try again:
1194 mm->free_area_cache = base;
1199 * A failed mmap() very likely causes application failure,
1200 * so fall back to the bottom-up function here. This scenario
1201 * can happen with large stack limits and large mmap()
1204 mm->free_area_cache = TASK_UNMAPPED_BASE;
1205 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1207 * Restore the topdown base:
1209 mm->free_area_cache = base;
1215 void arch_unmap_area_topdown(struct vm_area_struct *area)
1218 * Is this a new hole at the highest possible address?
1220 if (area->vm_end > area->vm_mm->free_area_cache)
1221 area->vm_mm->free_area_cache = area->vm_end;
1226 get_unmapped_area_prot(struct file *file, unsigned long addr, unsigned long len,
1227 unsigned long pgoff, unsigned long flags, int exec)
1229 if (flags & MAP_FIXED) {
1232 if (addr > TASK_SIZE - len)
1234 if (addr & ~PAGE_MASK)
1236 if (file && is_file_hugepages(file)) {
1238 * Check if the given range is hugepage aligned, and
1239 * can be made suitable for hugepages.
1241 ret = prepare_hugepage_range(addr, len);
1244 * Ensure that a normal request is not falling in a
1245 * reserved hugepage range. For some archs like IA-64,
1246 * there is a separate region for hugepages.
1248 ret = is_hugepage_only_range(addr, len);
1255 if (file && file->f_op && file->f_op->get_unmapped_area)
1256 return file->f_op->get_unmapped_area(file, addr, len,
1259 if (exec && current->mm->get_unmapped_exec_area)
1260 return current->mm->get_unmapped_exec_area(file, addr, len, pgoff, flags);
1262 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
1265 EXPORT_SYMBOL(get_unmapped_area_prot);
1268 #define SHLIB_BASE 0x00111000
1270 unsigned long arch_get_unmapped_exec_area(struct file *filp, unsigned long addr0,
1271 unsigned long len0, unsigned long pgoff, unsigned long flags)
1273 unsigned long addr = addr0, len = len0;
1274 struct mm_struct *mm = current->mm;
1275 struct vm_area_struct *vma;
1278 if (len > TASK_SIZE)
1281 if (!addr && !(flags & MAP_FIXED))
1282 addr = randomize_range(SHLIB_BASE, 0x01000000, len);
1285 addr = PAGE_ALIGN(addr);
1286 vma = find_vma(mm, addr);
1287 if (TASK_SIZE - len >= addr &&
1288 (!vma || addr + len <= vma->vm_start)) {
1295 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1296 /* At this point: (!vma || addr < vma->vm_end). */
1297 if (TASK_SIZE - len < addr) {
1300 if (!vma || addr + len <= vma->vm_start) {
1302 * Must not let a PROT_EXEC mapping get into the
1305 if (addr + len > mm->brk)
1309 * Up until the brk area we randomize addresses
1310 * as much as possible:
1312 if (addr >= 0x01000000) {
1313 tmp = randomize_range(0x01000000, mm->brk, len);
1314 vma = find_vma(mm, tmp);
1315 if (TASK_SIZE - len >= tmp &&
1316 (!vma || tmp + len <= vma->vm_start))
1320 * Ok, randomization didnt work out - return
1321 * the result of the linear search:
1329 return current->mm->get_unmapped_area(filp, addr0, len0, pgoff, flags);
1334 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1335 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1337 struct vm_area_struct *vma = NULL;
1340 /* Check the cache first. */
1341 /* (Cache hit rate is typically around 35%.) */
1342 vma = mm->mmap_cache;
1343 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1344 struct rb_node * rb_node;
1346 rb_node = mm->mm_rb.rb_node;
1350 struct vm_area_struct * vma_tmp;
1352 vma_tmp = rb_entry(rb_node,
1353 struct vm_area_struct, vm_rb);
1355 if (vma_tmp->vm_end > addr) {
1357 if (vma_tmp->vm_start <= addr)
1359 rb_node = rb_node->rb_left;
1361 rb_node = rb_node->rb_right;
1364 mm->mmap_cache = vma;
1370 EXPORT_SYMBOL(find_vma);
1372 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1373 struct vm_area_struct *
1374 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1375 struct vm_area_struct **pprev)
1377 struct vm_area_struct *vma = NULL, *prev = NULL;
1378 struct rb_node * rb_node;
1382 /* Guard against addr being lower than the first VMA */
1385 /* Go through the RB tree quickly. */
1386 rb_node = mm->mm_rb.rb_node;
1389 struct vm_area_struct *vma_tmp;
1390 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1392 if (addr < vma_tmp->vm_end) {
1393 rb_node = rb_node->rb_left;
1396 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1398 rb_node = rb_node->rb_right;
1404 return prev ? prev->vm_next : vma;
1408 static int over_stack_limit(unsigned long sz)
1410 if (sz < EXEC_STACK_BIAS)
1412 return (sz - EXEC_STACK_BIAS) > current->signal->rlim[RLIMIT_STACK].rlim_cur;
1415 #ifdef CONFIG_STACK_GROWSUP
1417 * vma is the first one with address > vma->vm_end. Have to extend vma.
1419 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1423 if (!(vma->vm_flags & VM_GROWSUP))
1427 * We must make sure the anon_vma is allocated
1428 * so that the anon_vma locking is not a noop.
1430 if (unlikely(anon_vma_prepare(vma)))
1435 * vma->vm_start/vm_end cannot change under us because the caller
1436 * is required to hold the mmap_sem in read mode. We need the
1437 * anon_vma lock to serialize against concurrent expand_stacks.
1439 address += 4 + PAGE_SIZE - 1;
1440 address &= PAGE_MASK;
1441 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1443 if (address < vma->vm_end) {
1444 anon_vma_unlock(vma);
1447 /* Overcommit.. vx check first to avoid vm_unacct_memory() */
1448 if (!vx_vmpages_avail(vma->vm_mm, grow) ||
1449 security_vm_enough_memory(grow)) {
1450 anon_vma_unlock(vma);
1454 if (over_stack_limit(address - vma->vm_start) ||
1455 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1456 current->signal->rlim[RLIMIT_AS].rlim_cur) {
1457 anon_vma_unlock(vma);
1458 vm_unacct_memory(grow);
1461 if ((vma->vm_flags & VM_LOCKED) && !capable(CAP_IPC_LOCK) &&
1462 ((vma->vm_mm->locked_vm + grow) << PAGE_SHIFT) >
1463 current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur) {
1464 anon_vma_unlock(vma);
1465 vm_unacct_memory(grow);
1468 vma->vm_end = address;
1469 // vma->vm_mm->total_vm += grow;
1470 vx_vmpages_add(vma->vm_mm, grow);
1471 if (vma->vm_flags & VM_LOCKED)
1472 // vma->vm_mm->locked_vm += grow;
1473 vx_vmlocked_add(vma->vm_mm, grow);
1474 __vm_stat_account(vma->vm_mm, vma->vm_flags, vma->vm_file, grow);
1475 anon_vma_unlock(vma);
1479 struct vm_area_struct *
1480 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1482 struct vm_area_struct *vma, *prev;
1485 vma = find_vma_prev(mm, addr, &prev);
1486 if (vma && (vma->vm_start <= addr))
1488 if (!prev || expand_stack(prev, addr))
1490 if (prev->vm_flags & VM_LOCKED) {
1491 make_pages_present(addr, prev->vm_end);
1497 * vma is the first one with address < vma->vm_start. Have to extend vma.
1499 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1504 * We must make sure the anon_vma is allocated
1505 * so that the anon_vma locking is not a noop.
1507 if (unlikely(anon_vma_prepare(vma)))
1512 * vma->vm_start/vm_end cannot change under us because the caller
1513 * is required to hold the mmap_sem in read mode. We need the
1514 * anon_vma lock to serialize against concurrent expand_stacks.
1516 address &= PAGE_MASK;
1517 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1519 if (address >= vma->vm_start) {
1520 anon_vma_unlock(vma);
1523 /* Overcommit.. vx check first to avoid vm_unacct_memory() */
1524 if (!vx_vmpages_avail(vma->vm_mm, grow) ||
1525 security_vm_enough_memory(grow)) {
1526 anon_vma_unlock(vma);
1530 if (over_stack_limit(vma->vm_end - address) ||
1531 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1532 current->signal->rlim[RLIMIT_AS].rlim_cur) {
1533 anon_vma_unlock(vma);
1534 vm_unacct_memory(grow);
1537 if ((vma->vm_flags & VM_LOCKED) && !capable(CAP_IPC_LOCK) &&
1538 ((vma->vm_mm->locked_vm + grow) << PAGE_SHIFT) >
1539 current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur) {
1540 anon_vma_unlock(vma);
1541 vm_unacct_memory(grow);
1544 vma->vm_start = address;
1545 vma->vm_pgoff -= grow;
1546 // vma->vm_mm->total_vm += grow;
1547 vx_vmpages_add(vma->vm_mm, grow);
1548 if (vma->vm_flags & VM_LOCKED)
1549 // vma->vm_mm->locked_vm += grow;
1550 vx_vmlocked_add(vma->vm_mm, grow);
1551 __vm_stat_account(vma->vm_mm, vma->vm_flags, vma->vm_file, grow);
1552 anon_vma_unlock(vma);
1556 struct vm_area_struct *
1557 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1559 struct vm_area_struct * vma;
1560 unsigned long start;
1563 vma = find_vma(mm,addr);
1566 if (vma->vm_start <= addr)
1568 if (!(vma->vm_flags & VM_GROWSDOWN))
1570 start = vma->vm_start;
1571 if (expand_stack(vma, addr))
1573 if (vma->vm_flags & VM_LOCKED) {
1574 make_pages_present(addr, start);
1581 * Try to free as many page directory entries as we can,
1582 * without having to work very hard at actually scanning
1583 * the page tables themselves.
1585 * Right now we try to free page tables if we have a nice
1586 * PGDIR-aligned area that got free'd up. We could be more
1587 * granular if we want to, but this is fast and simple,
1588 * and covers the bad cases.
1590 * "prev", if it exists, points to a vma before the one
1591 * we just free'd - but there's no telling how much before.
1593 static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev,
1594 unsigned long start, unsigned long end)
1596 unsigned long first = start & PGDIR_MASK;
1597 unsigned long last = end + PGDIR_SIZE - 1;
1598 unsigned long start_index, end_index;
1599 struct mm_struct *mm = tlb->mm;
1605 if (prev->vm_end > start) {
1606 if (last > prev->vm_start)
1607 last = prev->vm_start;
1612 struct vm_area_struct *next = prev->vm_next;
1615 if (next->vm_start < start) {
1619 if (last > next->vm_start)
1620 last = next->vm_start;
1622 if (prev->vm_end > first)
1623 first = prev->vm_end + PGDIR_SIZE - 1;
1627 if (last < first) /* for arches with discontiguous pgd indices */
1630 * If the PGD bits are not consecutive in the virtual address, the
1631 * old method of shifting the VA >> by PGDIR_SHIFT doesn't work.
1633 start_index = pgd_index(first);
1634 if (start_index < FIRST_USER_PGD_NR)
1635 start_index = FIRST_USER_PGD_NR;
1636 end_index = pgd_index(last);
1637 if (end_index > start_index) {
1638 clear_page_tables(tlb, start_index, end_index - start_index);
1639 flush_tlb_pgtables(mm, first & PGDIR_MASK, last & PGDIR_MASK);
1643 /* Normal function to fix up a mapping
1644 * This function is the default for when an area has no specific
1645 * function. This may be used as part of a more specific routine.
1647 * By the time this function is called, the area struct has been
1648 * removed from the process mapping list.
1650 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1652 size_t len = area->vm_end - area->vm_start;
1654 // area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1655 vx_vmpages_sub(area->vm_mm, len >> PAGE_SHIFT);
1657 if (area->vm_flags & VM_LOCKED)
1658 // area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1659 vx_vmlocked_sub(area->vm_mm, len >> PAGE_SHIFT);
1660 vm_stat_unaccount(area);
1661 area->vm_mm->unmap_area(area);
1662 remove_vm_struct(area);
1666 * Update the VMA and inode share lists.
1668 * Ok - we have the memory areas we should free on the 'free' list,
1669 * so release them, and do the vma updates.
1671 static void unmap_vma_list(struct mm_struct *mm,
1672 struct vm_area_struct *mpnt)
1675 struct vm_area_struct *next = mpnt->vm_next;
1676 unmap_vma(mm, mpnt);
1678 } while (mpnt != NULL);
1683 * Get rid of page table information in the indicated region.
1685 * Called with the page table lock held.
1687 static void unmap_region(struct mm_struct *mm,
1688 struct vm_area_struct *vma,
1689 struct vm_area_struct *prev,
1690 unsigned long start,
1693 struct mmu_gather *tlb;
1694 unsigned long nr_accounted = 0;
1697 tlb = tlb_gather_mmu(mm, 0);
1698 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1699 vm_unacct_memory(nr_accounted);
1701 if (is_hugepage_only_range(start, end - start))
1702 hugetlb_free_pgtables(tlb, prev, start, end);
1704 free_pgtables(tlb, prev, start, end);
1705 tlb_finish_mmu(tlb, start, end);
1709 * Create a list of vma's touched by the unmap, removing them from the mm's
1710 * vma list as we go..
1713 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1714 struct vm_area_struct *prev, unsigned long end)
1716 struct vm_area_struct **insertion_point;
1717 struct vm_area_struct *tail_vma = NULL;
1719 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1721 rb_erase(&vma->vm_rb, &mm->mm_rb);
1725 } while (vma && vma->vm_start < end);
1726 *insertion_point = vma;
1727 tail_vma->vm_next = NULL;
1728 mm->mmap_cache = NULL; /* Kill the cache. */
1732 * Split a vma into two pieces at address 'addr', a new vma is allocated
1733 * either for the first part or the the tail.
1735 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1736 unsigned long addr, int new_below)
1738 struct mempolicy *pol;
1739 struct vm_area_struct *new;
1741 if (mm->map_count >= sysctl_max_map_count)
1744 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1748 /* most fields are the same, copy all, and then fixup */
1754 new->vm_start = addr;
1755 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1758 pol = mpol_copy(vma_policy(vma));
1760 kmem_cache_free(vm_area_cachep, new);
1761 return PTR_ERR(pol);
1763 vma_set_policy(new, pol);
1766 get_file(new->vm_file);
1768 if (new->vm_ops && new->vm_ops->open)
1769 new->vm_ops->open(new);
1772 unsigned long old_end = vma->vm_end;
1774 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1775 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1776 if (vma->vm_flags & VM_EXEC)
1777 arch_remove_exec_range(mm, old_end);
1779 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1784 /* Munmap is split into 2 main parts -- this part which finds
1785 * what needs doing, and the areas themselves, which do the
1786 * work. This now handles partial unmappings.
1787 * Jeremy Fitzhardinge <jeremy@goop.org>
1789 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1792 struct vm_area_struct *mpnt, *prev, *last;
1794 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1797 if ((len = PAGE_ALIGN(len)) == 0)
1800 /* Find the first overlapping VMA */
1801 mpnt = find_vma_prev(mm, start, &prev);
1804 /* we have start < mpnt->vm_end */
1806 if (is_vm_hugetlb_page(mpnt)) {
1807 int ret = is_aligned_hugepage_range(start, len);
1813 /* if it doesn't overlap, we have nothing.. */
1815 if (mpnt->vm_start >= end)
1819 * If we need to split any vma, do it now to save pain later.
1821 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1822 * unmapped vm_area_struct will remain in use: so lower split_vma
1823 * places tmp vma above, and higher split_vma places tmp vma below.
1825 if (start > mpnt->vm_start) {
1826 if (split_vma(mm, mpnt, start, 0))
1831 /* Does it split the last one? */
1832 last = find_vma(mm, end);
1833 if (last && end > last->vm_start) {
1834 if (split_vma(mm, last, end, 1))
1837 mpnt = prev? prev->vm_next: mm->mmap;
1840 * Remove the vma's, and unmap the actual pages
1842 detach_vmas_to_be_unmapped(mm, mpnt, prev, end);
1843 spin_lock(&mm->page_table_lock);
1844 unmap_region(mm, mpnt, prev, start, end);
1845 spin_unlock(&mm->page_table_lock);
1847 /* Fix up all other VM information */
1848 unmap_vma_list(mm, mpnt);
1853 EXPORT_SYMBOL(do_munmap);
1855 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1858 struct mm_struct *mm = current->mm;
1860 profile_munmap(addr);
1862 down_write(&mm->mmap_sem);
1863 ret = do_munmap(mm, addr, len);
1864 up_write(&mm->mmap_sem);
1869 * this is really a simplified "do_mmap". it only handles
1870 * anonymous maps. eventually we may be able to do some
1871 * brk-specific accounting here.
1873 unsigned long __do_brk(unsigned long addr, unsigned long len)
1875 struct mm_struct * mm = current->mm;
1876 struct vm_area_struct * vma, * prev;
1877 unsigned long flags;
1878 struct rb_node ** rb_link, * rb_parent;
1879 pgoff_t pgoff = addr >> PAGE_SHIFT;
1881 len = PAGE_ALIGN(len);
1885 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1891 if (mm->def_flags & VM_LOCKED) {
1892 unsigned long locked, lock_limit;
1893 locked = mm->locked_vm << PAGE_SHIFT;
1894 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1896 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1898 if (!vx_vmlocked_avail(mm, len >> PAGE_SHIFT))
1903 * Clear old maps. this also does some error checking for us
1906 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1907 if (vma && vma->vm_start < addr + len) {
1908 if (do_munmap(mm, addr, len))
1913 /* Check against address space limits *after* clearing old maps... */
1914 if ((mm->total_vm << PAGE_SHIFT) + len
1915 > current->signal->rlim[RLIMIT_AS].rlim_cur)
1918 if (mm->map_count > sysctl_max_map_count)
1921 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
1922 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
1925 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1927 /* Can we just expand an old private anonymous mapping? */
1928 if (vma_merge(mm, prev, addr, addr + len, flags,
1929 NULL, NULL, pgoff, NULL))
1933 * create a vma struct for an anonymous mapping
1935 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1937 vm_unacct_memory(len >> PAGE_SHIFT);
1940 memset(vma, 0, sizeof(*vma));
1943 vma->vm_start = addr;
1944 vma->vm_end = addr + len;
1945 vma->vm_pgoff = pgoff;
1946 vma->vm_flags = flags;
1947 vma->vm_page_prot = protection_map[flags & 0x0f];
1948 vma_link(mm, vma, prev, rb_link, rb_parent);
1950 // mm->total_vm += len >> PAGE_SHIFT;
1951 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1952 if (flags & VM_LOCKED) {
1953 // mm->locked_vm += len >> PAGE_SHIFT;
1954 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1955 make_pages_present(addr, addr + len);
1960 EXPORT_SYMBOL(__do_brk);
1962 unsigned long do_brk(unsigned long addr, unsigned long len)
1966 down_write(¤t->mm->mmap_sem);
1967 ret = __do_brk(addr, len);
1968 up_write(¤t->mm->mmap_sem);
1972 EXPORT_SYMBOL(do_brk);
1975 /* Release all mmaps. */
1976 void exit_mmap(struct mm_struct *mm)
1978 struct mmu_gather *tlb;
1979 struct vm_area_struct *vma;
1980 unsigned long nr_accounted = 0;
1984 spin_lock(&mm->page_table_lock);
1986 tlb = tlb_gather_mmu(mm, 1);
1988 /* Use ~0UL here to ensure all VMAs in the mm are unmapped */
1989 mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0,
1990 ~0UL, &nr_accounted, NULL);
1991 vm_unacct_memory(nr_accounted);
1992 BUG_ON(mm->map_count); /* This is just debugging */
1993 clear_page_tables(tlb, FIRST_USER_PGD_NR, USER_PTRS_PER_PGD);
1994 tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm));
1997 mm->mmap = mm->mmap_cache = NULL;
1998 mm->mm_rb = RB_ROOT;
2000 vx_rsspages_sub(mm, mm->rss);
2001 // mm->total_vm = 0;
2002 vx_vmpages_sub(mm, mm->total_vm);
2003 // mm->locked_vm = 0;
2004 vx_vmlocked_sub(mm, mm->locked_vm);
2006 spin_unlock(&mm->page_table_lock);
2009 * Walk the list again, actually closing and freeing it
2010 * without holding any MM locks.
2013 struct vm_area_struct *next = vma->vm_next;
2014 remove_vm_struct(vma);
2019 /* Insert vm structure into process list sorted by address
2020 * and into the inode's i_mmap tree. If vm_file is non-NULL
2021 * then i_mmap_lock is taken here.
2023 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2025 struct vm_area_struct * __vma, * prev;
2026 struct rb_node ** rb_link, * rb_parent;
2029 * The vm_pgoff of a purely anonymous vma should be irrelevant
2030 * until its first write fault, when page's anon_vma and index
2031 * are set. But now set the vm_pgoff it will almost certainly
2032 * end up with (unless mremap moves it elsewhere before that
2033 * first wfault), so /proc/pid/maps tells a consistent story.
2035 * By setting it to reflect the virtual start address of the
2036 * vma, merges and splits can happen in a seamless way, just
2037 * using the existing file pgoff checks and manipulations.
2038 * Similarly in do_mmap_pgoff and in do_brk.
2040 if (!vma->vm_file) {
2041 BUG_ON(vma->anon_vma);
2042 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2044 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2045 if (__vma && __vma->vm_start < vma->vm_end)
2047 vma_link(mm, vma, prev, rb_link, rb_parent);
2052 * Copy the vma structure to a new location in the same mm,
2053 * prior to moving page table entries, to effect an mremap move.
2055 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2056 unsigned long addr, unsigned long len, pgoff_t pgoff)
2058 struct vm_area_struct *vma = *vmap;
2059 unsigned long vma_start = vma->vm_start;
2060 struct mm_struct *mm = vma->vm_mm;
2061 struct vm_area_struct *new_vma, *prev;
2062 struct rb_node **rb_link, *rb_parent;
2063 struct mempolicy *pol;
2066 * If anonymous vma has not yet been faulted, update new pgoff
2067 * to match new location, to increase its chance of merging.
2069 if (!vma->vm_file && !vma->anon_vma)
2070 pgoff = addr >> PAGE_SHIFT;
2072 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2073 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2074 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2077 * Source vma may have been merged into new_vma
2079 if (vma_start >= new_vma->vm_start &&
2080 vma_start < new_vma->vm_end)
2083 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2086 pol = mpol_copy(vma_policy(vma));
2088 kmem_cache_free(vm_area_cachep, new_vma);
2091 vma_set_policy(new_vma, pol);
2092 new_vma->vm_start = addr;
2093 new_vma->vm_end = addr + len;
2094 new_vma->vm_pgoff = pgoff;
2095 if (new_vma->vm_file)
2096 get_file(new_vma->vm_file);
2097 if (new_vma->vm_ops && new_vma->vm_ops->open)
2098 new_vma->vm_ops->open(new_vma);
2099 vma_link(mm, new_vma, prev, rb_link, rb_parent);