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>
27 #include <asm/uaccess.h>
28 #include <asm/pgalloc.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 = 0; /* default is 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;
95 struct address_space *mapping = file->f_mapping;
96 spin_lock(&mapping->i_mmap_lock);
97 __remove_shared_vm_struct(vma, file, mapping);
98 spin_unlock(&mapping->i_mmap_lock);
100 if (vma->vm_ops && vma->vm_ops->close)
101 vma->vm_ops->close(vma);
104 anon_vma_unlink(vma);
105 mpol_free(vma_policy(vma));
106 kmem_cache_free(vm_area_cachep, vma);
110 * sys_brk() for the most part doesn't need the global kernel
111 * lock, except when an application is doing something nasty
112 * like trying to un-brk an area that has already been mapped
113 * to a regular file. in this case, the unmapping will need
114 * to invoke file system routines that need the global lock.
116 asmlinkage unsigned long sys_brk(unsigned long brk)
118 unsigned long rlim, retval;
119 unsigned long newbrk, oldbrk;
120 struct mm_struct *mm = current->mm;
122 down_write(&mm->mmap_sem);
124 if (brk < mm->end_code)
126 newbrk = PAGE_ALIGN(brk);
127 oldbrk = PAGE_ALIGN(mm->brk);
128 if (oldbrk == newbrk)
131 /* Always allow shrinking brk. */
132 if (brk <= mm->brk) {
133 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
138 /* Check against rlimit.. */
139 rlim = current->rlim[RLIMIT_DATA].rlim_cur;
140 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
143 /* Check against existing mmap mappings. */
144 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
147 /* Ok, looks good - let it rip. */
148 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
154 up_write(&mm->mmap_sem);
159 static int browse_rb(struct rb_root *root)
162 struct rb_node *nd, *pn = NULL;
163 unsigned long prev = 0, pend = 0;
165 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
166 struct vm_area_struct *vma;
167 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
168 if (vma->vm_start < prev)
169 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
170 if (vma->vm_start < pend)
171 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
172 if (vma->vm_start > vma->vm_end)
173 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
178 for (nd = pn; nd; nd = rb_prev(nd)) {
182 printk("backwards %d, forwards %d\n", j, i), i = 0;
186 void validate_mm(struct mm_struct *mm)
190 struct vm_area_struct *tmp = mm->mmap;
195 if (i != mm->map_count)
196 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
197 i = browse_rb(&mm->mm_rb);
198 if (i != mm->map_count)
199 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
204 #define validate_mm(mm) do { } while (0)
207 static struct vm_area_struct *
208 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
209 struct vm_area_struct **pprev, struct rb_node ***rb_link,
210 struct rb_node ** rb_parent)
212 struct vm_area_struct * vma;
213 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
215 __rb_link = &mm->mm_rb.rb_node;
216 rb_prev = __rb_parent = NULL;
220 struct vm_area_struct *vma_tmp;
222 __rb_parent = *__rb_link;
223 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
225 if (vma_tmp->vm_end > addr) {
227 if (vma_tmp->vm_start <= addr)
229 __rb_link = &__rb_parent->rb_left;
231 rb_prev = __rb_parent;
232 __rb_link = &__rb_parent->rb_right;
238 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
239 *rb_link = __rb_link;
240 *rb_parent = __rb_parent;
245 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
246 struct vm_area_struct *prev, struct rb_node *rb_parent)
248 if (vma->vm_flags & VM_EXEC)
249 arch_add_exec_range(mm, vma->vm_end);
251 vma->vm_next = prev->vm_next;
256 vma->vm_next = rb_entry(rb_parent,
257 struct vm_area_struct, vm_rb);
263 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
264 struct rb_node **rb_link, struct rb_node *rb_parent)
266 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
267 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
270 static inline void __vma_link_file(struct vm_area_struct *vma)
276 struct address_space *mapping = file->f_mapping;
278 if (vma->vm_flags & VM_DENYWRITE)
279 atomic_dec(&file->f_dentry->d_inode->i_writecount);
280 if (vma->vm_flags & VM_SHARED)
281 mapping->i_mmap_writable++;
283 flush_dcache_mmap_lock(mapping);
284 if (unlikely(vma->vm_flags & VM_NONLINEAR))
285 list_add_tail(&vma->shared.vm_set.list,
286 &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_prio_tree_init(vma);
299 __vma_link_list(mm, vma, prev, rb_parent);
300 __vma_link_rb(mm, vma, rb_link, rb_parent);
301 __vma_link_file(vma);
302 __anon_vma_link(vma);
305 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
306 struct vm_area_struct *prev, struct rb_node **rb_link,
307 struct rb_node *rb_parent)
309 struct address_space *mapping = NULL;
312 mapping = vma->vm_file->f_mapping;
315 spin_lock(&mapping->i_mmap_lock);
317 __vma_link(mm, vma, prev, rb_link, rb_parent);
318 anon_vma_unlock(vma);
320 spin_unlock(&mapping->i_mmap_lock);
322 mark_mm_hugetlb(mm, vma);
328 * Insert vm structure into process list sorted by address and into the
329 * inode's i_mmap tree. The caller should hold mm->mmap_sem and
330 * ->f_mappping->i_mmap_lock if vm_file is non-NULL.
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);
342 mark_mm_hugetlb(mm, vma);
348 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
349 struct vm_area_struct *prev)
351 prev->vm_next = vma->vm_next;
352 rb_erase(&vma->vm_rb, &mm->mm_rb);
353 if (mm->mmap_cache == vma)
354 mm->mmap_cache = prev;
355 if (vma->vm_flags & VM_EXEC)
356 arch_remove_exec_range(mm, vma->vm_end);
360 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
361 * is already present in an i_mmap tree without adjusting the tree.
362 * The following helper function should be used when such adjustments
363 * are necessary. The "insert" vma (if any) is to be inserted
364 * before we drop the necessary locks.
366 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
367 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
369 struct mm_struct *mm = vma->vm_mm;
370 struct vm_area_struct *next = vma->vm_next;
371 struct address_space *mapping = NULL;
372 struct prio_tree_root *root = NULL;
373 struct file *file = vma->vm_file;
374 struct anon_vma *anon_vma = NULL;
375 long adjust_next = 0;
378 if (next && !insert) {
379 if (end >= next->vm_end) {
380 again: remove_next = 1 + (end > next->vm_end);
382 anon_vma = next->anon_vma;
383 } else if (end < vma->vm_end || end > next->vm_start) {
385 * vma shrinks, and !insert tells it's not
386 * split_vma inserting another: so it must
387 * be mprotect shifting the boundary down.
389 * vma expands, overlapping part of the next:
390 * must be mprotect shifting the boundary up.
392 BUG_ON(vma->vm_end != next->vm_start);
393 adjust_next = end - next->vm_start;
394 anon_vma = next->anon_vma;
399 mapping = file->f_mapping;
400 if (!(vma->vm_flags & VM_NONLINEAR))
401 root = &mapping->i_mmap;
402 spin_lock(&mapping->i_mmap_lock);
406 * When changing only vma->vm_end, we don't really need
407 * anon_vma lock: but is that case worth optimizing out?
410 anon_vma = vma->anon_vma;
412 spin_lock(&anon_vma->lock);
415 flush_dcache_mmap_lock(mapping);
416 vma_prio_tree_remove(vma, root);
418 vma_prio_tree_remove(next, root);
421 vma->vm_start = start;
423 vma->vm_pgoff = pgoff;
425 next->vm_start += adjust_next;
426 next->vm_pgoff += adjust_next >> PAGE_SHIFT;
431 vma_prio_tree_init(next);
432 vma_prio_tree_insert(next, root);
434 vma_prio_tree_init(vma);
435 vma_prio_tree_insert(vma, root);
436 flush_dcache_mmap_unlock(mapping);
441 * vma_merge has merged next into vma, and needs
442 * us to remove next before dropping the locks.
444 __vma_unlink(mm, next, vma);
446 __remove_shared_vm_struct(next, file, mapping);
448 __anon_vma_merge(vma, next);
451 * split_vma has split insert from vma, and needs
452 * us to insert it before dropping the locks
453 * (it may either follow vma or precede it).
455 __insert_vm_struct(mm, insert);
459 spin_unlock(&anon_vma->lock);
461 spin_unlock(&mapping->i_mmap_lock);
467 mpol_free(vma_policy(next));
468 kmem_cache_free(vm_area_cachep, next);
470 * In mprotect's case 6 (see comments on vma_merge),
471 * we must remove another next too. It would clutter
472 * up the code too much to do both in one go.
474 if (remove_next == 2) {
484 * If the vma has a ->close operation then the driver probably needs to release
485 * per-vma resources, so we don't attempt to merge those.
487 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
489 static inline int is_mergeable_vma(struct vm_area_struct *vma,
490 struct file *file, unsigned long vm_flags)
492 if (vma->vm_flags != vm_flags)
494 if (vma->vm_file != file)
496 if (vma->vm_ops && vma->vm_ops->close)
501 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
502 struct anon_vma *anon_vma2)
504 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
508 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
509 * in front of (at a lower virtual address and file offset than) the vma.
511 * We cannot merge two vmas if they have differently assigned (non-NULL)
512 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
514 * We don't check here for the merged mmap wrapping around the end of pagecache
515 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
516 * wrap, nor mmaps which cover the final page at index -1UL.
519 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
520 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
522 if (is_mergeable_vma(vma, file, vm_flags) &&
523 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
524 if (vma->vm_pgoff == vm_pgoff)
531 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
532 * beyond (at a higher virtual address and file offset than) the vma.
534 * We cannot merge two vmas if they have differently assigned (non-NULL)
535 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
538 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
539 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
541 if (is_mergeable_vma(vma, file, vm_flags) &&
542 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
544 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
545 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
552 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
553 * whether that can be merged with its predecessor or its successor.
554 * Or both (it neatly fills a hole).
556 * In most cases - when called for mmap, brk or mremap - [addr,end) is
557 * certain not to be mapped by the time vma_merge is called; but when
558 * called for mprotect, it is certain to be already mapped (either at
559 * an offset within prev, or at the start of next), and the flags of
560 * this area are about to be changed to vm_flags - and the no-change
561 * case has already been eliminated.
563 * The following mprotect cases have to be considered, where AAAA is
564 * the area passed down from mprotect_fixup, never extending beyond one
565 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
567 * AAAA AAAA AAAA AAAA
568 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
569 * cannot merge might become might become might become
570 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
571 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
572 * mremap move: PPPPNNNNNNNN 8
574 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
575 * might become case 1 below case 2 below case 3 below
577 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
578 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
580 struct vm_area_struct *vma_merge(struct mm_struct *mm,
581 struct vm_area_struct *prev, unsigned long addr,
582 unsigned long end, unsigned long vm_flags,
583 struct anon_vma *anon_vma, struct file *file,
584 pgoff_t pgoff, struct mempolicy *policy)
586 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
587 struct vm_area_struct *area, *next;
590 * We later require that vma->vm_flags == vm_flags,
591 * so this tests vma->vm_flags & VM_SPECIAL, too.
593 if (vm_flags & VM_SPECIAL)
597 next = prev->vm_next;
601 if (next && next->vm_end == end) /* cases 6, 7, 8 */
602 next = next->vm_next;
605 * Can it merge with the predecessor?
607 if (prev && prev->vm_end == addr &&
608 mpol_equal(vma_policy(prev), policy) &&
609 can_vma_merge_after(prev, vm_flags,
610 anon_vma, file, pgoff)) {
612 * OK, it can. Can we now merge in the successor as well?
614 if (next && end == next->vm_start &&
615 mpol_equal(policy, vma_policy(next)) &&
616 can_vma_merge_before(next, vm_flags,
617 anon_vma, file, pgoff+pglen) &&
618 is_mergeable_anon_vma(prev->anon_vma,
621 vma_adjust(prev, prev->vm_start,
622 next->vm_end, prev->vm_pgoff, NULL);
623 } else /* cases 2, 5, 7 */
624 vma_adjust(prev, prev->vm_start,
625 end, prev->vm_pgoff, NULL);
626 if (prev->vm_flags & VM_EXEC)
627 arch_add_exec_range(mm, prev->vm_end);
632 * Can this new request be merged in front of next?
634 if (next && end == next->vm_start &&
635 mpol_equal(policy, vma_policy(next)) &&
636 can_vma_merge_before(next, vm_flags,
637 anon_vma, file, pgoff+pglen)) {
638 if (prev && addr < prev->vm_end) /* case 4 */
639 vma_adjust(prev, prev->vm_start,
640 addr, prev->vm_pgoff, NULL);
641 else /* cases 3, 8 */
642 vma_adjust(area, addr, next->vm_end,
643 next->vm_pgoff - pglen, NULL);
651 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
652 * neighbouring vmas for a suitable anon_vma, before it goes off
653 * to allocate a new anon_vma. It checks because a repetitive
654 * sequence of mprotects and faults may otherwise lead to distinct
655 * anon_vmas being allocated, preventing vma merge in subsequent
658 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
660 struct vm_area_struct *near;
661 unsigned long vm_flags;
668 * Since only mprotect tries to remerge vmas, match flags
669 * which might be mprotected into each other later on.
670 * Neither mlock nor madvise tries to remerge at present,
671 * so leave their flags as obstructing a merge.
673 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
674 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
676 if (near->anon_vma && vma->vm_end == near->vm_start &&
677 mpol_equal(vma_policy(vma), vma_policy(near)) &&
678 can_vma_merge_before(near, vm_flags,
679 NULL, vma->vm_file, vma->vm_pgoff +
680 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
681 return near->anon_vma;
684 * It is potentially slow to have to call find_vma_prev here.
685 * But it's only on the first write fault on the vma, not
686 * every time, and we could devise a way to avoid it later
687 * (e.g. stash info in next's anon_vma_node when assigning
688 * an anon_vma, or when trying vma_merge). Another time.
690 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
695 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
696 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
698 if (near->anon_vma && near->vm_end == vma->vm_start &&
699 mpol_equal(vma_policy(near), vma_policy(vma)) &&
700 can_vma_merge_after(near, vm_flags,
701 NULL, vma->vm_file, vma->vm_pgoff))
702 return near->anon_vma;
705 * There's no absolute need to look only at touching neighbours:
706 * we could search further afield for "compatible" anon_vmas.
707 * But it would probably just be a waste of time searching,
708 * or lead to too many vmas hanging off the same anon_vma.
709 * We're trying to allow mprotect remerging later on,
710 * not trying to minimize memory used for anon_vmas.
716 * The caller must hold down_write(current->mm->mmap_sem).
719 unsigned long do_mmap_pgoff(struct mm_struct *mm, struct file * file,
720 unsigned long addr, unsigned long len,
721 unsigned long prot, unsigned long flags,
724 struct vm_area_struct * vma, * prev;
726 unsigned int vm_flags;
727 int correct_wcount = 0;
729 struct rb_node ** rb_link, * rb_parent;
731 unsigned long charged = 0;
734 if (is_file_hugepages(file))
737 if (!file->f_op || !file->f_op->mmap)
740 if ((prot & PROT_EXEC) &&
741 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
748 /* Careful about overflows.. */
749 len = PAGE_ALIGN(len);
750 if (!len || len > TASK_SIZE)
753 /* offset overflow? */
754 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
757 /* Too many mappings? */
758 if (mm->map_count > sysctl_max_map_count)
761 /* Obtain the address to map to. we verify (or select) it and ensure
762 * that it represents a valid section of the address space.
764 addr = get_unmapped_area(file, addr, len, pgoff, flags, prot & PROT_EXEC);
765 if (addr & ~PAGE_MASK)
768 /* Do simple checking here so the lower-level routines won't have
769 * to. we assume access permissions have been handled by the open
770 * of the memory object, so we don't do any here.
772 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
773 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
775 if (flags & MAP_LOCKED) {
778 vm_flags |= VM_LOCKED;
780 /* mlock MCL_FUTURE? */
781 if (vm_flags & VM_LOCKED) {
782 unsigned long locked, lock_limit;
783 locked = mm->locked_vm << PAGE_SHIFT;
784 lock_limit = current->rlim[RLIMIT_MEMLOCK].rlim_cur;
786 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
790 inode = file ? file->f_dentry->d_inode : NULL;
793 switch (flags & MAP_TYPE) {
795 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
799 * Make sure we don't allow writing to an append-only
802 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
806 * Make sure there are no mandatory locks on the file.
808 if (locks_verify_locked(inode))
811 vm_flags |= VM_SHARED | VM_MAYSHARE;
812 if (!(file->f_mode & FMODE_WRITE))
813 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
817 if (!(file->f_mode & FMODE_READ))
825 switch (flags & MAP_TYPE) {
827 vm_flags |= VM_SHARED | VM_MAYSHARE;
831 * Set pgoff according to addr for anon_vma.
833 pgoff = addr >> PAGE_SHIFT;
840 error = security_file_mmap(file, prot, flags);
847 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
848 if (vma && vma->vm_start < addr + len) {
849 if (do_munmap(mm, addr, len))
854 /* Check against address space limit. */
855 if ((mm->total_vm << PAGE_SHIFT) + len
856 > current->rlim[RLIMIT_AS].rlim_cur)
859 /* check context space, maybe only Private writable mapping? */
860 if (!vx_vmpages_avail(mm, len >> PAGE_SHIFT))
863 if (accountable && (!(flags & MAP_NORESERVE) ||
864 sysctl_overcommit_memory > 1)) {
865 if (vm_flags & VM_SHARED) {
866 /* Check memory availability in shmem_file_setup? */
867 vm_flags |= VM_ACCOUNT;
868 } else if (vm_flags & VM_WRITE) {
870 * Private writable mapping: check memory availability
872 charged = len >> PAGE_SHIFT;
873 if (security_vm_enough_memory(charged))
875 vm_flags |= VM_ACCOUNT;
880 * Can we just expand an old private anonymous mapping?
881 * The VM_SHARED test is necessary because shmem_zero_setup
882 * will create the file object for a shared anonymous map below.
884 if (!file && !(vm_flags & VM_SHARED) &&
885 vma_merge(mm, prev, addr, addr + len, vm_flags,
886 NULL, NULL, pgoff, NULL))
890 * Determine the object being mapped and call the appropriate
891 * specific mapper. the address has already been validated, but
892 * not unmapped, but the maps are removed from the list.
894 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
899 memset(vma, 0, sizeof(*vma));
902 vma->vm_start = addr;
903 vma->vm_end = addr + len;
904 vma->vm_flags = vm_flags;
905 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
906 vma->vm_pgoff = pgoff;
910 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
912 if (vm_flags & VM_DENYWRITE) {
913 error = deny_write_access(file);
920 error = file->f_op->mmap(file, vma);
922 goto unmap_and_free_vma;
923 } else if (vm_flags & VM_SHARED) {
924 error = shmem_zero_setup(vma);
929 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
930 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
931 * that memory reservation must be checked; but that reservation
932 * belongs to shared memory object, not to vma: so now clear it.
934 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
935 vma->vm_flags &= ~VM_ACCOUNT;
937 /* Can addr have changed??
939 * Answer: Yes, several device drivers can do it in their
940 * f_op->mmap method. -DaveM
942 addr = vma->vm_start;
944 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
945 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
946 vma_link(mm, vma, prev, rb_link, rb_parent);
948 atomic_inc(&inode->i_writecount);
952 atomic_inc(&inode->i_writecount);
955 mpol_free(vma_policy(vma));
956 kmem_cache_free(vm_area_cachep, vma);
959 // mm->total_vm += len >> PAGE_SHIFT;
960 vx_vmpages_add(mm, len >> PAGE_SHIFT);
961 if (vm_flags & VM_LOCKED) {
962 // mm->locked_vm += len >> PAGE_SHIFT;
963 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
964 make_pages_present(addr, addr + len);
966 if (flags & MAP_POPULATE) {
967 up_write(&mm->mmap_sem);
968 sys_remap_file_pages(addr, len, 0,
969 pgoff, flags & MAP_NONBLOCK);
970 down_write(&mm->mmap_sem);
976 atomic_inc(&inode->i_writecount);
980 /* Undo any partial mapping done by a device driver. */
981 zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
983 kmem_cache_free(vm_area_cachep, vma);
986 vm_unacct_memory(charged);
990 EXPORT_SYMBOL(do_mmap_pgoff);
992 /* Get an address range which is currently unmapped.
993 * For shmat() with addr=0.
995 * Ugly calling convention alert:
996 * Return value with the low bits set means error value,
998 * if (ret & ~PAGE_MASK)
1001 * This function "knows" that -ENOMEM has the bits set.
1003 #ifndef HAVE_ARCH_UNMAPPED_AREA
1004 static inline unsigned long
1005 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1006 unsigned long len, unsigned long pgoff, unsigned long flags, unsigned long exec)
1008 struct mm_struct *mm = current->mm;
1009 struct vm_area_struct *vma;
1010 unsigned long start_addr;
1012 if (len > TASK_SIZE)
1016 addr = PAGE_ALIGN(addr);
1017 vma = find_vma(mm, addr);
1018 if (TASK_SIZE - len >= addr &&
1019 (!vma || addr + len <= vma->vm_start))
1022 start_addr = addr = mm->free_area_cache;
1025 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1026 /* At this point: (!vma || addr < vma->vm_end). */
1027 if (TASK_SIZE - len < addr) {
1029 * Start a new search - just in case we missed
1032 if (start_addr != TASK_UNMAPPED_BASE) {
1033 start_addr = addr = TASK_UNMAPPED_BASE;
1038 if (!vma || addr + len <= vma->vm_start) {
1040 * Remember the place where we stopped the search:
1042 mm->free_area_cache = addr + len;
1049 extern unsigned long
1050 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
1051 unsigned long, unsigned long, unsigned long);
1055 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1056 unsigned long pgoff, unsigned long flags, unsigned long exec)
1058 if (flags & MAP_FIXED) {
1061 if (addr > TASK_SIZE - len)
1063 if (addr & ~PAGE_MASK)
1065 if (file && is_file_hugepages(file)) {
1067 * Check if the given range is hugepage aligned, and
1068 * can be made suitable for hugepages.
1070 ret = prepare_hugepage_range(addr, len);
1073 * Ensure that a normal request is not falling in a
1074 * reserved hugepage range. For some archs like IA-64,
1075 * there is a separate region for hugepages.
1077 ret = is_hugepage_only_range(addr, len);
1084 if (file && file->f_op && file->f_op->get_unmapped_area)
1085 return file->f_op->get_unmapped_area(file, addr, len,
1088 return arch_get_unmapped_area(file, addr, len, pgoff, flags, exec);
1091 EXPORT_SYMBOL(get_unmapped_area);
1093 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1094 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1096 struct vm_area_struct *vma = NULL;
1099 /* Check the cache first. */
1100 /* (Cache hit rate is typically around 35%.) */
1101 vma = mm->mmap_cache;
1102 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1103 struct rb_node * rb_node;
1105 rb_node = mm->mm_rb.rb_node;
1109 struct vm_area_struct * vma_tmp;
1111 vma_tmp = rb_entry(rb_node,
1112 struct vm_area_struct, vm_rb);
1114 if (vma_tmp->vm_end > addr) {
1116 if (vma_tmp->vm_start <= addr)
1118 rb_node = rb_node->rb_left;
1120 rb_node = rb_node->rb_right;
1123 mm->mmap_cache = vma;
1129 EXPORT_SYMBOL(find_vma);
1131 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1132 struct vm_area_struct *
1133 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1134 struct vm_area_struct **pprev)
1136 struct vm_area_struct *vma = NULL, *prev = NULL;
1137 struct rb_node * rb_node;
1141 /* Guard against addr being lower than the first VMA */
1144 /* Go through the RB tree quickly. */
1145 rb_node = mm->mm_rb.rb_node;
1148 struct vm_area_struct *vma_tmp;
1149 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1151 if (addr < vma_tmp->vm_end) {
1152 rb_node = rb_node->rb_left;
1155 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1157 rb_node = rb_node->rb_right;
1163 return prev ? prev->vm_next : vma;
1167 static int over_stack_limit(unsigned long sz)
1169 if (sz < EXEC_STACK_BIAS)
1171 return (sz - EXEC_STACK_BIAS) > current->rlim[RLIMIT_STACK].rlim_cur;
1174 #ifdef CONFIG_STACK_GROWSUP
1176 * vma is the first one with address > vma->vm_end. Have to extend vma.
1178 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1182 if (!(vma->vm_flags & VM_GROWSUP))
1186 * We must make sure the anon_vma is allocated
1187 * so that the anon_vma locking is not a noop.
1189 if (unlikely(anon_vma_prepare(vma)))
1194 * vma->vm_start/vm_end cannot change under us because the caller
1195 * is required to hold the mmap_sem in read mode. We need the
1196 * anon_vma lock to serialize against concurrent expand_stacks.
1198 address += 4 + PAGE_SIZE - 1;
1199 address &= PAGE_MASK;
1200 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1203 if (security_vm_enough_memory(grow) ||
1204 !vx_vmpages_avail(vma->vm_mm, grow)) {
1205 anon_vma_unlock(vma);
1209 if (over_stack_limit(address - vma->vm_start) ||
1210 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1211 current->rlim[RLIMIT_AS].rlim_cur) {
1212 anon_vma_unlock(vma);
1213 vm_unacct_memory(grow);
1217 vma->vm_end = address;
1218 // vma->vm_mm->total_vm += grow;
1219 vx_vmpages_add(vma->vm_mm, grow);
1220 if (vma->vm_flags & VM_LOCKED)
1221 // vma->vm_mm->locked_vm += grow;
1222 vx_vmlocked_add(vma->vm_mm, grow);
1223 anon_vma_unlock(vma);
1227 struct vm_area_struct *
1228 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1230 struct vm_area_struct *vma, *prev;
1233 vma = find_vma_prev(mm, addr, &prev);
1234 if (vma && (vma->vm_start <= addr))
1236 if (!prev || expand_stack(prev, addr))
1238 if (prev->vm_flags & VM_LOCKED) {
1239 make_pages_present(addr, prev->vm_end);
1245 * vma is the first one with address < vma->vm_start. Have to extend vma.
1247 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1252 * We must make sure the anon_vma is allocated
1253 * so that the anon_vma locking is not a noop.
1255 if (unlikely(anon_vma_prepare(vma)))
1260 * vma->vm_start/vm_end cannot change under us because the caller
1261 * is required to hold the mmap_sem in read mode. We need the
1262 * anon_vma lock to serialize against concurrent expand_stacks.
1264 address &= PAGE_MASK;
1265 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1268 if (security_vm_enough_memory(grow) ||
1269 !vx_vmpages_avail(vma->vm_mm, grow)) {
1270 anon_vma_unlock(vma);
1274 if (over_stack_limit(vma->vm_end - address) ||
1275 ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
1276 current->rlim[RLIMIT_AS].rlim_cur) {
1277 anon_vma_unlock(vma);
1278 vm_unacct_memory(grow);
1282 vma->vm_start = address;
1283 vma->vm_pgoff -= grow;
1284 // vma->vm_mm->total_vm += grow;
1285 vx_vmpages_add(vma->vm_mm, grow);
1286 if (vma->vm_flags & VM_LOCKED)
1287 // vma->vm_mm->locked_vm += grow;
1288 vx_vmlocked_add(vma->vm_mm, grow);
1289 anon_vma_unlock(vma);
1293 struct vm_area_struct *
1294 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1296 struct vm_area_struct * vma;
1297 unsigned long start;
1300 vma = find_vma(mm,addr);
1303 if (vma->vm_start <= addr)
1305 if (!(vma->vm_flags & VM_GROWSDOWN))
1307 start = vma->vm_start;
1308 if (expand_stack(vma, addr))
1310 if (vma->vm_flags & VM_LOCKED) {
1311 make_pages_present(addr, start);
1318 * Try to free as many page directory entries as we can,
1319 * without having to work very hard at actually scanning
1320 * the page tables themselves.
1322 * Right now we try to free page tables if we have a nice
1323 * PGDIR-aligned area that got free'd up. We could be more
1324 * granular if we want to, but this is fast and simple,
1325 * and covers the bad cases.
1327 * "prev", if it exists, points to a vma before the one
1328 * we just free'd - but there's no telling how much before.
1330 static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev,
1331 unsigned long start, unsigned long end)
1333 unsigned long first = start & PGDIR_MASK;
1334 unsigned long last = end + PGDIR_SIZE - 1;
1335 unsigned long start_index, end_index;
1336 struct mm_struct *mm = tlb->mm;
1342 if (prev->vm_end > start) {
1343 if (last > prev->vm_start)
1344 last = prev->vm_start;
1349 struct vm_area_struct *next = prev->vm_next;
1352 if (next->vm_start < start) {
1356 if (last > next->vm_start)
1357 last = next->vm_start;
1359 if (prev->vm_end > first)
1360 first = prev->vm_end + PGDIR_SIZE - 1;
1364 if (last < first) /* for arches with discontiguous pgd indices */
1367 * If the PGD bits are not consecutive in the virtual address, the
1368 * old method of shifting the VA >> by PGDIR_SHIFT doesn't work.
1370 start_index = pgd_index(first);
1371 if (start_index < FIRST_USER_PGD_NR)
1372 start_index = FIRST_USER_PGD_NR;
1373 end_index = pgd_index(last);
1374 if (end_index > start_index) {
1375 clear_page_tables(tlb, start_index, end_index - start_index);
1376 flush_tlb_pgtables(mm, first & PGDIR_MASK, last & PGDIR_MASK);
1380 /* Normal function to fix up a mapping
1381 * This function is the default for when an area has no specific
1382 * function. This may be used as part of a more specific routine.
1384 * By the time this function is called, the area struct has been
1385 * removed from the process mapping list.
1387 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1389 size_t len = area->vm_end - area->vm_start;
1390 unsigned long old_end = area->vm_end;
1392 // area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1393 vx_vmpages_sub(area->vm_mm, len >> PAGE_SHIFT);
1395 if (area->vm_flags & VM_LOCKED)
1396 // area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1397 vx_vmlocked_sub(area->vm_mm, len >> PAGE_SHIFT);
1399 * Is this a new hole at the lowest possible address?
1401 if (area->vm_start >= TASK_UNMAPPED_BASE &&
1402 area->vm_start < area->vm_mm->free_area_cache)
1403 area->vm_mm->free_area_cache = area->vm_start;
1405 * Is this a new hole at the highest possible address?
1407 if (area->vm_start > area->vm_mm->non_executable_cache)
1408 area->vm_mm->non_executable_cache = area->vm_start;
1409 remove_vm_struct(area);
1410 if (unlikely(area->vm_flags & VM_EXEC))
1411 arch_remove_exec_range(mm, old_end);
1415 * Update the VMA and inode share lists.
1417 * Ok - we have the memory areas we should free on the 'free' list,
1418 * so release them, and do the vma updates.
1420 static void unmap_vma_list(struct mm_struct *mm,
1421 struct vm_area_struct *mpnt)
1424 struct vm_area_struct *next = mpnt->vm_next;
1425 unmap_vma(mm, mpnt);
1427 } while (mpnt != NULL);
1432 * Get rid of page table information in the indicated region.
1434 * Called with the page table lock held.
1436 static void unmap_region(struct mm_struct *mm,
1437 struct vm_area_struct *vma,
1438 struct vm_area_struct *prev,
1439 unsigned long start,
1442 struct mmu_gather *tlb;
1443 unsigned long nr_accounted = 0;
1446 tlb = tlb_gather_mmu(mm, 0);
1447 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1448 vm_unacct_memory(nr_accounted);
1450 if (is_hugepage_only_range(start, end - start))
1451 hugetlb_free_pgtables(tlb, prev, start, end);
1453 free_pgtables(tlb, prev, start, end);
1454 tlb_finish_mmu(tlb, start, end);
1458 * Create a list of vma's touched by the unmap, removing them from the mm's
1459 * vma list as we go..
1462 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1463 struct vm_area_struct *prev, unsigned long end)
1465 struct vm_area_struct **insertion_point;
1466 struct vm_area_struct *tail_vma = NULL;
1468 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1470 rb_erase(&vma->vm_rb, &mm->mm_rb);
1474 } while (vma && vma->vm_start < end);
1475 *insertion_point = vma;
1476 tail_vma->vm_next = NULL;
1477 mm->mmap_cache = NULL; /* Kill the cache. */
1481 * Split a vma into two pieces at address 'addr', a new vma is allocated
1482 * either for the first part or the the tail.
1484 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1485 unsigned long addr, int new_below)
1487 struct mempolicy *pol;
1488 struct vm_area_struct *new;
1490 if (mm->map_count >= sysctl_max_map_count)
1493 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1497 /* most fields are the same, copy all, and then fixup */
1503 new->vm_start = addr;
1504 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1507 pol = mpol_copy(vma_policy(vma));
1509 kmem_cache_free(vm_area_cachep, new);
1510 return PTR_ERR(pol);
1512 vma_set_policy(new, pol);
1515 get_file(new->vm_file);
1517 if (new->vm_ops && new->vm_ops->open)
1518 new->vm_ops->open(new);
1521 unsigned long old_end = vma->vm_end;
1523 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1524 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1525 if (vma->vm_flags & VM_EXEC)
1526 arch_remove_exec_range(mm, old_end);
1528 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1533 /* Munmap is split into 2 main parts -- this part which finds
1534 * what needs doing, and the areas themselves, which do the
1535 * work. This now handles partial unmappings.
1536 * Jeremy Fitzhardinge <jeremy@goop.org>
1538 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1541 struct vm_area_struct *mpnt, *prev, *last;
1543 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1546 if ((len = PAGE_ALIGN(len)) == 0)
1549 /* Find the first overlapping VMA */
1550 mpnt = find_vma_prev(mm, start, &prev);
1553 /* we have start < mpnt->vm_end */
1555 if (is_vm_hugetlb_page(mpnt)) {
1556 int ret = is_aligned_hugepage_range(start, len);
1562 /* if it doesn't overlap, we have nothing.. */
1564 if (mpnt->vm_start >= end)
1567 /* Something will probably happen, so notify. */
1568 if (mpnt->vm_file && (mpnt->vm_flags & VM_EXEC))
1569 profile_exec_unmap(mm);
1572 * If we need to split any vma, do it now to save pain later.
1574 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1575 * unmapped vm_area_struct will remain in use: so lower split_vma
1576 * places tmp vma above, and higher split_vma places tmp vma below.
1578 if (start > mpnt->vm_start) {
1579 if (split_vma(mm, mpnt, start, 0))
1584 /* Does it split the last one? */
1585 last = find_vma(mm, end);
1586 if (last && end > last->vm_start) {
1587 if (split_vma(mm, last, end, 1))
1590 mpnt = prev? prev->vm_next: mm->mmap;
1593 * Remove the vma's, and unmap the actual pages
1595 detach_vmas_to_be_unmapped(mm, mpnt, prev, end);
1596 spin_lock(&mm->page_table_lock);
1597 unmap_region(mm, mpnt, prev, start, end);
1598 spin_unlock(&mm->page_table_lock);
1600 /* Fix up all other VM information */
1601 unmap_vma_list(mm, mpnt);
1606 EXPORT_SYMBOL(do_munmap);
1608 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1611 struct mm_struct *mm = current->mm;
1613 down_write(&mm->mmap_sem);
1614 ret = do_munmap(mm, addr, len);
1615 up_write(&mm->mmap_sem);
1620 * this is really a simplified "do_mmap". it only handles
1621 * anonymous maps. eventually we may be able to do some
1622 * brk-specific accounting here.
1624 unsigned long do_brk(unsigned long addr, unsigned long len)
1626 struct mm_struct * mm = current->mm;
1627 struct vm_area_struct * vma, * prev;
1628 unsigned long flags;
1629 struct rb_node ** rb_link, * rb_parent;
1630 pgoff_t pgoff = addr >> PAGE_SHIFT;
1632 len = PAGE_ALIGN(len);
1636 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1642 if (mm->def_flags & VM_LOCKED) {
1643 unsigned long locked, lock_limit;
1644 locked = mm->locked_vm << PAGE_SHIFT;
1645 lock_limit = current->rlim[RLIMIT_MEMLOCK].rlim_cur;
1647 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1649 /* vserver checks ? */
1653 * Clear old maps. this also does some error checking for us
1656 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1657 if (vma && vma->vm_start < addr + len) {
1658 if (do_munmap(mm, addr, len))
1663 /* Check against address space limits *after* clearing old maps... */
1664 if ((mm->total_vm << PAGE_SHIFT) + len
1665 > current->rlim[RLIMIT_AS].rlim_cur)
1668 if (mm->map_count > sysctl_max_map_count)
1671 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
1672 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
1675 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1677 /* Can we just expand an old private anonymous mapping? */
1678 if (vma_merge(mm, prev, addr, addr + len, flags,
1679 NULL, NULL, pgoff, NULL))
1683 * create a vma struct for an anonymous mapping
1685 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1687 vm_unacct_memory(len >> PAGE_SHIFT);
1690 memset(vma, 0, sizeof(*vma));
1693 vma->vm_start = addr;
1694 vma->vm_end = addr + len;
1695 vma->vm_pgoff = pgoff;
1696 vma->vm_flags = flags;
1697 vma->vm_page_prot = protection_map[flags & 0x0f];
1698 vma_link(mm, vma, prev, rb_link, rb_parent);
1700 // mm->total_vm += len >> PAGE_SHIFT;
1701 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1702 if (flags & VM_LOCKED) {
1703 // mm->locked_vm += len >> PAGE_SHIFT;
1704 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1705 make_pages_present(addr, addr + len);
1710 EXPORT_SYMBOL(do_brk);
1712 /* Release all mmaps. */
1713 void exit_mmap(struct mm_struct *mm)
1715 struct mmu_gather *tlb;
1716 struct vm_area_struct *vma;
1717 unsigned long nr_accounted = 0;
1719 profile_exit_mmap(mm);
1723 spin_lock(&mm->page_table_lock);
1725 tlb = tlb_gather_mmu(mm, 1);
1727 /* Use ~0UL here to ensure all VMAs in the mm are unmapped */
1728 mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0,
1729 ~0UL, &nr_accounted, NULL);
1730 vm_unacct_memory(nr_accounted);
1731 BUG_ON(mm->map_count); /* This is just debugging */
1732 clear_page_tables(tlb, FIRST_USER_PGD_NR, USER_PTRS_PER_PGD);
1733 tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm));
1736 mm->mmap = mm->mmap_cache = NULL;
1737 mm->mm_rb = RB_ROOT;
1739 vx_rsspages_sub(mm, mm->rss);
1740 // mm->total_vm = 0;
1741 vx_vmpages_sub(mm, mm->total_vm);
1742 // mm->locked_vm = 0;
1743 vx_vmlocked_sub(mm, mm->locked_vm);
1744 arch_flush_exec_range(mm);
1746 spin_unlock(&mm->page_table_lock);
1749 * Walk the list again, actually closing and freeing it
1750 * without holding any MM locks.
1753 struct vm_area_struct *next = vma->vm_next;
1754 remove_vm_struct(vma);
1759 /* Insert vm structure into process list sorted by address
1760 * and into the inode's i_mmap tree. If vm_file is non-NULL
1761 * then i_mmap_lock is taken here.
1763 void insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1765 struct vm_area_struct * __vma, * prev;
1766 struct rb_node ** rb_link, * rb_parent;
1769 * The vm_pgoff of a purely anonymous vma should be irrelevant
1770 * until its first write fault, when page's anon_vma and index
1771 * are set. But now set the vm_pgoff it will almost certainly
1772 * end up with (unless mremap moves it elsewhere before that
1773 * first wfault), so /proc/pid/maps tells a consistent story.
1775 * By setting it to reflect the virtual start address of the
1776 * vma, merges and splits can happen in a seamless way, just
1777 * using the existing file pgoff checks and manipulations.
1778 * Similarly in do_mmap_pgoff and in do_brk.
1780 if (!vma->vm_file) {
1781 BUG_ON(vma->anon_vma);
1782 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1784 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1785 if (__vma && __vma->vm_start < vma->vm_end)
1787 vma_link(mm, vma, prev, rb_link, rb_parent);
1791 * Copy the vma structure to a new location in the same mm,
1792 * prior to moving page table entries, to effect an mremap move.
1794 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
1795 unsigned long addr, unsigned long len, pgoff_t pgoff)
1797 struct vm_area_struct *vma = *vmap;
1798 unsigned long vma_start = vma->vm_start;
1799 struct mm_struct *mm = vma->vm_mm;
1800 struct vm_area_struct *new_vma, *prev;
1801 struct rb_node **rb_link, *rb_parent;
1802 struct mempolicy *pol;
1805 * If anonymous vma has not yet been faulted, update new pgoff
1806 * to match new location, to increase its chance of merging.
1808 if (!vma->vm_file && !vma->anon_vma)
1809 pgoff = addr >> PAGE_SHIFT;
1811 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1812 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
1813 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
1816 * Source vma may have been merged into new_vma
1818 if (vma_start >= new_vma->vm_start &&
1819 vma_start < new_vma->vm_end)
1822 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1825 pol = mpol_copy(vma_policy(vma));
1827 kmem_cache_free(vm_area_cachep, new_vma);
1830 vma_set_policy(new_vma, pol);
1831 new_vma->vm_start = addr;
1832 new_vma->vm_end = addr + len;
1833 new_vma->vm_pgoff = pgoff;
1834 if (new_vma->vm_file)
1835 get_file(new_vma->vm_file);
1836 if (new_vma->vm_ops && new_vma->vm_ops->open)
1837 new_vma->vm_ops->open(new_vma);
1838 vma_link(mm, new_vma, prev, rb_link, rb_parent);