1 /*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
6 * Address space accounting code <alan@redhat.com>
7 */
8
9 #include <linux/slab.h>
10 #include <linux/mm.h>
11 #include <linux/shm.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swap.h>
15 #include <linux/syscalls.h>
16 #include <linux/init.h>
17 #include <linux/file.h>
18 #include <linux/fs.h>
19 #include <linux/personality.h>
20 #include <linux/security.h>
21 #include <linux/hugetlb.h>
22 #include <linux/profile.h>
23 #include <linux/module.h>
24 #include <linux/acct.h>
25 #include <linux/mount.h>
26 #include <linux/mempolicy.h>
27 #include <linux/rmap.h>
28
29 #include <asm/uaccess.h>
30 #include <asm/cacheflush.h>
31 #include <asm/tlb.h>
32
33 /*
34 * WARNING: the debugging will use recursive algorithms so never enable this
35 * unless you know what you are doing.
36 */
37 #undef DEBUG_MM_RB
38
39 /* description of effects of mapping type and prot in current implementation.
40 * this is due to the limited x86 page protection hardware. The expected
41 * behavior is in parens:
42 *
43 * map_type prot
44 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
45 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
46 * w: (no) no w: (no) no w: (yes) yes w: (no) no
47 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
48 *
49 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
50 * w: (no) no w: (no) no w: (copy) copy w: (no) no
51 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
52 *
53 */
54 pgprot_t protection_map[16] = {
55 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
56 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
57 };
58
59 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
60 int sysctl_overcommit_ratio = 50; /* default is 50% */
61 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
62 atomic_t vm_committed_space = ATOMIC_INIT(0);
63
64 /*
65 * Check that a process has enough memory to allocate a new virtual
66 * mapping. 0 means there is enough memory for the allocation to
67 * succeed and -ENOMEM implies there is not.
68 *
69 * We currently support three overcommit policies, which are set via the
70 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
71 *
72 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
73 * Additional code 2002 Jul 20 by Robert Love.
74 *
75 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
76 *
77 * Note this is a helper function intended to be used by LSMs which
78 * wish to use this logic.
79 */
80 int __vm_enough_memory(long pages, int cap_sys_admin)
81 {
82 unsigned long free, allowed;
83
84 vm_acct_memory(pages);
85
86 /*
87 * Sometimes we want to use more memory than we have
88 */
89 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
90 return 0;
91
92 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
93 unsigned long n;
94
95 free = get_page_cache_size();
96 free += nr_swap_pages;
97
98 /*
99 * Any slabs which are created with the
100 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
101 * which are reclaimable, under pressure. The dentry
102 * cache and most inode caches should fall into this
103 */
104 free += atomic_read(&slab_reclaim_pages);
105
106 /*
107 * Leave the last 3% for root
108 */
109 if (!cap_sys_admin)
110 free -= free / 32;
111
112 if (free > pages)
113 return 0;
114
115 /*
116 * nr_free_pages() is very expensive on large systems,
117 * only call if we're about to fail.
118 */
119 n = nr_free_pages();
120 if (!cap_sys_admin)
121 n -= n / 32;
122 free += n;
123
124 if (free > pages)
125 return 0;
126 vm_unacct_memory(pages);
127 return -ENOMEM;
128 }
129
130 allowed = (totalram_pages - hugetlb_total_pages())
131 * sysctl_overcommit_ratio / 100;
132 /*
133 * Leave the last 3% for root
134 */
135 if (!cap_sys_admin)
136 allowed -= allowed / 32;
137 allowed += total_swap_pages;
138
139 /* Don't let a single process grow too big:
140 leave 3% of the size of this process for other processes */
141 allowed -= current->mm->total_vm / 32;
142
143 if (atomic_read(&vm_committed_space) < allowed)
144 return 0;
145
146 vm_unacct_memory(pages);
147
148 return -ENOMEM;
149 }
150
151 EXPORT_SYMBOL(sysctl_overcommit_memory);
152 EXPORT_SYMBOL(sysctl_overcommit_ratio);
153 EXPORT_SYMBOL(sysctl_max_map_count);
154 EXPORT_SYMBOL(vm_committed_space);
155 EXPORT_SYMBOL(__vm_enough_memory);
156
157 /*
158 * Requires inode->i_mapping->i_mmap_lock
159 */
160 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
161 struct file *file, struct address_space *mapping)
162 {
163 if (vma->vm_flags & VM_DENYWRITE)
164 atomic_inc(&file->f_dentry->d_inode->i_writecount);
165 if (vma->vm_flags & VM_SHARED)
166 mapping->i_mmap_writable--;
167
168 flush_dcache_mmap_lock(mapping);
169 if (unlikely(vma->vm_flags & VM_NONLINEAR))
170 list_del_init(&vma->shared.vm_set.list);
171 else
172 vma_prio_tree_remove(vma, &mapping->i_mmap);
173 flush_dcache_mmap_unlock(mapping);
174 }
175
176 /*
177 * Remove one vm structure and free it.
178 */
179 static void remove_vm_struct(struct vm_area_struct *vma)
180 {
181 struct file *file = vma->vm_file;
182
183 might_sleep();
184 if (file) {
185 struct address_space *mapping = file->f_mapping;
186 spin_lock(&mapping->i_mmap_lock);
187 __remove_shared_vm_struct(vma, file, mapping);
188 spin_unlock(&mapping->i_mmap_lock);
189 }
190 if (vma->vm_ops && vma->vm_ops->close)
191 vma->vm_ops->close(vma);
192 if (file)
193 fput(file);
194 anon_vma_unlink(vma);
195 mpol_free(vma_policy(vma));
196 kmem_cache_free(vm_area_cachep, vma);
197 }
198
199 /*
200 * sys_brk() for the most part doesn't need the global kernel
201 * lock, except when an application is doing something nasty
202 * like trying to un-brk an area that has already been mapped
203 * to a regular file. in this case, the unmapping will need
204 * to invoke file system routines that need the global lock.
205 */
206 asmlinkage unsigned long sys_brk(unsigned long brk)
207 {
208 unsigned long rlim, retval;
209 unsigned long newbrk, oldbrk;
210 struct mm_struct *mm = current->mm;
211
212 down_write(&mm->mmap_sem);
213
214 if (brk < mm->end_code)
215 goto out;
216 newbrk = PAGE_ALIGN(brk);
217 oldbrk = PAGE_ALIGN(mm->brk);
218 if (oldbrk == newbrk)
219 goto set_brk;
220
221 /* Always allow shrinking brk. */
222 if (brk <= mm->brk) {
223 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
224 goto set_brk;
225 goto out;
226 }
227
228 /* Check against rlimit.. */
229 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
230 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
231 goto out;
232
233 /* Check against existing mmap mappings. */
234 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
235 goto out;
236
237 /* Ok, looks good - let it rip. */
238 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
239 goto out;
240 set_brk:
241 mm->brk = brk;
242 out:
243 retval = mm->brk;
244 up_write(&mm->mmap_sem);
245 return retval;
246 }
247
248 #ifdef DEBUG_MM_RB
249 static int browse_rb(struct rb_root *root)
250 {
251 int i = 0, j;
252 struct rb_node *nd, *pn = NULL;
253 unsigned long prev = 0, pend = 0;
254
255 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
256 struct vm_area_struct *vma;
257 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
258 if (vma->vm_start < prev)
259 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
260 if (vma->vm_start < pend)
261 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
262 if (vma->vm_start > vma->vm_end)
263 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
264 i++;
265 pn = nd;
266 }
267 j = 0;
268 for (nd = pn; nd; nd = rb_prev(nd)) {
269 j++;
270 }
271 if (i != j)
272 printk("backwards %d, forwards %d\n", j, i), i = 0;
273 return i;
274 }
275
276 void validate_mm(struct mm_struct *mm)
277 {
278 int bug = 0;
279 int i = 0;
280 struct vm_area_struct *tmp = mm->mmap;
281 while (tmp) {
282 tmp = tmp->vm_next;
283 i++;
284 }
285 if (i != mm->map_count)
286 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
287 i = browse_rb(&mm->mm_rb);
288 if (i != mm->map_count)
289 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
290 if (bug)
291 BUG();
292 }
293 #else
294 #define validate_mm(mm) do { } while (0)
295 #endif
296
297 static struct vm_area_struct *
298 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
299 struct vm_area_struct **pprev, struct rb_node ***rb_link,
300 struct rb_node ** rb_parent)
301 {
302 struct vm_area_struct * vma;
303 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
304
305 __rb_link = &mm->mm_rb.rb_node;
306 rb_prev = __rb_parent = NULL;
307 vma = NULL;
308
309 while (*__rb_link) {
310 struct vm_area_struct *vma_tmp;
311
312 __rb_parent = *__rb_link;
313 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
314
315 if (vma_tmp->vm_end > addr) {
316 vma = vma_tmp;
317 if (vma_tmp->vm_start <= addr)
318 return vma;
319 __rb_link = &__rb_parent->rb_left;
320 } else {
321 rb_prev = __rb_parent;
322 __rb_link = &__rb_parent->rb_right;
323 }
324 }
325
326 *pprev = NULL;
327 if (rb_prev)
328 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
329 *rb_link = __rb_link;
330 *rb_parent = __rb_parent;
331 return vma;
332 }
333
334 static inline void
335 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
336 struct vm_area_struct *prev, struct rb_node *rb_parent)
337 {
338 if (prev) {
339 vma->vm_next = prev->vm_next;
340 prev->vm_next = vma;
341 } else {
342 mm->mmap = vma;
343 if (rb_parent)
344 vma->vm_next = rb_entry(rb_parent,
345 struct vm_area_struct, vm_rb);
346 else
347 vma->vm_next = NULL;
348 }
349 }
350
351 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
352 struct rb_node **rb_link, struct rb_node *rb_parent)
353 {
354 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
355 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
356 }
357
358 static inline void __vma_link_file(struct vm_area_struct *vma)
359 {
360 struct file * file;
361
362 file = vma->vm_file;
363 if (file) {
364 struct address_space *mapping = file->f_mapping;
365
366 if (vma->vm_flags & VM_DENYWRITE)
367 atomic_dec(&file->f_dentry->d_inode->i_writecount);
368 if (vma->vm_flags & VM_SHARED)
369 mapping->i_mmap_writable++;
370
371 flush_dcache_mmap_lock(mapping);
372 if (unlikely(vma->vm_flags & VM_NONLINEAR))
373 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
374 else
375 vma_prio_tree_insert(vma, &mapping->i_mmap);
376 flush_dcache_mmap_unlock(mapping);
377 }
378 }
379
380 static void
381 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
382 struct vm_area_struct *prev, struct rb_node **rb_link,
383 struct rb_node *rb_parent)
384 {
385 __vma_link_list(mm, vma, prev, rb_parent);
386 __vma_link_rb(mm, vma, rb_link, rb_parent);
387 __anon_vma_link(vma);
388 }
389
390 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
391 struct vm_area_struct *prev, struct rb_node **rb_link,
392 struct rb_node *rb_parent)
393 {
394 struct address_space *mapping = NULL;
395
396 if (vma->vm_file)
397 mapping = vma->vm_file->f_mapping;
398
399 if (mapping) {
400 spin_lock(&mapping->i_mmap_lock);
401 vma->vm_truncate_count = mapping->truncate_count;
402 }
403 anon_vma_lock(vma);
404
405 __vma_link(mm, vma, prev, rb_link, rb_parent);
406 __vma_link_file(vma);
407
408 anon_vma_unlock(vma);
409 if (mapping)
410 spin_unlock(&mapping->i_mmap_lock);
411
412 mm->map_count++;
413 validate_mm(mm);
414 }
415
416 /*
417 * Helper for vma_adjust in the split_vma insert case:
418 * insert vm structure into list and rbtree and anon_vma,
419 * but it has already been inserted into prio_tree earlier.
420 */
421 static void
422 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
423 {
424 struct vm_area_struct * __vma, * prev;
425 struct rb_node ** rb_link, * rb_parent;
426
427 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
428 if (__vma && __vma->vm_start < vma->vm_end)
429 BUG();
430 __vma_link(mm, vma, prev, rb_link, rb_parent);
431 mm->map_count++;
432 }
433
434 static inline void
435 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
436 struct vm_area_struct *prev)
437 {
438 prev->vm_next = vma->vm_next;
439 rb_erase(&vma->vm_rb, &mm->mm_rb);
440 if (mm->mmap_cache == vma)
441 mm->mmap_cache = prev;
442 }
443
444 /*
445 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
446 * is already present in an i_mmap tree without adjusting the tree.
447 * The following helper function should be used when such adjustments
448 * are necessary. The "insert" vma (if any) is to be inserted
449 * before we drop the necessary locks.
450 */
451 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
452 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
453 {
454 struct mm_struct *mm = vma->vm_mm;
455 struct vm_area_struct *next = vma->vm_next;
456 struct vm_area_struct *importer = NULL;
457 struct address_space *mapping = NULL;
458 struct prio_tree_root *root = NULL;
459 struct file *file = vma->vm_file;
460 struct anon_vma *anon_vma = NULL;
461 long adjust_next = 0;
462 int remove_next = 0;
463
464 if (next && !insert) {
465 if (end >= next->vm_end) {
466 /*
467 * vma expands, overlapping all the next, and
468 * perhaps the one after too (mprotect case 6).
469 */
470 again: remove_next = 1 + (end > next->vm_end);
471 end = next->vm_end;
472 anon_vma = next->anon_vma;
473 importer = vma;
474 } else if (end > next->vm_start) {
475 /*
476 * vma expands, overlapping part of the next:
477 * mprotect case 5 shifting the boundary up.
478 */
479 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
480 anon_vma = next->anon_vma;
481 importer = vma;
482 } else if (end < vma->vm_end) {
483 /*
484 * vma shrinks, and !insert tells it's not
485 * split_vma inserting another: so it must be
486 * mprotect case 4 shifting the boundary down.
487 */
488 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
489 anon_vma = next->anon_vma;
490 importer = next;
491 }
492 }
493
494 if (file) {
495 mapping = file->f_mapping;
496 if (!(vma->vm_flags & VM_NONLINEAR))
497 root = &mapping->i_mmap;
498 spin_lock(&mapping->i_mmap_lock);
499 if (importer &&
500 vma->vm_truncate_count != next->vm_truncate_count) {
501 /*
502 * unmap_mapping_range might be in progress:
503 * ensure that the expanding vma is rescanned.
504 */
505 importer->vm_truncate_count = 0;
506 }
507 if (insert) {
508 insert->vm_truncate_count = vma->vm_truncate_count;
509 /*
510 * Put into prio_tree now, so instantiated pages
511 * are visible to arm/parisc __flush_dcache_page
512 * throughout; but we cannot insert into address
513 * space until vma start or end is updated.
514 */
515 __vma_link_file(insert);
516 }
517 }
518
519 /*
520 * When changing only vma->vm_end, we don't really need
521 * anon_vma lock: but is that case worth optimizing out?
522 */
523 if (vma->anon_vma)
524 anon_vma = vma->anon_vma;
525 if (anon_vma) {
526 spin_lock(&anon_vma->lock);
527 /*
528 * Easily overlooked: when mprotect shifts the boundary,
529 * make sure the expanding vma has anon_vma set if the
530 * shrinking vma had, to cover any anon pages imported.
531 */
532 if (importer && !importer->anon_vma) {
533 importer->anon_vma = anon_vma;
534 __anon_vma_link(importer);
535 }
536 }
537
538 if (root) {
539 flush_dcache_mmap_lock(mapping);
540 vma_prio_tree_remove(vma, root);
541 if (adjust_next)
542 vma_prio_tree_remove(next, root);
543 }
544
545 vma->vm_start = start;
546 vma->vm_end = end;
547 vma->vm_pgoff = pgoff;
548 if (adjust_next) {
549 next->vm_start += adjust_next << PAGE_SHIFT;
550 next->vm_pgoff += adjust_next;
551 }
552
553 if (root) {
554 if (adjust_next)
555 vma_prio_tree_insert(next, root);
556 vma_prio_tree_insert(vma, root);
557 flush_dcache_mmap_unlock(mapping);
558 }
559
560 if (remove_next) {
561 /*
562 * vma_merge has merged next into vma, and needs
563 * us to remove next before dropping the locks.
564 */
565 __vma_unlink(mm, next, vma);
566 if (file)
567 __remove_shared_vm_struct(next, file, mapping);
568 if (next->anon_vma)
569 __anon_vma_merge(vma, next);
570 } else if (insert) {
571 /*
572 * split_vma has split insert from vma, and needs
573 * us to insert it before dropping the locks
574 * (it may either follow vma or precede it).
575 */
576 __insert_vm_struct(mm, insert);
577 }
578
579 if (anon_vma)
580 spin_unlock(&anon_vma->lock);
581 if (mapping)
582 spin_unlock(&mapping->i_mmap_lock);
583
584 if (remove_next) {
585 if (file)
586 fput(file);
587 mm->map_count--;
588 mpol_free(vma_policy(next));
589 kmem_cache_free(vm_area_cachep, next);
590 /*
591 * In mprotect's case 6 (see comments on vma_merge),
592 * we must remove another next too. It would clutter
593 * up the code too much to do both in one go.
594 */
595 if (remove_next == 2) {
596 next = vma->vm_next;
597 goto again;
598 }
599 }
600
601 validate_mm(mm);
602 }
603
604 /*
605 * If the vma has a ->close operation then the driver probably needs to release
606 * per-vma resources, so we don't attempt to merge those.
607 */
608 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
609
610 static inline int is_mergeable_vma(struct vm_area_struct *vma,
611 struct file *file, unsigned long vm_flags)
612 {
613 if (vma->vm_flags != vm_flags)
614 return 0;
615 if (vma->vm_file != file)
616 return 0;
617 if (vma->vm_ops && vma->vm_ops->close)
618 return 0;
619 return 1;
620 }
621
622 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
623 struct anon_vma *anon_vma2)
624 {
625 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
626 }
627
628 /*
629 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
630 * in front of (at a lower virtual address and file offset than) the vma.
631 *
632 * We cannot merge two vmas if they have differently assigned (non-NULL)
633 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
634 *
635 * We don't check here for the merged mmap wrapping around the end of pagecache
636 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
637 * wrap, nor mmaps which cover the final page at index -1UL.
638 */
639 static int
640 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
641 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
642 {
643 if (is_mergeable_vma(vma, file, vm_flags) &&
644 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
645 if (vma->vm_pgoff == vm_pgoff)
646 return 1;
647 }
648 return 0;
649 }
650
651 /*
652 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
653 * beyond (at a higher virtual address and file offset than) the vma.
654 *
655 * We cannot merge two vmas if they have differently assigned (non-NULL)
656 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
657 */
658 static int
659 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
660 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
661 {
662 if (is_mergeable_vma(vma, file, vm_flags) &&
663 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
664 pgoff_t vm_pglen;
665 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
666 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
667 return 1;
668 }
669 return 0;
670 }
671
672 /*
673 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
674 * whether that can be merged with its predecessor or its successor.
675 * Or both (it neatly fills a hole).
676 *
677 * In most cases - when called for mmap, brk or mremap - [addr,end) is
678 * certain not to be mapped by the time vma_merge is called; but when
679 * called for mprotect, it is certain to be already mapped (either at
680 * an offset within prev, or at the start of next), and the flags of
681 * this area are about to be changed to vm_flags - and the no-change
682 * case has already been eliminated.
683 *
684 * The following mprotect cases have to be considered, where AAAA is
685 * the area passed down from mprotect_fixup, never extending beyond one
686 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
687 *
688 * AAAA AAAA AAAA AAAA
689 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
690 * cannot merge might become might become might become
691 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
692 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
693 * mremap move: PPPPNNNNNNNN 8
694 * AAAA
695 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
696 * might become case 1 below case 2 below case 3 below
697 *
698 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
699 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
700 */
701 struct vm_area_struct *vma_merge(struct mm_struct *mm,
702 struct vm_area_struct *prev, unsigned long addr,
703 unsigned long end, unsigned long vm_flags,
704 struct anon_vma *anon_vma, struct file *file,
705 pgoff_t pgoff, struct mempolicy *policy)
706 {
707 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
708 struct vm_area_struct *area, *next;
709
710 /*
711 * We later require that vma->vm_flags == vm_flags,
712 * so this tests vma->vm_flags & VM_SPECIAL, too.
713 */
714 if (vm_flags & VM_SPECIAL)
715 return NULL;
716
717 if (prev)
718 next = prev->vm_next;
719 else
720 next = mm->mmap;
721 area = next;
722 if (next && next->vm_end == end) /* cases 6, 7, 8 */
723 next = next->vm_next;
724
725 /*
726 * Can it merge with the predecessor?
727 */
728 if (prev && prev->vm_end == addr &&
729 mpol_equal(vma_policy(prev), policy) &&
730 can_vma_merge_after(prev, vm_flags,
731 anon_vma, file, pgoff)) {
732 /*
733 * OK, it can. Can we now merge in the successor as well?
734 */
735 if (next && end == next->vm_start &&
736 mpol_equal(policy, vma_policy(next)) &&
737 can_vma_merge_before(next, vm_flags,
738 anon_vma, file, pgoff+pglen) &&
739 is_mergeable_anon_vma(prev->anon_vma,
740 next->anon_vma)) {
741 /* cases 1, 6 */
742 vma_adjust(prev, prev->vm_start,
743 next->vm_end, prev->vm_pgoff, NULL);
744 } else /* cases 2, 5, 7 */
745 vma_adjust(prev, prev->vm_start,
746 end, prev->vm_pgoff, NULL);
747 return prev;
748 }
749
750 /*
751 * Can this new request be merged in front of next?
752 */
753 if (next && end == next->vm_start &&
754 mpol_equal(policy, vma_policy(next)) &&
755 can_vma_merge_before(next, vm_flags,
756 anon_vma, file, pgoff+pglen)) {
757 if (prev && addr < prev->vm_end) /* case 4 */
758 vma_adjust(prev, prev->vm_start,
759 addr, prev->vm_pgoff, NULL);
760 else /* cases 3, 8 */
761 vma_adjust(area, addr, next->vm_end,
762 next->vm_pgoff - pglen, NULL);
763 return area;
764 }
765
766 return NULL;
767 }
768
769 /*
770 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
771 * neighbouring vmas for a suitable anon_vma, before it goes off
772 * to allocate a new anon_vma. It checks because a repetitive
773 * sequence of mprotects and faults may otherwise lead to distinct
774 * anon_vmas being allocated, preventing vma merge in subsequent
775 * mprotect.
776 */
777 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
778 {
779 struct vm_area_struct *near;
780 unsigned long vm_flags;
781
782 near = vma->vm_next;
783 if (!near)
784 goto try_prev;
785
786 /*
787 * Since only mprotect tries to remerge vmas, match flags
788 * which might be mprotected into each other later on.
789 * Neither mlock nor madvise tries to remerge at present,
790 * so leave their flags as obstructing a merge.
791 */
792 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
793 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
794
795 if (near->anon_vma && vma->vm_end == near->vm_start &&
796 mpol_equal(vma_policy(vma), vma_policy(near)) &&
797 can_vma_merge_before(near, vm_flags,
798 NULL, vma->vm_file, vma->vm_pgoff +
799 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
800 return near->anon_vma;
801 try_prev:
802 /*
803 * It is potentially slow to have to call find_vma_prev here.
804 * But it's only on the first write fault on the vma, not
805 * every time, and we could devise a way to avoid it later
806 * (e.g. stash info in next's anon_vma_node when assigning
807 * an anon_vma, or when trying vma_merge). Another time.
808 */
809 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
810 BUG();
811 if (!near)
812 goto none;
813
814 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
815 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
816
817 if (near->anon_vma && near->vm_end == vma->vm_start &&
818 mpol_equal(vma_policy(near), vma_policy(vma)) &&
819 can_vma_merge_after(near, vm_flags,
820 NULL, vma->vm_file, vma->vm_pgoff))
821 return near->anon_vma;
822 none:
823 /*
824 * There's no absolute need to look only at touching neighbours:
825 * we could search further afield for "compatible" anon_vmas.
826 * But it would probably just be a waste of time searching,
827 * or lead to too many vmas hanging off the same anon_vma.
828 * We're trying to allow mprotect remerging later on,
829 * not trying to minimize memory used for anon_vmas.
830 */
831 return NULL;
832 }
833
834 #ifdef CONFIG_PROC_FS
835 void __vm_stat_account(struct mm_struct *mm, unsigned long flags,
836 struct file *file, long pages)
837 {
838 const unsigned long stack_flags
839 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
840
841 #ifdef CONFIG_HUGETLB
842 if (flags & VM_HUGETLB) {
843 if (!(flags & VM_DONTCOPY))
844 mm->shared_vm += pages;
845 return;
846 }
847 #endif /* CONFIG_HUGETLB */
848
849 if (file) {
850 mm->shared_vm += pages;
851 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
852 mm->exec_vm += pages;
853 } else if (flags & stack_flags)
854 mm->stack_vm += pages;
855 if (flags & (VM_RESERVED|VM_IO))
856 mm->reserved_vm += pages;
857 }
858 #endif /* CONFIG_PROC_FS */
859
860 /*
861 * The caller must hold down_write(current->mm->mmap_sem).
862 */
863
864 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
865 unsigned long len, unsigned long prot,
866 unsigned long flags, unsigned long pgoff)
867 {
868 struct mm_struct * mm = current->mm;
869 struct vm_area_struct * vma, * prev;
870 struct inode *inode;
871 unsigned int vm_flags;
872 int correct_wcount = 0;
873 int error;
874 struct rb_node ** rb_link, * rb_parent;
875 int accountable = 1;
876 unsigned long charged = 0;
877
878 if (file) {
879 if (is_file_hugepages(file))
880 accountable = 0;
881
882 if (!file->f_op || !file->f_op->mmap)
883 return -ENODEV;
884
885 if ((prot & PROT_EXEC) &&
886 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
887 return -EPERM;
888 }
889 /*
890 * Does the application expect PROT_READ to imply PROT_EXEC?
891 *
892 * (the exception is when the underlying filesystem is noexec
893 * mounted, in which case we dont add PROT_EXEC.)
894 */
895 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
896 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
897 prot |= PROT_EXEC;
898
899 if (!len)
900 return addr;
901
902 /* Careful about overflows.. */
903 len = PAGE_ALIGN(len);
904 if (!len || len > TASK_SIZE)
905 return -EINVAL;
906
907 /* offset overflow? */
908 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
909 return -EINVAL;
910
911 /* Too many mappings? */
912 if (mm->map_count > sysctl_max_map_count)
913 return -ENOMEM;
914
915 /* Obtain the address to map to. we verify (or select) it and ensure
916 * that it represents a valid section of the address space.
917 */
918 addr = get_unmapped_area(file, addr, len, pgoff, flags);
919 if (addr & ~PAGE_MASK)
920 return addr;
921
922 /* Do simple checking here so the lower-level routines won't have
923 * to. we assume access permissions have been handled by the open
924 * of the memory object, so we don't do any here.
925 */
926 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
927 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
928
929 if (flags & MAP_LOCKED) {
930 if (!can_do_mlock())
931 return -EPERM;
932 vm_flags |= VM_LOCKED;
933 }
934 /* mlock MCL_FUTURE? */
935 if (vm_flags & VM_LOCKED) {
936 unsigned long locked, lock_limit;
937 locked = mm->locked_vm << PAGE_SHIFT;
938 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
939 locked += len;
940 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
941 return -EAGAIN;
942 }
943
944 inode = file ? file->f_dentry->d_inode : NULL;
945
946 if (file) {
947 switch (flags & MAP_TYPE) {
948 case MAP_SHARED:
949 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
950 return -EACCES;
951
952 /*
953 * Make sure we don't allow writing to an append-only
954 * file..
955 */
956 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
957 return -EACCES;
958
959 /*
960 * Make sure there are no mandatory locks on the file.
961 */
962 if (locks_verify_locked(inode))
963 return -EAGAIN;
964
965 vm_flags |= VM_SHARED | VM_MAYSHARE;
966 if (!(file->f_mode & FMODE_WRITE))
967 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
968
969 /* fall through */
970 case MAP_PRIVATE:
971 if (!(file->f_mode & FMODE_READ))
972 return -EACCES;
973 break;
974
975 default:
976 return -EINVAL;
977 }
978 } else {
979 switch (flags & MAP_TYPE) {
980 case MAP_SHARED:
981 vm_flags |= VM_SHARED | VM_MAYSHARE;
982 break;
983 case MAP_PRIVATE:
984 /*
985 * Set pgoff according to addr for anon_vma.
986 */
987 pgoff = addr >> PAGE_SHIFT;
988 break;
989 default:
990 return -EINVAL;
991 }
992 }
993
994 error = security_file_mmap(file, prot, flags);
995 if (error)
996 return error;
997
998 /* Clear old maps */
999 error = -ENOMEM;
1000 munmap_back:
1001 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1002 if (vma && vma->vm_start < addr + len) {
1003 if (do_munmap(mm, addr, len))
1004 return -ENOMEM;
1005 goto munmap_back;
1006 }
1007
1008 /* Check against address space limit. */
1009 if ((mm->total_vm << PAGE_SHIFT) + len
1010 > current->signal->rlim[RLIMIT_AS].rlim_cur)
1011 return -ENOMEM;
1012
1013 if (accountable && (!(flags & MAP_NORESERVE) ||
1014 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1015 if (vm_flags & VM_SHARED) {
1016 /* Check memory availability in shmem_file_setup? */
1017 vm_flags |= VM_ACCOUNT;
1018 } else if (vm_flags & VM_WRITE) {
1019 /*
1020 * Private writable mapping: check memory availability
1021 */
1022 charged = len >> PAGE_SHIFT;
1023 if (security_vm_enough_memory(charged))
1024 return -ENOMEM;
1025 vm_flags |= VM_ACCOUNT;
1026 }
1027 }
1028
1029 /*
1030 * Can we just expand an old private anonymous mapping?
1031 * The VM_SHARED test is necessary because shmem_zero_setup
1032 * will create the file object for a shared anonymous map below.
1033 */
1034 if (!file && !(vm_flags & VM_SHARED) &&
1035 vma_merge(mm, prev, addr, addr + len, vm_flags,
1036 NULL, NULL, pgoff, NULL))
1037 goto out;
1038
1039 /*
1040 * Determine the object being mapped and call the appropriate
1041 * specific mapper. the address has already been validated, but
1042 * not unmapped, but the maps are removed from the list.
1043 */
1044 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1045 if (!vma) {
1046 error = -ENOMEM;
1047 goto unacct_error;
1048 }
1049 memset(vma, 0, sizeof(*vma));
1050
1051 vma->vm_mm = mm;
1052 vma->vm_start = addr;
1053 vma->vm_end = addr + len;
1054 vma->vm_flags = vm_flags;
1055 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1056 vma->vm_pgoff = pgoff;
1057
1058 if (file) {
1059 error = -EINVAL;
1060 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1061 goto free_vma;
1062 if (vm_flags & VM_DENYWRITE) {
1063 error = deny_write_access(file);
1064 if (error)
1065 goto free_vma;
1066 correct_wcount = 1;
1067 }
1068 vma->vm_file = file;
1069 get_file(file);
1070 error = file->f_op->mmap(file, vma);
1071 if (error)
1072 goto unmap_and_free_vma;
1073 } else if (vm_flags & VM_SHARED) {
1074 error = shmem_zero_setup(vma);
1075 if (error)
1076 goto free_vma;
1077 }
1078
1079 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1080 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1081 * that memory reservation must be checked; but that reservation
1082 * belongs to shared memory object, not to vma: so now clear it.
1083 */
1084 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1085 vma->vm_flags &= ~VM_ACCOUNT;
1086
1087 /* Can addr have changed??
1088 *
1089 * Answer: Yes, several device drivers can do it in their
1090 * f_op->mmap method. -DaveM
1091 */
1092 addr = vma->vm_start;
1093 pgoff = vma->vm_pgoff;
1094 vm_flags = vma->vm_flags;
1095
1096 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1097 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1098 file = vma->vm_file;
1099 vma_link(mm, vma, prev, rb_link, rb_parent);
1100 if (correct_wcount)
1101 atomic_inc(&inode->i_writecount);
1102 } else {
1103 if (file) {
1104 if (correct_wcount)
1105 atomic_inc(&inode->i_writecount);
1106 fput(file);
1107 }
1108 mpol_free(vma_policy(vma));
1109 kmem_cache_free(vm_area_cachep, vma);
1110 }
1111 out:
1112 mm->total_vm += len >> PAGE_SHIFT;
1113 __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1114 if (vm_flags & VM_LOCKED) {
1115 mm->locked_vm += len >> PAGE_SHIFT;
1116 make_pages_present(addr, addr + len);
1117 }
1118 if (flags & MAP_POPULATE) {
1119 up_write(&mm->mmap_sem);
1120 sys_remap_file_pages(addr, len, 0,
1121 pgoff, flags & MAP_NONBLOCK);
1122 down_write(&mm->mmap_sem);
1123 }
1124 acct_update_integrals();
1125 update_mem_hiwater();
1126 return addr;
1127
1128 unmap_and_free_vma:
1129 if (correct_wcount)
1130 atomic_inc(&inode->i_writecount);
1131 vma->vm_file = NULL;
1132 fput(file);
1133
1134 /* Undo any partial mapping done by a device driver. */
1135 zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1136 free_vma:
1137 kmem_cache_free(vm_area_cachep, vma);
1138 unacct_error:
1139 if (charged)
1140 vm_unacct_memory(charged);
1141 return error;
1142 }
1143
1144 EXPORT_SYMBOL(do_mmap_pgoff);
1145
1146 /* Get an address range which is currently unmapped.
1147 * For shmat() with addr=0.
1148 *
1149 * Ugly calling convention alert:
1150 * Return value with the low bits set means error value,
1151 * ie
1152 * if (ret & ~PAGE_MASK)
1153 * error = ret;
1154 *
1155 * This function "knows" that -ENOMEM has the bits set.
1156 */
1157 #ifndef HAVE_ARCH_UNMAPPED_AREA
1158 unsigned long
1159 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1160 unsigned long len, unsigned long pgoff, unsigned long flags)
1161 {
1162 struct mm_struct *mm = current->mm;
1163 struct vm_area_struct *vma;
1164 unsigned long start_addr;
1165
1166 if (len > TASK_SIZE)
1167 return -ENOMEM;
1168
1169 if (addr) {
1170 addr = PAGE_ALIGN(addr);
1171 vma = find_vma(mm, addr);
1172 if (TASK_SIZE - len >= addr &&
1173 (!vma || addr + len <= vma->vm_start))
1174 return addr;
1175 }
1176 start_addr = addr = mm->free_area_cache;
1177
1178 full_search:
1179 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1180 /* At this point: (!vma || addr < vma->vm_end). */
1181 if (TASK_SIZE - len < addr) {
1182 /*
1183 * Start a new search - just in case we missed
1184 * some holes.
1185 */
1186 if (start_addr != TASK_UNMAPPED_BASE) {
1187 start_addr = addr = TASK_UNMAPPED_BASE;
1188 goto full_search;
1189 }
1190 return -ENOMEM;
1191 }
1192 if (!vma || addr + len <= vma->vm_start) {
1193 /*
1194 * Remember the place where we stopped the search:
1195 */
1196 mm->free_area_cache = addr + len;
1197 return addr;
1198 }
1199 addr = vma->vm_end;
1200 }
1201 }
1202 #endif
1203
1204 void arch_unmap_area(struct vm_area_struct *area)
1205 {
1206 /*
1207 * Is this a new hole at the lowest possible address?
1208 */
1209 if (area->vm_start >= TASK_UNMAPPED_BASE &&
1210 area->vm_start < area->vm_mm->free_area_cache)
1211 area->vm_mm->free_area_cache = area->vm_start;
1212 }
1213
1214 /*
1215 * This mmap-allocator allocates new areas top-down from below the
1216 * stack's low limit (the base):
1217 */
1218 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1219 unsigned long
1220 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1221 const unsigned long len, const unsigned long pgoff,
1222 const unsigned long flags)
1223 {
1224 struct vm_area_struct *vma, *prev_vma;
1225 struct mm_struct *mm = current->mm;
1226 unsigned long base = mm->mmap_base, addr = addr0;
1227 int first_time = 1;
1228
1229 /* requested length too big for entire address space */
1230 if (len > TASK_SIZE)
1231 return -ENOMEM;
1232
1233 /* dont allow allocations above current base */
1234 if (mm->free_area_cache > base)
1235 mm->free_area_cache = base;
1236
1237 /* requesting a specific address */
1238 if (addr) {
1239 addr = PAGE_ALIGN(addr);
1240 vma = find_vma(mm, addr);
1241 if (TASK_SIZE - len >= addr &&
1242 (!vma || addr + len <= vma->vm_start))
1243 return addr;
1244 }
1245
1246 try_again:
1247 /* make sure it can fit in the remaining address space */
1248 if (mm->free_area_cache < len)
1249 goto fail;
1250
1251 /* either no address requested or cant fit in requested address hole */
1252 addr = (mm->free_area_cache - len) & PAGE_MASK;
1253 do {
1254 /*
1255 * Lookup failure means no vma is above this address,
1256 * i.e. return with success:
1257 */
1258 if (!(vma = find_vma_prev(mm, addr, &prev_vma)))
1259 return addr;
1260
1261 /*
1262 * new region fits between prev_vma->vm_end and
1263 * vma->vm_start, use it:
1264 */
1265 if (addr+len <= vma->vm_start &&
1266 (!prev_vma || (addr >= prev_vma->vm_end)))
1267 /* remember the address as a hint for next time */
1268 return (mm->free_area_cache = addr);
1269 else
1270 /* pull free_area_cache down to the first hole */
1271 if (mm->free_area_cache == vma->vm_end)
1272 mm->free_area_cache = vma->vm_start;
1273
1274 /* try just below the current vma->vm_start */
1275 addr = vma->vm_start-len;
1276 } while (len <= vma->vm_start);
1277
1278 fail:
1279 /*
1280 * if hint left us with no space for the requested
1281 * mapping then try again:
1282 */
1283 if (first_time) {
1284 mm->free_area_cache = base;
1285 first_time = 0;
1286 goto try_again;
1287 }
1288 /*
1289 * A failed mmap() very likely causes application failure,
1290 * so fall back to the bottom-up function here. This scenario
1291 * can happen with large stack limits and large mmap()
1292 * allocations.
1293 */
1294 mm->free_area_cache = TASK_UNMAPPED_BASE;
1295 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1296 /*
1297 * Restore the topdown base:
1298 */
1299 mm->free_area_cache = base;
1300
1301 return addr;
1302 }
1303 #endif
1304
1305 void arch_unmap_area_topdown(struct vm_area_struct *area)
1306 {
1307 /*
1308 * Is this a new hole at the highest possible address?
1309 */
1310 if (area->vm_end > area->vm_mm->free_area_cache)
1311 area->vm_mm->free_area_cache = area->vm_end;
1312 }
1313
1314 unsigned long
1315 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1316 unsigned long pgoff, unsigned long flags)
1317 {
1318 if (flags & MAP_FIXED) {
1319 unsigned long ret;
1320
1321 if (addr > TASK_SIZE - len)
1322 return -ENOMEM;
1323 if (addr & ~PAGE_MASK)
1324 return -EINVAL;
1325 if (file && is_file_hugepages(file)) {
1326 /*
1327 * Check if the given range is hugepage aligned, and
1328 * can be made suitable for hugepages.
1329 */
1330 ret = prepare_hugepage_range(addr, len);
1331 } else {
1332 /*
1333 * Ensure that a normal request is not falling in a
1334 * reserved hugepage range. For some archs like IA-64,
1335 * there is a separate region for hugepages.
1336 */
1337 ret = is_hugepage_only_range(addr, len);
1338 }
1339 if (ret)
1340 return -EINVAL;
1341 return addr;
1342 }
1343
1344 if (file && file->f_op && file->f_op->get_unmapped_area)
1345 return file->f_op->get_unmapped_area(file, addr, len,
1346 pgoff, flags);
1347
1348 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
1349 }
1350
1351 EXPORT_SYMBOL(get_unmapped_area);
1352
1353 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1354 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1355 {
1356 struct vm_area_struct *vma = NULL;
1357
1358 if (mm) {
1359 /* Check the cache first. */
1360 /* (Cache hit rate is typically around 35%.) */
1361 vma = mm->mmap_cache;
1362 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1363 struct rb_node * rb_node;
1364
1365 rb_node = mm->mm_rb.rb_node;
1366 vma = NULL;
1367
1368 while (rb_node) {
1369 struct vm_area_struct * vma_tmp;
1370
1371 vma_tmp = rb_entry(rb_node,
1372 struct vm_area_struct, vm_rb);
1373
1374 if (vma_tmp->vm_end > addr) {
1375 vma = vma_tmp;
1376 if (vma_tmp->vm_start <= addr)
1377 break;
1378 rb_node = rb_node->rb_left;
1379 } else
1380 rb_node = rb_node->rb_right;
1381 }
1382 if (vma)
1383 mm->mmap_cache = vma;
1384 }
1385 }
1386 return vma;
1387 }
1388
1389 EXPORT_SYMBOL(find_vma);
1390
1391 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1392 struct vm_area_struct *
1393 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1394 struct vm_area_struct **pprev)
1395 {
1396 struct vm_area_struct *vma = NULL, *prev = NULL;
1397 struct rb_node * rb_node;
1398 if (!mm)
1399 goto out;
1400
1401 /* Guard against addr being lower than the first VMA */
1402 vma = mm->mmap;
1403
1404 /* Go through the RB tree quickly. */
1405 rb_node = mm->mm_rb.rb_node;
1406
1407 while (rb_node) {
1408 struct vm_area_struct *vma_tmp;
1409 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1410
1411 if (addr < vma_tmp->vm_end) {
1412 rb_node = rb_node->rb_left;
1413 } else {
1414 prev = vma_tmp;
1415 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1416 break;
1417 rb_node = rb_node->rb_right;
1418 }
1419 }
1420
1421 out:
1422 *pprev = prev;
1423 return prev ? prev->vm_next : vma;
1424 }
1425
1426 /*
1427 * Verify that the stack growth is acceptable and
1428 * update accounting. This is shared with both the
1429 * grow-up and grow-down cases.
1430 */
1431 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1432 {
1433 struct mm_struct *mm = vma->vm_mm;
1434 struct rlimit *rlim = current->signal->rlim;
1435
1436 /* address space limit tests */
1437 if (mm->total_vm + grow > rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT)
1438 return -ENOMEM;
1439
1440 /* Stack limit test */
1441 if (size > rlim[RLIMIT_STACK].rlim_cur)
1442 return -ENOMEM;
1443
1444 /* mlock limit tests */
1445 if (vma->vm_flags & VM_LOCKED) {
1446 unsigned long locked;
1447 unsigned long limit;
1448 locked = mm->locked_vm + grow;
1449 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1450 if (locked > limit && !capable(CAP_IPC_LOCK))
1451 return -ENOMEM;
1452 }
1453
1454 /*
1455 * Overcommit.. This must be the final test, as it will
1456 * update security statistics.
1457 */
1458 if (security_vm_enough_memory(grow))
1459 return -ENOMEM;
1460
1461 /* Ok, everything looks good - let it rip */
1462 mm->total_vm += grow;
1463 if (vma->vm_flags & VM_LOCKED)
1464 mm->locked_vm += grow;
1465 __vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1466 acct_update_integrals();
1467 update_mem_hiwater();
1468 return 0;
1469 }
1470
1471 #ifdef CONFIG_STACK_GROWSUP
1472 /*
1473 * vma is the first one with address > vma->vm_end. Have to extend vma.
1474 */
1475 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1476 {
1477 int error;
1478
1479 if (!(vma->vm_flags & VM_GROWSUP))
1480 return -EFAULT;
1481
1482 /*
1483 * We must make sure the anon_vma is allocated
1484 * so that the anon_vma locking is not a noop.
1485 */
1486 if (unlikely(anon_vma_prepare(vma)))
1487 return -ENOMEM;
1488 anon_vma_lock(vma);
1489
1490 /*
1491 * vma->vm_start/vm_end cannot change under us because the caller
1492 * is required to hold the mmap_sem in read mode. We need the
1493 * anon_vma lock to serialize against concurrent expand_stacks.
1494 */
1495 address += 4 + PAGE_SIZE - 1;
1496 address &= PAGE_MASK;
1497 error = 0;
1498
1499 /* Somebody else might have raced and expanded it already */
1500 if (address > vma->vm_end) {
1501 unsigned long size, grow;
1502
1503 size = address - vma->vm_start;
1504 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1505
1506 error = acct_stack_growth(vma, size, grow);
1507 if (!error)
1508 vma->vm_end = address;
1509 }
1510 anon_vma_unlock(vma);
1511 return error;
1512 }
1513
1514 struct vm_area_struct *
1515 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1516 {
1517 struct vm_area_struct *vma, *prev;
1518
1519 addr &= PAGE_MASK;
1520 vma = find_vma_prev(mm, addr, &prev);
1521 if (vma && (vma->vm_start <= addr))
1522 return vma;
1523 if (!prev || expand_stack(prev, addr))
1524 return NULL;
1525 if (prev->vm_flags & VM_LOCKED) {
1526 make_pages_present(addr, prev->vm_end);
1527 }
1528 return prev;
1529 }
1530 #else
1531 /*
1532 * vma is the first one with address < vma->vm_start. Have to extend vma.
1533 */
1534 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1535 {
1536 int error;
1537
1538 /*
1539 * We must make sure the anon_vma is allocated
1540 * so that the anon_vma locking is not a noop.
1541 */
1542 if (unlikely(anon_vma_prepare(vma)))
1543 return -ENOMEM;
1544 anon_vma_lock(vma);
1545
1546 /*
1547 * vma->vm_start/vm_end cannot change under us because the caller
1548 * is required to hold the mmap_sem in read mode. We need the
1549 * anon_vma lock to serialize against concurrent expand_stacks.
1550 */
1551 address &= PAGE_MASK;
1552 error = 0;
1553
1554 /* Somebody else might have raced and expanded it already */
1555 if (address < vma->vm_start) {
1556 unsigned long size, grow;
1557
1558 size = vma->vm_end - address;
1559 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1560
1561 error = acct_stack_growth(vma, size, grow);
1562 if (!error) {
1563 vma->vm_start = address;
1564 vma->vm_pgoff -= grow;
1565 }
1566 }
1567 anon_vma_unlock(vma);
1568 return error;
1569 }
1570
1571 struct vm_area_struct *
1572 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1573 {
1574 struct vm_area_struct * vma;
1575 unsigned long start;
1576
1577 addr &= PAGE_MASK;
1578 vma = find_vma(mm,addr);
1579 if (!vma)
1580 return NULL;
1581 if (vma->vm_start <= addr)
1582 return vma;
1583 if (!(vma->vm_flags & VM_GROWSDOWN))
1584 return NULL;
1585 start = vma->vm_start;
1586 if (expand_stack(vma, addr))
1587 return NULL;
1588 if (vma->vm_flags & VM_LOCKED) {
1589 make_pages_present(addr, start);
1590 }
1591 return vma;
1592 }
1593 #endif
1594
1595 /*
1596 * Try to free as many page directory entries as we can,
1597 * without having to work very hard at actually scanning
1598 * the page tables themselves.
1599 *
1600 * Right now we try to free page tables if we have a nice
1601 * PGDIR-aligned area that got free'd up. We could be more
1602 * granular if we want to, but this is fast and simple,
1603 * and covers the bad cases.
1604 *
1605 * "prev", if it exists, points to a vma before the one
1606 * we just free'd - but there's no telling how much before.
1607 */
1608 static void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *prev,
1609 unsigned long start, unsigned long end)
1610 {
1611 unsigned long first = start & PGDIR_MASK;
1612 unsigned long last = end + PGDIR_SIZE - 1;
1613 struct mm_struct *mm = tlb->mm;
1614
1615 if (last > MM_VM_SIZE(mm) || last < end)
1616 last = MM_VM_SIZE(mm);
1617
1618 if (!prev) {
1619 prev = mm->mmap;
1620 if (!prev)
1621 goto no_mmaps;
1622 if (prev->vm_end > start) {
1623 if (last > prev->vm_start)
1624 last = prev->vm_start;
1625 goto no_mmaps;
1626 }
1627 }
1628 for (;;) {
1629 struct vm_area_struct *next = prev->vm_next;
1630
1631 if (next) {
1632 if (next->vm_start < start) {
1633 prev = next;
1634 continue;
1635 }
1636 if (last > next->vm_start)
1637 last = next->vm_start;
1638 }
1639 if (prev->vm_end > first)
1640 first = prev->vm_end;
1641 break;
1642 }
1643 no_mmaps:
1644 if (last < first) /* for arches with discontiguous pgd indices */
1645 return;
1646 if (first < FIRST_USER_PGD_NR * PGDIR_SIZE)
1647 first = FIRST_USER_PGD_NR * PGDIR_SIZE;
1648 /* No point trying to free anything if we're in the same pte page */
1649 if ((first & PMD_MASK) < (last & PMD_MASK)) {
1650 clear_page_range(tlb, first, last);
1651 flush_tlb_pgtables(mm, first, last);
1652 }
1653 }
1654
1655 /* Normal function to fix up a mapping
1656 * This function is the default for when an area has no specific
1657 * function. This may be used as part of a more specific routine.
1658 *
1659 * By the time this function is called, the area struct has been
1660 * removed from the process mapping list.
1661 */
1662 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1663 {
1664 size_t len = area->vm_end - area->vm_start;
1665
1666 area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1667 if (area->vm_flags & VM_LOCKED)
1668 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1669 vm_stat_unaccount(area);
1670 area->vm_mm->unmap_area(area);
1671 remove_vm_struct(area);
1672 }
1673
1674 /*
1675 * Update the VMA and inode share lists.
1676 *
1677 * Ok - we have the memory areas we should free on the 'free' list,
1678 * so release them, and do the vma updates.
1679 */
1680 static void unmap_vma_list(struct mm_struct *mm,
1681 struct vm_area_struct *mpnt)
1682 {
1683 do {
1684 struct vm_area_struct *next = mpnt->vm_next;
1685 unmap_vma(mm, mpnt);
1686 mpnt = next;
1687 } while (mpnt != NULL);
1688 validate_mm(mm);
1689 }
1690
1691 /*
1692 * Get rid of page table information in the indicated region.
1693 *
1694 * Called with the page table lock held.
1695 */
1696 static void unmap_region(struct mm_struct *mm,
1697 struct vm_area_struct *vma,
1698 struct vm_area_struct *prev,
1699 unsigned long start,
1700 unsigned long end)
1701 {
1702 struct mmu_gather *tlb;
1703 unsigned long nr_accounted = 0;
1704
1705 lru_add_drain();
1706 tlb = tlb_gather_mmu(mm, 0);
1707 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1708 vm_unacct_memory(nr_accounted);
1709
1710 if (is_hugepage_only_range(start, end - start))
1711 hugetlb_free_pgtables(tlb, prev, start, end);
1712 else
1713 free_pgtables(tlb, prev, start, end);
1714 tlb_finish_mmu(tlb, start, end);
1715 }
1716
1717 /*
1718 * Create a list of vma's touched by the unmap, removing them from the mm's
1719 * vma list as we go..
1720 */
1721 static void
1722 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1723 struct vm_area_struct *prev, unsigned long end)
1724 {
1725 struct vm_area_struct **insertion_point;
1726 struct vm_area_struct *tail_vma = NULL;
1727
1728 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1729 do {
1730 rb_erase(&vma->vm_rb, &mm->mm_rb);
1731 mm->map_count--;
1732 tail_vma = vma;
1733 vma = vma->vm_next;
1734 } while (vma && vma->vm_start < end);
1735 *insertion_point = vma;
1736 tail_vma->vm_next = NULL;
1737 mm->mmap_cache = NULL; /* Kill the cache. */
1738 }
1739
1740 /*
1741 * Split a vma into two pieces at address 'addr', a new vma is allocated
1742 * either for the first part or the the tail.
1743 */
1744 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1745 unsigned long addr, int new_below)
1746 {
1747 struct mempolicy *pol;
1748 struct vm_area_struct *new;
1749
1750 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1751 return -EINVAL;
1752
1753 if (mm->map_count >= sysctl_max_map_count)
1754 return -ENOMEM;
1755
1756 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1757 if (!new)
1758 return -ENOMEM;
1759
1760 /* most fields are the same, copy all, and then fixup */
1761 *new = *vma;
1762
1763 if (new_below)
1764 new->vm_end = addr;
1765 else {
1766 new->vm_start = addr;
1767 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1768 }
1769
1770 pol = mpol_copy(vma_policy(vma));
1771 if (IS_ERR(pol)) {
1772 kmem_cache_free(vm_area_cachep, new);
1773 return PTR_ERR(pol);
1774 }
1775 vma_set_policy(new, pol);
1776
1777 if (new->vm_file)
1778 get_file(new->vm_file);
1779
1780 if (new->vm_ops && new->vm_ops->open)
1781 new->vm_ops->open(new);
1782
1783 if (new_below)
1784 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1785 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1786 else
1787 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1788
1789 return 0;
1790 }
1791
1792 /* Munmap is split into 2 main parts -- this part which finds
1793 * what needs doing, and the areas themselves, which do the
1794 * work. This now handles partial unmappings.
1795 * Jeremy Fitzhardinge <jeremy@goop.org>
1796 */
1797 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1798 {
1799 unsigned long end;
1800 struct vm_area_struct *mpnt, *prev, *last;
1801
1802 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1803 return -EINVAL;
1804
1805 if ((len = PAGE_ALIGN(len)) == 0)
1806 return -EINVAL;
1807
1808 /* Find the first overlapping VMA */
1809 mpnt = find_vma_prev(mm, start, &prev);
1810 if (!mpnt)
1811 return 0;
1812 /* we have start < mpnt->vm_end */
1813
1814 /* if it doesn't overlap, we have nothing.. */
1815 end = start + len;
1816 if (mpnt->vm_start >= end)
1817 return 0;
1818
1819 /*
1820 * If we need to split any vma, do it now to save pain later.
1821 *
1822 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1823 * unmapped vm_area_struct will remain in use: so lower split_vma
1824 * places tmp vma above, and higher split_vma places tmp vma below.
1825 */
1826 if (start > mpnt->vm_start) {
1827 int error = split_vma(mm, mpnt, start, 0);
1828 if (error)
1829 return error;
1830 prev = mpnt;
1831 }
1832
1833 /* Does it split the last one? */
1834 last = find_vma(mm, end);
1835 if (last && end > last->vm_start) {
1836 int error = split_vma(mm, last, end, 1);
1837 if (error)
1838 return error;
1839 }
1840 mpnt = prev? prev->vm_next: mm->mmap;
1841
1842 /*
1843 * Remove the vma's, and unmap the actual pages
1844 */
1845 detach_vmas_to_be_unmapped(mm, mpnt, prev, end);
1846 spin_lock(&mm->page_table_lock);
1847 unmap_region(mm, mpnt, prev, start, end);
1848 spin_unlock(&mm->page_table_lock);
1849
1850 /* Fix up all other VM information */
1851 unmap_vma_list(mm, mpnt);
1852
1853 return 0;
1854 }
1855
1856 EXPORT_SYMBOL(do_munmap);
1857
1858 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1859 {
1860 int ret;
1861 struct mm_struct *mm = current->mm;
1862
1863 profile_munmap(addr);
1864
1865 down_write(&mm->mmap_sem);
1866 ret = do_munmap(mm, addr, len);
1867 up_write(&mm->mmap_sem);
1868 return ret;
1869 }
1870
1871 static inline void verify_mm_writelocked(struct mm_struct *mm)
1872 {
1873 #ifdef CONFIG_DEBUG_KERNEL
1874 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1875 WARN_ON(1);
1876 up_read(&mm->mmap_sem);
1877 }
1878 #endif
1879 }
1880
1881 /*
1882 * this is really a simplified "do_mmap". it only handles
1883 * anonymous maps. eventually we may be able to do some
1884 * brk-specific accounting here.
1885 */
1886 unsigned long do_brk(unsigned long addr, unsigned long len)
1887 {
1888 struct mm_struct * mm = current->mm;
1889 struct vm_area_struct * vma, * prev;
1890 unsigned long flags;
1891 struct rb_node ** rb_link, * rb_parent;
1892 pgoff_t pgoff = addr >> PAGE_SHIFT;
1893
1894 len = PAGE_ALIGN(len);
1895 if (!len)
1896 return addr;
1897
1898 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1899 return -EINVAL;
1900
1901 /*
1902 * mlock MCL_FUTURE?
1903 */
1904 if (mm->def_flags & VM_LOCKED) {
1905 unsigned long locked, lock_limit;
1906 locked = mm->locked_vm << PAGE_SHIFT;
1907 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1908 locked += len;
1909 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1910 return -EAGAIN;
1911 }
1912
1913 /*
1914 * mm->mmap_sem is required to protect against another thread
1915 * changing the mappings in case we sleep.
1916 */
1917 verify_mm_writelocked(mm);
1918
1919 /*
1920 * Clear old maps. this also does some error checking for us
1921 */
1922 munmap_back:
1923 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1924 if (vma && vma->vm_start < addr + len) {
1925 if (do_munmap(mm, addr, len))
1926 return -ENOMEM;
1927 goto munmap_back;
1928 }
1929
1930 /* Check against address space limits *after* clearing old maps... */
1931 if ((mm->total_vm << PAGE_SHIFT) + len
1932 > current->signal->rlim[RLIMIT_AS].rlim_cur)
1933 return -ENOMEM;
1934
1935 if (mm->map_count > sysctl_max_map_count)
1936 return -ENOMEM;
1937
1938 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1939 return -ENOMEM;
1940
1941 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1942
1943 /* Can we just expand an old private anonymous mapping? */
1944 if (vma_merge(mm, prev, addr, addr + len, flags,
1945 NULL, NULL, pgoff, NULL))
1946 goto out;
1947
1948 /*
1949 * create a vma struct for an anonymous mapping
1950 */
1951 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1952 if (!vma) {
1953 vm_unacct_memory(len >> PAGE_SHIFT);
1954 return -ENOMEM;
1955 }
1956 memset(vma, 0, sizeof(*vma));
1957
1958 vma->vm_mm = mm;
1959 vma->vm_start = addr;
1960 vma->vm_end = addr + len;
1961 vma->vm_pgoff = pgoff;
1962 vma->vm_flags = flags;
1963 vma->vm_page_prot = protection_map[flags & 0x0f];
1964 vma_link(mm, vma, prev, rb_link, rb_parent);
1965 out:
1966 mm->total_vm += len >> PAGE_SHIFT;
1967 if (flags & VM_LOCKED) {
1968 mm->locked_vm += len >> PAGE_SHIFT;
1969 make_pages_present(addr, addr + len);
1970 }
1971 acct_update_integrals();
1972 update_mem_hiwater();
1973 return addr;
1974 }
1975
1976 EXPORT_SYMBOL(do_brk);
1977
1978 /* Release all mmaps. */
1979 void exit_mmap(struct mm_struct *mm)
1980 {
1981 struct mmu_gather *tlb;
1982 struct vm_area_struct *vma;
1983 unsigned long nr_accounted = 0;
1984
1985 lru_add_drain();
1986
1987 spin_lock(&mm->page_table_lock);
1988
1989 tlb = tlb_gather_mmu(mm, 1);
1990 flush_cache_mm(mm);
1991 /* Use ~0UL here to ensure all VMAs in the mm are unmapped */
1992 mm->map_count -= unmap_vmas(&tlb, mm, mm->mmap, 0,
1993 ~0UL, &nr_accounted, NULL);
1994 vm_unacct_memory(nr_accounted);
1995 BUG_ON(mm->map_count); /* This is just debugging */
1996 clear_page_range(tlb, FIRST_USER_PGD_NR * PGDIR_SIZE, MM_VM_SIZE(mm));
1997
1998 tlb_finish_mmu(tlb, 0, MM_VM_SIZE(mm));
1999
2000 vma = mm->mmap;
2001 mm->mmap = mm->mmap_cache = NULL;
2002 mm->mm_rb = RB_ROOT;
2003 mm->rss = 0;
2004 mm->total_vm = 0;
2005 mm->locked_vm = 0;
2006
2007 spin_unlock(&mm->page_table_lock);
2008
2009 /*
2010 * Walk the list again, actually closing and freeing it
2011 * without holding any MM locks.
2012 */
2013 while (vma) {
2014 struct vm_area_struct *next = vma->vm_next;
2015 remove_vm_struct(vma);
2016 vma = next;
2017 }
2018 }
2019
2020 /* Insert vm structure into process list sorted by address
2021 * and into the inode's i_mmap tree. If vm_file is non-NULL
2022 * then i_mmap_lock is taken here.
2023 */
2024 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2025 {
2026 struct vm_area_struct * __vma, * prev;
2027 struct rb_node ** rb_link, * rb_parent;
2028
2029 /*
2030 * The vm_pgoff of a purely anonymous vma should be irrelevant
2031 * until its first write fault, when page's anon_vma and index
2032 * are set. But now set the vm_pgoff it will almost certainly
2033 * end up with (unless mremap moves it elsewhere before that
2034 * first wfault), so /proc/pid/maps tells a consistent story.
2035 *
2036 * By setting it to reflect the virtual start address of the
2037 * vma, merges and splits can happen in a seamless way, just
2038 * using the existing file pgoff checks and manipulations.
2039 * Similarly in do_mmap_pgoff and in do_brk.
2040 */
2041 if (!vma->vm_file) {
2042 BUG_ON(vma->anon_vma);
2043 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2044 }
2045 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2046 if (__vma && __vma->vm_start < vma->vm_end)
2047 return -ENOMEM;
2048 vma_link(mm, vma, prev, rb_link, rb_parent);
2049 return 0;
2050 }
2051
2052 /*
2053 * Copy the vma structure to a new location in the same mm,
2054 * prior to moving page table entries, to effect an mremap move.
2055 */
2056 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2057 unsigned long addr, unsigned long len, pgoff_t pgoff)
2058 {
2059 struct vm_area_struct *vma = *vmap;
2060 unsigned long vma_start = vma->vm_start;
2061 struct mm_struct *mm = vma->vm_mm;
2062 struct vm_area_struct *new_vma, *prev;
2063 struct rb_node **rb_link, *rb_parent;
2064 struct mempolicy *pol;
2065
2066 /*
2067 * If anonymous vma has not yet been faulted, update new pgoff
2068 * to match new location, to increase its chance of merging.
2069 */
2070 if (!vma->vm_file && !vma->anon_vma)
2071 pgoff = addr >> PAGE_SHIFT;
2072
2073 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2074 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2075 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2076 if (new_vma) {
2077 /*
2078 * Source vma may have been merged into new_vma
2079 */
2080 if (vma_start >= new_vma->vm_start &&
2081 vma_start < new_vma->vm_end)
2082 *vmap = new_vma;
2083 } else {
2084 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2085 if (new_vma) {
2086 *new_vma = *vma;
2087 pol = mpol_copy(vma_policy(vma));
2088 if (IS_ERR(pol)) {
2089 kmem_cache_free(vm_area_cachep, new_vma);
2090 return NULL;
2091 }
2092 vma_set_policy(new_vma, pol);
2093 new_vma->vm_start = addr;
2094 new_vma->vm_end = addr + len;
2095 new_vma->vm_pgoff = pgoff;
2096 if (new_vma->vm_file)
2097 get_file(new_vma->vm_file);
2098 if (new_vma->vm_ops && new_vma->vm_ops->open)
2099 new_vma->vm_ops->open(new_vma);
2100 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2101 }
2102 }
2103 return new_vma;
2104 }
2105
|
This page was automatically generated by the
LXR engine.
|