blob: 4f8def03428ce967d91f7a1fa738733d8a386758 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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/mount.h>
25#include <linux/mempolicy.h>
26#include <linux/rmap.h>
27
28#include <asm/uaccess.h>
29#include <asm/cacheflush.h>
30#include <asm/tlb.h>
31
Hugh Dickinse0da3822005-04-19 13:29:15 -070032static void unmap_region(struct mm_struct *mm,
33 struct vm_area_struct *vma, struct vm_area_struct *prev,
34 unsigned long start, unsigned long end);
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/*
37 * WARNING: the debugging will use recursive algorithms so never enable this
38 * unless you know what you are doing.
39 */
40#undef DEBUG_MM_RB
41
42/* description of effects of mapping type and prot in current implementation.
43 * this is due to the limited x86 page protection hardware. The expected
44 * behavior is in parens:
45 *
46 * map_type prot
47 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
48 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
49 * w: (no) no w: (no) no w: (yes) yes w: (no) no
50 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
51 *
52 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
53 * w: (no) no w: (no) no w: (copy) copy w: (no) no
54 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
55 *
56 */
57pgprot_t protection_map[16] = {
58 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
59 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
60};
61
62int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070064int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065atomic_t vm_committed_space = ATOMIC_INIT(0);
66
67/*
68 * Check that a process has enough memory to allocate a new virtual
69 * mapping. 0 means there is enough memory for the allocation to
70 * succeed and -ENOMEM implies there is not.
71 *
72 * We currently support three overcommit policies, which are set via the
73 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
74 *
75 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
76 * Additional code 2002 Jul 20 by Robert Love.
77 *
78 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
79 *
80 * Note this is a helper function intended to be used by LSMs which
81 * wish to use this logic.
82 */
83int __vm_enough_memory(long pages, int cap_sys_admin)
84{
85 unsigned long free, allowed;
86
87 vm_acct_memory(pages);
88
89 /*
90 * Sometimes we want to use more memory than we have
91 */
92 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
93 return 0;
94
95 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
96 unsigned long n;
97
98 free = get_page_cache_size();
99 free += nr_swap_pages;
100
101 /*
102 * Any slabs which are created with the
103 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
104 * which are reclaimable, under pressure. The dentry
105 * cache and most inode caches should fall into this
106 */
107 free += atomic_read(&slab_reclaim_pages);
108
109 /*
110 * Leave the last 3% for root
111 */
112 if (!cap_sys_admin)
113 free -= free / 32;
114
115 if (free > pages)
116 return 0;
117
118 /*
119 * nr_free_pages() is very expensive on large systems,
120 * only call if we're about to fail.
121 */
122 n = nr_free_pages();
123 if (!cap_sys_admin)
124 n -= n / 32;
125 free += n;
126
127 if (free > pages)
128 return 0;
129 vm_unacct_memory(pages);
130 return -ENOMEM;
131 }
132
133 allowed = (totalram_pages - hugetlb_total_pages())
134 * sysctl_overcommit_ratio / 100;
135 /*
136 * Leave the last 3% for root
137 */
138 if (!cap_sys_admin)
139 allowed -= allowed / 32;
140 allowed += total_swap_pages;
141
142 /* Don't let a single process grow too big:
143 leave 3% of the size of this process for other processes */
144 allowed -= current->mm->total_vm / 32;
145
Simon Derr2f60f8d2005-08-04 19:52:03 -0700146 /*
147 * cast `allowed' as a signed long because vm_committed_space
148 * sometimes has a negative value
149 */
150 if (atomic_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 return 0;
152
153 vm_unacct_memory(pages);
154
155 return -ENOMEM;
156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158EXPORT_SYMBOL(__vm_enough_memory);
159
160/*
161 * Requires inode->i_mapping->i_mmap_lock
162 */
163static void __remove_shared_vm_struct(struct vm_area_struct *vma,
164 struct file *file, struct address_space *mapping)
165{
166 if (vma->vm_flags & VM_DENYWRITE)
167 atomic_inc(&file->f_dentry->d_inode->i_writecount);
168 if (vma->vm_flags & VM_SHARED)
169 mapping->i_mmap_writable--;
170
171 flush_dcache_mmap_lock(mapping);
172 if (unlikely(vma->vm_flags & VM_NONLINEAR))
173 list_del_init(&vma->shared.vm_set.list);
174 else
175 vma_prio_tree_remove(vma, &mapping->i_mmap);
176 flush_dcache_mmap_unlock(mapping);
177}
178
179/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700180 * Unlink a file-based vm structure from its prio_tree, to hide
181 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700183void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
185 struct file *file = vma->vm_file;
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 if (file) {
188 struct address_space *mapping = file->f_mapping;
189 spin_lock(&mapping->i_mmap_lock);
190 __remove_shared_vm_struct(vma, file, mapping);
191 spin_unlock(&mapping->i_mmap_lock);
192 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700193}
194
195/*
196 * Close a vm structure and free it, returning the next.
197 */
198static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
199{
200 struct vm_area_struct *next = vma->vm_next;
201
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700202 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 if (vma->vm_ops && vma->vm_ops->close)
204 vma->vm_ops->close(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700205 if (vma->vm_file)
206 fput(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 mpol_free(vma_policy(vma));
208 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700209 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212asmlinkage unsigned long sys_brk(unsigned long brk)
213{
214 unsigned long rlim, retval;
215 unsigned long newbrk, oldbrk;
216 struct mm_struct *mm = current->mm;
217
218 down_write(&mm->mmap_sem);
219
220 if (brk < mm->end_code)
221 goto out;
222 newbrk = PAGE_ALIGN(brk);
223 oldbrk = PAGE_ALIGN(mm->brk);
224 if (oldbrk == newbrk)
225 goto set_brk;
226
227 /* Always allow shrinking brk. */
228 if (brk <= mm->brk) {
229 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
230 goto set_brk;
231 goto out;
232 }
233
234 /* Check against rlimit.. */
235 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
236 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
237 goto out;
238
239 /* Check against existing mmap mappings. */
240 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
241 goto out;
242
243 /* Ok, looks good - let it rip. */
244 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
245 goto out;
246set_brk:
247 mm->brk = brk;
248out:
249 retval = mm->brk;
250 up_write(&mm->mmap_sem);
251 return retval;
252}
253
254#ifdef DEBUG_MM_RB
255static int browse_rb(struct rb_root *root)
256{
257 int i = 0, j;
258 struct rb_node *nd, *pn = NULL;
259 unsigned long prev = 0, pend = 0;
260
261 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
262 struct vm_area_struct *vma;
263 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
264 if (vma->vm_start < prev)
265 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
266 if (vma->vm_start < pend)
267 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
268 if (vma->vm_start > vma->vm_end)
269 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
270 i++;
271 pn = nd;
272 }
273 j = 0;
274 for (nd = pn; nd; nd = rb_prev(nd)) {
275 j++;
276 }
277 if (i != j)
278 printk("backwards %d, forwards %d\n", j, i), i = 0;
279 return i;
280}
281
282void validate_mm(struct mm_struct *mm)
283{
284 int bug = 0;
285 int i = 0;
286 struct vm_area_struct *tmp = mm->mmap;
287 while (tmp) {
288 tmp = tmp->vm_next;
289 i++;
290 }
291 if (i != mm->map_count)
292 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
293 i = browse_rb(&mm->mm_rb);
294 if (i != mm->map_count)
295 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
296 if (bug)
297 BUG();
298}
299#else
300#define validate_mm(mm) do { } while (0)
301#endif
302
303static struct vm_area_struct *
304find_vma_prepare(struct mm_struct *mm, unsigned long addr,
305 struct vm_area_struct **pprev, struct rb_node ***rb_link,
306 struct rb_node ** rb_parent)
307{
308 struct vm_area_struct * vma;
309 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
310
311 __rb_link = &mm->mm_rb.rb_node;
312 rb_prev = __rb_parent = NULL;
313 vma = NULL;
314
315 while (*__rb_link) {
316 struct vm_area_struct *vma_tmp;
317
318 __rb_parent = *__rb_link;
319 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
320
321 if (vma_tmp->vm_end > addr) {
322 vma = vma_tmp;
323 if (vma_tmp->vm_start <= addr)
324 return vma;
325 __rb_link = &__rb_parent->rb_left;
326 } else {
327 rb_prev = __rb_parent;
328 __rb_link = &__rb_parent->rb_right;
329 }
330 }
331
332 *pprev = NULL;
333 if (rb_prev)
334 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
335 *rb_link = __rb_link;
336 *rb_parent = __rb_parent;
337 return vma;
338}
339
340static inline void
341__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
342 struct vm_area_struct *prev, struct rb_node *rb_parent)
343{
344 if (prev) {
345 vma->vm_next = prev->vm_next;
346 prev->vm_next = vma;
347 } else {
348 mm->mmap = vma;
349 if (rb_parent)
350 vma->vm_next = rb_entry(rb_parent,
351 struct vm_area_struct, vm_rb);
352 else
353 vma->vm_next = NULL;
354 }
355}
356
357void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
358 struct rb_node **rb_link, struct rb_node *rb_parent)
359{
360 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
361 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
362}
363
364static inline void __vma_link_file(struct vm_area_struct *vma)
365{
366 struct file * file;
367
368 file = vma->vm_file;
369 if (file) {
370 struct address_space *mapping = file->f_mapping;
371
372 if (vma->vm_flags & VM_DENYWRITE)
373 atomic_dec(&file->f_dentry->d_inode->i_writecount);
374 if (vma->vm_flags & VM_SHARED)
375 mapping->i_mmap_writable++;
376
377 flush_dcache_mmap_lock(mapping);
378 if (unlikely(vma->vm_flags & VM_NONLINEAR))
379 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
380 else
381 vma_prio_tree_insert(vma, &mapping->i_mmap);
382 flush_dcache_mmap_unlock(mapping);
383 }
384}
385
386static void
387__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
388 struct vm_area_struct *prev, struct rb_node **rb_link,
389 struct rb_node *rb_parent)
390{
391 __vma_link_list(mm, vma, prev, rb_parent);
392 __vma_link_rb(mm, vma, rb_link, rb_parent);
393 __anon_vma_link(vma);
394}
395
396static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
397 struct vm_area_struct *prev, struct rb_node **rb_link,
398 struct rb_node *rb_parent)
399{
400 struct address_space *mapping = NULL;
401
402 if (vma->vm_file)
403 mapping = vma->vm_file->f_mapping;
404
405 if (mapping) {
406 spin_lock(&mapping->i_mmap_lock);
407 vma->vm_truncate_count = mapping->truncate_count;
408 }
409 anon_vma_lock(vma);
410
411 __vma_link(mm, vma, prev, rb_link, rb_parent);
412 __vma_link_file(vma);
413
414 anon_vma_unlock(vma);
415 if (mapping)
416 spin_unlock(&mapping->i_mmap_lock);
417
418 mm->map_count++;
419 validate_mm(mm);
420}
421
422/*
423 * Helper for vma_adjust in the split_vma insert case:
424 * insert vm structure into list and rbtree and anon_vma,
425 * but it has already been inserted into prio_tree earlier.
426 */
427static void
428__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
429{
430 struct vm_area_struct * __vma, * prev;
431 struct rb_node ** rb_link, * rb_parent;
432
433 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
434 if (__vma && __vma->vm_start < vma->vm_end)
435 BUG();
436 __vma_link(mm, vma, prev, rb_link, rb_parent);
437 mm->map_count++;
438}
439
440static inline void
441__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
442 struct vm_area_struct *prev)
443{
444 prev->vm_next = vma->vm_next;
445 rb_erase(&vma->vm_rb, &mm->mm_rb);
446 if (mm->mmap_cache == vma)
447 mm->mmap_cache = prev;
448}
449
450/*
451 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
452 * is already present in an i_mmap tree without adjusting the tree.
453 * The following helper function should be used when such adjustments
454 * are necessary. The "insert" vma (if any) is to be inserted
455 * before we drop the necessary locks.
456 */
457void vma_adjust(struct vm_area_struct *vma, unsigned long start,
458 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
459{
460 struct mm_struct *mm = vma->vm_mm;
461 struct vm_area_struct *next = vma->vm_next;
462 struct vm_area_struct *importer = NULL;
463 struct address_space *mapping = NULL;
464 struct prio_tree_root *root = NULL;
465 struct file *file = vma->vm_file;
466 struct anon_vma *anon_vma = NULL;
467 long adjust_next = 0;
468 int remove_next = 0;
469
470 if (next && !insert) {
471 if (end >= next->vm_end) {
472 /*
473 * vma expands, overlapping all the next, and
474 * perhaps the one after too (mprotect case 6).
475 */
476again: remove_next = 1 + (end > next->vm_end);
477 end = next->vm_end;
478 anon_vma = next->anon_vma;
479 importer = vma;
480 } else if (end > next->vm_start) {
481 /*
482 * vma expands, overlapping part of the next:
483 * mprotect case 5 shifting the boundary up.
484 */
485 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
486 anon_vma = next->anon_vma;
487 importer = vma;
488 } else if (end < vma->vm_end) {
489 /*
490 * vma shrinks, and !insert tells it's not
491 * split_vma inserting another: so it must be
492 * mprotect case 4 shifting the boundary down.
493 */
494 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
495 anon_vma = next->anon_vma;
496 importer = next;
497 }
498 }
499
500 if (file) {
501 mapping = file->f_mapping;
502 if (!(vma->vm_flags & VM_NONLINEAR))
503 root = &mapping->i_mmap;
504 spin_lock(&mapping->i_mmap_lock);
505 if (importer &&
506 vma->vm_truncate_count != next->vm_truncate_count) {
507 /*
508 * unmap_mapping_range might be in progress:
509 * ensure that the expanding vma is rescanned.
510 */
511 importer->vm_truncate_count = 0;
512 }
513 if (insert) {
514 insert->vm_truncate_count = vma->vm_truncate_count;
515 /*
516 * Put into prio_tree now, so instantiated pages
517 * are visible to arm/parisc __flush_dcache_page
518 * throughout; but we cannot insert into address
519 * space until vma start or end is updated.
520 */
521 __vma_link_file(insert);
522 }
523 }
524
525 /*
526 * When changing only vma->vm_end, we don't really need
527 * anon_vma lock: but is that case worth optimizing out?
528 */
529 if (vma->anon_vma)
530 anon_vma = vma->anon_vma;
531 if (anon_vma) {
532 spin_lock(&anon_vma->lock);
533 /*
534 * Easily overlooked: when mprotect shifts the boundary,
535 * make sure the expanding vma has anon_vma set if the
536 * shrinking vma had, to cover any anon pages imported.
537 */
538 if (importer && !importer->anon_vma) {
539 importer->anon_vma = anon_vma;
540 __anon_vma_link(importer);
541 }
542 }
543
544 if (root) {
545 flush_dcache_mmap_lock(mapping);
546 vma_prio_tree_remove(vma, root);
547 if (adjust_next)
548 vma_prio_tree_remove(next, root);
549 }
550
551 vma->vm_start = start;
552 vma->vm_end = end;
553 vma->vm_pgoff = pgoff;
554 if (adjust_next) {
555 next->vm_start += adjust_next << PAGE_SHIFT;
556 next->vm_pgoff += adjust_next;
557 }
558
559 if (root) {
560 if (adjust_next)
561 vma_prio_tree_insert(next, root);
562 vma_prio_tree_insert(vma, root);
563 flush_dcache_mmap_unlock(mapping);
564 }
565
566 if (remove_next) {
567 /*
568 * vma_merge has merged next into vma, and needs
569 * us to remove next before dropping the locks.
570 */
571 __vma_unlink(mm, next, vma);
572 if (file)
573 __remove_shared_vm_struct(next, file, mapping);
574 if (next->anon_vma)
575 __anon_vma_merge(vma, next);
576 } else if (insert) {
577 /*
578 * split_vma has split insert from vma, and needs
579 * us to insert it before dropping the locks
580 * (it may either follow vma or precede it).
581 */
582 __insert_vm_struct(mm, insert);
583 }
584
585 if (anon_vma)
586 spin_unlock(&anon_vma->lock);
587 if (mapping)
588 spin_unlock(&mapping->i_mmap_lock);
589
590 if (remove_next) {
591 if (file)
592 fput(file);
593 mm->map_count--;
594 mpol_free(vma_policy(next));
595 kmem_cache_free(vm_area_cachep, next);
596 /*
597 * In mprotect's case 6 (see comments on vma_merge),
598 * we must remove another next too. It would clutter
599 * up the code too much to do both in one go.
600 */
601 if (remove_next == 2) {
602 next = vma->vm_next;
603 goto again;
604 }
605 }
606
607 validate_mm(mm);
608}
609
610/*
611 * If the vma has a ->close operation then the driver probably needs to release
612 * per-vma resources, so we don't attempt to merge those.
613 */
614#define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
615
616static inline int is_mergeable_vma(struct vm_area_struct *vma,
617 struct file *file, unsigned long vm_flags)
618{
619 if (vma->vm_flags != vm_flags)
620 return 0;
621 if (vma->vm_file != file)
622 return 0;
623 if (vma->vm_ops && vma->vm_ops->close)
624 return 0;
625 return 1;
626}
627
628static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
629 struct anon_vma *anon_vma2)
630{
631 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
632}
633
634/*
635 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
636 * in front of (at a lower virtual address and file offset than) the vma.
637 *
638 * We cannot merge two vmas if they have differently assigned (non-NULL)
639 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
640 *
641 * We don't check here for the merged mmap wrapping around the end of pagecache
642 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
643 * wrap, nor mmaps which cover the final page at index -1UL.
644 */
645static int
646can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
647 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
648{
649 if (is_mergeable_vma(vma, file, vm_flags) &&
650 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
651 if (vma->vm_pgoff == vm_pgoff)
652 return 1;
653 }
654 return 0;
655}
656
657/*
658 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
659 * beyond (at a higher virtual address and file offset than) the vma.
660 *
661 * We cannot merge two vmas if they have differently assigned (non-NULL)
662 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
663 */
664static int
665can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
666 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
667{
668 if (is_mergeable_vma(vma, file, vm_flags) &&
669 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
670 pgoff_t vm_pglen;
671 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
672 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
673 return 1;
674 }
675 return 0;
676}
677
678/*
679 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
680 * whether that can be merged with its predecessor or its successor.
681 * Or both (it neatly fills a hole).
682 *
683 * In most cases - when called for mmap, brk or mremap - [addr,end) is
684 * certain not to be mapped by the time vma_merge is called; but when
685 * called for mprotect, it is certain to be already mapped (either at
686 * an offset within prev, or at the start of next), and the flags of
687 * this area are about to be changed to vm_flags - and the no-change
688 * case has already been eliminated.
689 *
690 * The following mprotect cases have to be considered, where AAAA is
691 * the area passed down from mprotect_fixup, never extending beyond one
692 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
693 *
694 * AAAA AAAA AAAA AAAA
695 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
696 * cannot merge might become might become might become
697 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
698 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
699 * mremap move: PPPPNNNNNNNN 8
700 * AAAA
701 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
702 * might become case 1 below case 2 below case 3 below
703 *
704 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
705 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
706 */
707struct vm_area_struct *vma_merge(struct mm_struct *mm,
708 struct vm_area_struct *prev, unsigned long addr,
709 unsigned long end, unsigned long vm_flags,
710 struct anon_vma *anon_vma, struct file *file,
711 pgoff_t pgoff, struct mempolicy *policy)
712{
713 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
714 struct vm_area_struct *area, *next;
715
716 /*
717 * We later require that vma->vm_flags == vm_flags,
718 * so this tests vma->vm_flags & VM_SPECIAL, too.
719 */
720 if (vm_flags & VM_SPECIAL)
721 return NULL;
722
723 if (prev)
724 next = prev->vm_next;
725 else
726 next = mm->mmap;
727 area = next;
728 if (next && next->vm_end == end) /* cases 6, 7, 8 */
729 next = next->vm_next;
730
731 /*
732 * Can it merge with the predecessor?
733 */
734 if (prev && prev->vm_end == addr &&
735 mpol_equal(vma_policy(prev), policy) &&
736 can_vma_merge_after(prev, vm_flags,
737 anon_vma, file, pgoff)) {
738 /*
739 * OK, it can. Can we now merge in the successor as well?
740 */
741 if (next && end == next->vm_start &&
742 mpol_equal(policy, vma_policy(next)) &&
743 can_vma_merge_before(next, vm_flags,
744 anon_vma, file, pgoff+pglen) &&
745 is_mergeable_anon_vma(prev->anon_vma,
746 next->anon_vma)) {
747 /* cases 1, 6 */
748 vma_adjust(prev, prev->vm_start,
749 next->vm_end, prev->vm_pgoff, NULL);
750 } else /* cases 2, 5, 7 */
751 vma_adjust(prev, prev->vm_start,
752 end, prev->vm_pgoff, NULL);
753 return prev;
754 }
755
756 /*
757 * Can this new request be merged in front of next?
758 */
759 if (next && end == next->vm_start &&
760 mpol_equal(policy, vma_policy(next)) &&
761 can_vma_merge_before(next, vm_flags,
762 anon_vma, file, pgoff+pglen)) {
763 if (prev && addr < prev->vm_end) /* case 4 */
764 vma_adjust(prev, prev->vm_start,
765 addr, prev->vm_pgoff, NULL);
766 else /* cases 3, 8 */
767 vma_adjust(area, addr, next->vm_end,
768 next->vm_pgoff - pglen, NULL);
769 return area;
770 }
771
772 return NULL;
773}
774
775/*
776 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
777 * neighbouring vmas for a suitable anon_vma, before it goes off
778 * to allocate a new anon_vma. It checks because a repetitive
779 * sequence of mprotects and faults may otherwise lead to distinct
780 * anon_vmas being allocated, preventing vma merge in subsequent
781 * mprotect.
782 */
783struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
784{
785 struct vm_area_struct *near;
786 unsigned long vm_flags;
787
788 near = vma->vm_next;
789 if (!near)
790 goto try_prev;
791
792 /*
793 * Since only mprotect tries to remerge vmas, match flags
794 * which might be mprotected into each other later on.
795 * Neither mlock nor madvise tries to remerge at present,
796 * so leave their flags as obstructing a merge.
797 */
798 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
799 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
800
801 if (near->anon_vma && vma->vm_end == near->vm_start &&
802 mpol_equal(vma_policy(vma), vma_policy(near)) &&
803 can_vma_merge_before(near, vm_flags,
804 NULL, vma->vm_file, vma->vm_pgoff +
805 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
806 return near->anon_vma;
807try_prev:
808 /*
809 * It is potentially slow to have to call find_vma_prev here.
810 * But it's only on the first write fault on the vma, not
811 * every time, and we could devise a way to avoid it later
812 * (e.g. stash info in next's anon_vma_node when assigning
813 * an anon_vma, or when trying vma_merge). Another time.
814 */
815 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
816 BUG();
817 if (!near)
818 goto none;
819
820 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
821 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
822
823 if (near->anon_vma && near->vm_end == vma->vm_start &&
824 mpol_equal(vma_policy(near), vma_policy(vma)) &&
825 can_vma_merge_after(near, vm_flags,
826 NULL, vma->vm_file, vma->vm_pgoff))
827 return near->anon_vma;
828none:
829 /*
830 * There's no absolute need to look only at touching neighbours:
831 * we could search further afield for "compatible" anon_vmas.
832 * But it would probably just be a waste of time searching,
833 * or lead to too many vmas hanging off the same anon_vma.
834 * We're trying to allow mprotect remerging later on,
835 * not trying to minimize memory used for anon_vmas.
836 */
837 return NULL;
838}
839
840#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700841void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 struct file *file, long pages)
843{
844 const unsigned long stack_flags
845 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
846
847#ifdef CONFIG_HUGETLB
848 if (flags & VM_HUGETLB) {
849 if (!(flags & VM_DONTCOPY))
850 mm->shared_vm += pages;
851 return;
852 }
853#endif /* CONFIG_HUGETLB */
854
855 if (file) {
856 mm->shared_vm += pages;
857 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
858 mm->exec_vm += pages;
859 } else if (flags & stack_flags)
860 mm->stack_vm += pages;
861 if (flags & (VM_RESERVED|VM_IO))
862 mm->reserved_vm += pages;
863}
864#endif /* CONFIG_PROC_FS */
865
866/*
867 * The caller must hold down_write(current->mm->mmap_sem).
868 */
869
870unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
871 unsigned long len, unsigned long prot,
872 unsigned long flags, unsigned long pgoff)
873{
874 struct mm_struct * mm = current->mm;
875 struct vm_area_struct * vma, * prev;
876 struct inode *inode;
877 unsigned int vm_flags;
878 int correct_wcount = 0;
879 int error;
880 struct rb_node ** rb_link, * rb_parent;
881 int accountable = 1;
882 unsigned long charged = 0, reqprot = prot;
883
884 if (file) {
885 if (is_file_hugepages(file))
886 accountable = 0;
887
888 if (!file->f_op || !file->f_op->mmap)
889 return -ENODEV;
890
891 if ((prot & PROT_EXEC) &&
892 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
893 return -EPERM;
894 }
895 /*
896 * Does the application expect PROT_READ to imply PROT_EXEC?
897 *
898 * (the exception is when the underlying filesystem is noexec
899 * mounted, in which case we dont add PROT_EXEC.)
900 */
901 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
902 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
903 prot |= PROT_EXEC;
904
905 if (!len)
906 return -EINVAL;
907
908 /* Careful about overflows.. */
909 len = PAGE_ALIGN(len);
910 if (!len || len > TASK_SIZE)
911 return -ENOMEM;
912
913 /* offset overflow? */
914 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
915 return -EOVERFLOW;
916
917 /* Too many mappings? */
918 if (mm->map_count > sysctl_max_map_count)
919 return -ENOMEM;
920
921 /* Obtain the address to map to. we verify (or select) it and ensure
922 * that it represents a valid section of the address space.
923 */
924 addr = get_unmapped_area(file, addr, len, pgoff, flags);
925 if (addr & ~PAGE_MASK)
926 return addr;
927
928 /* Do simple checking here so the lower-level routines won't have
929 * to. we assume access permissions have been handled by the open
930 * of the memory object, so we don't do any here.
931 */
932 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
933 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
934
935 if (flags & MAP_LOCKED) {
936 if (!can_do_mlock())
937 return -EPERM;
938 vm_flags |= VM_LOCKED;
939 }
940 /* mlock MCL_FUTURE? */
941 if (vm_flags & VM_LOCKED) {
942 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700943 locked = len >> PAGE_SHIFT;
944 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700946 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
948 return -EAGAIN;
949 }
950
951 inode = file ? file->f_dentry->d_inode : NULL;
952
953 if (file) {
954 switch (flags & MAP_TYPE) {
955 case MAP_SHARED:
956 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
957 return -EACCES;
958
959 /*
960 * Make sure we don't allow writing to an append-only
961 * file..
962 */
963 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
964 return -EACCES;
965
966 /*
967 * Make sure there are no mandatory locks on the file.
968 */
969 if (locks_verify_locked(inode))
970 return -EAGAIN;
971
972 vm_flags |= VM_SHARED | VM_MAYSHARE;
973 if (!(file->f_mode & FMODE_WRITE))
974 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
975
976 /* fall through */
977 case MAP_PRIVATE:
978 if (!(file->f_mode & FMODE_READ))
979 return -EACCES;
980 break;
981
982 default:
983 return -EINVAL;
984 }
985 } else {
986 switch (flags & MAP_TYPE) {
987 case MAP_SHARED:
988 vm_flags |= VM_SHARED | VM_MAYSHARE;
989 break;
990 case MAP_PRIVATE:
991 /*
992 * Set pgoff according to addr for anon_vma.
993 */
994 pgoff = addr >> PAGE_SHIFT;
995 break;
996 default:
997 return -EINVAL;
998 }
999 }
1000
1001 error = security_file_mmap(file, reqprot, prot, flags);
1002 if (error)
1003 return error;
1004
1005 /* Clear old maps */
1006 error = -ENOMEM;
1007munmap_back:
1008 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1009 if (vma && vma->vm_start < addr + len) {
1010 if (do_munmap(mm, addr, len))
1011 return -ENOMEM;
1012 goto munmap_back;
1013 }
1014
1015 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001016 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 return -ENOMEM;
1018
1019 if (accountable && (!(flags & MAP_NORESERVE) ||
1020 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1021 if (vm_flags & VM_SHARED) {
1022 /* Check memory availability in shmem_file_setup? */
1023 vm_flags |= VM_ACCOUNT;
1024 } else if (vm_flags & VM_WRITE) {
1025 /*
1026 * Private writable mapping: check memory availability
1027 */
1028 charged = len >> PAGE_SHIFT;
1029 if (security_vm_enough_memory(charged))
1030 return -ENOMEM;
1031 vm_flags |= VM_ACCOUNT;
1032 }
1033 }
1034
1035 /*
1036 * Can we just expand an old private anonymous mapping?
1037 * The VM_SHARED test is necessary because shmem_zero_setup
1038 * will create the file object for a shared anonymous map below.
1039 */
1040 if (!file && !(vm_flags & VM_SHARED) &&
1041 vma_merge(mm, prev, addr, addr + len, vm_flags,
1042 NULL, NULL, pgoff, NULL))
1043 goto out;
1044
1045 /*
1046 * Determine the object being mapped and call the appropriate
1047 * specific mapper. the address has already been validated, but
1048 * not unmapped, but the maps are removed from the list.
1049 */
1050 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1051 if (!vma) {
1052 error = -ENOMEM;
1053 goto unacct_error;
1054 }
1055 memset(vma, 0, sizeof(*vma));
1056
1057 vma->vm_mm = mm;
1058 vma->vm_start = addr;
1059 vma->vm_end = addr + len;
1060 vma->vm_flags = vm_flags;
1061 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1062 vma->vm_pgoff = pgoff;
1063
1064 if (file) {
1065 error = -EINVAL;
1066 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1067 goto free_vma;
1068 if (vm_flags & VM_DENYWRITE) {
1069 error = deny_write_access(file);
1070 if (error)
1071 goto free_vma;
1072 correct_wcount = 1;
1073 }
1074 vma->vm_file = file;
1075 get_file(file);
1076 error = file->f_op->mmap(file, vma);
1077 if (error)
1078 goto unmap_and_free_vma;
Nick Pigginb5810032005-10-29 18:16:12 -07001079 if ((vma->vm_flags & (VM_SHARED | VM_WRITE | VM_RESERVED))
1080 == (VM_WRITE | VM_RESERVED)) {
1081 printk(KERN_WARNING "program %s is using MAP_PRIVATE, "
1082 "PROT_WRITE mmap of VM_RESERVED memory, which "
1083 "is deprecated. Please report this to "
1084 "linux-kernel@vger.kernel.org\n",current->comm);
1085 if (vma->vm_ops && vma->vm_ops->close)
1086 vma->vm_ops->close(vma);
1087 error = -EACCES;
1088 goto unmap_and_free_vma;
1089 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 } else if (vm_flags & VM_SHARED) {
1091 error = shmem_zero_setup(vma);
1092 if (error)
1093 goto free_vma;
1094 }
1095
1096 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1097 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1098 * that memory reservation must be checked; but that reservation
1099 * belongs to shared memory object, not to vma: so now clear it.
1100 */
1101 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1102 vma->vm_flags &= ~VM_ACCOUNT;
1103
1104 /* Can addr have changed??
1105 *
1106 * Answer: Yes, several device drivers can do it in their
1107 * f_op->mmap method. -DaveM
1108 */
1109 addr = vma->vm_start;
1110 pgoff = vma->vm_pgoff;
1111 vm_flags = vma->vm_flags;
1112
1113 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1114 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1115 file = vma->vm_file;
1116 vma_link(mm, vma, prev, rb_link, rb_parent);
1117 if (correct_wcount)
1118 atomic_inc(&inode->i_writecount);
1119 } else {
1120 if (file) {
1121 if (correct_wcount)
1122 atomic_inc(&inode->i_writecount);
1123 fput(file);
1124 }
1125 mpol_free(vma_policy(vma));
1126 kmem_cache_free(vm_area_cachep, vma);
1127 }
1128out:
1129 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001130 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 if (vm_flags & VM_LOCKED) {
1132 mm->locked_vm += len >> PAGE_SHIFT;
1133 make_pages_present(addr, addr + len);
1134 }
1135 if (flags & MAP_POPULATE) {
1136 up_write(&mm->mmap_sem);
1137 sys_remap_file_pages(addr, len, 0,
1138 pgoff, flags & MAP_NONBLOCK);
1139 down_write(&mm->mmap_sem);
1140 }
1141 return addr;
1142
1143unmap_and_free_vma:
1144 if (correct_wcount)
1145 atomic_inc(&inode->i_writecount);
1146 vma->vm_file = NULL;
1147 fput(file);
1148
1149 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001150 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1151 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152free_vma:
1153 kmem_cache_free(vm_area_cachep, vma);
1154unacct_error:
1155 if (charged)
1156 vm_unacct_memory(charged);
1157 return error;
1158}
1159
1160EXPORT_SYMBOL(do_mmap_pgoff);
1161
1162/* Get an address range which is currently unmapped.
1163 * For shmat() with addr=0.
1164 *
1165 * Ugly calling convention alert:
1166 * Return value with the low bits set means error value,
1167 * ie
1168 * if (ret & ~PAGE_MASK)
1169 * error = ret;
1170 *
1171 * This function "knows" that -ENOMEM has the bits set.
1172 */
1173#ifndef HAVE_ARCH_UNMAPPED_AREA
1174unsigned long
1175arch_get_unmapped_area(struct file *filp, unsigned long addr,
1176 unsigned long len, unsigned long pgoff, unsigned long flags)
1177{
1178 struct mm_struct *mm = current->mm;
1179 struct vm_area_struct *vma;
1180 unsigned long start_addr;
1181
1182 if (len > TASK_SIZE)
1183 return -ENOMEM;
1184
1185 if (addr) {
1186 addr = PAGE_ALIGN(addr);
1187 vma = find_vma(mm, addr);
1188 if (TASK_SIZE - len >= addr &&
1189 (!vma || addr + len <= vma->vm_start))
1190 return addr;
1191 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001192 if (len > mm->cached_hole_size) {
1193 start_addr = addr = mm->free_area_cache;
1194 } else {
1195 start_addr = addr = TASK_UNMAPPED_BASE;
1196 mm->cached_hole_size = 0;
1197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199full_search:
1200 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1201 /* At this point: (!vma || addr < vma->vm_end). */
1202 if (TASK_SIZE - len < addr) {
1203 /*
1204 * Start a new search - just in case we missed
1205 * some holes.
1206 */
1207 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001208 addr = TASK_UNMAPPED_BASE;
1209 start_addr = addr;
1210 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 goto full_search;
1212 }
1213 return -ENOMEM;
1214 }
1215 if (!vma || addr + len <= vma->vm_start) {
1216 /*
1217 * Remember the place where we stopped the search:
1218 */
1219 mm->free_area_cache = addr + len;
1220 return addr;
1221 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001222 if (addr + mm->cached_hole_size < vma->vm_start)
1223 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 addr = vma->vm_end;
1225 }
1226}
1227#endif
1228
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001229void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
1231 /*
1232 * Is this a new hole at the lowest possible address?
1233 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001234 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1235 mm->free_area_cache = addr;
1236 mm->cached_hole_size = ~0UL;
1237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238}
1239
1240/*
1241 * This mmap-allocator allocates new areas top-down from below the
1242 * stack's low limit (the base):
1243 */
1244#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1245unsigned long
1246arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1247 const unsigned long len, const unsigned long pgoff,
1248 const unsigned long flags)
1249{
1250 struct vm_area_struct *vma;
1251 struct mm_struct *mm = current->mm;
1252 unsigned long addr = addr0;
1253
1254 /* requested length too big for entire address space */
1255 if (len > TASK_SIZE)
1256 return -ENOMEM;
1257
1258 /* requesting a specific address */
1259 if (addr) {
1260 addr = PAGE_ALIGN(addr);
1261 vma = find_vma(mm, addr);
1262 if (TASK_SIZE - len >= addr &&
1263 (!vma || addr + len <= vma->vm_start))
1264 return addr;
1265 }
1266
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001267 /* check if free_area_cache is useful for us */
1268 if (len <= mm->cached_hole_size) {
1269 mm->cached_hole_size = 0;
1270 mm->free_area_cache = mm->mmap_base;
1271 }
1272
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 /* either no address requested or can't fit in requested address hole */
1274 addr = mm->free_area_cache;
1275
1276 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001277 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 vma = find_vma(mm, addr-len);
1279 if (!vma || addr <= vma->vm_start)
1280 /* remember the address as a hint for next time */
1281 return (mm->free_area_cache = addr-len);
1282 }
1283
Chris Wright73219d12005-06-21 17:14:52 -07001284 if (mm->mmap_base < len)
1285 goto bottomup;
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 addr = mm->mmap_base-len;
1288
1289 do {
1290 /*
1291 * Lookup failure means no vma is above this address,
1292 * else if new region fits below vma->vm_start,
1293 * return with success:
1294 */
1295 vma = find_vma(mm, addr);
1296 if (!vma || addr+len <= vma->vm_start)
1297 /* remember the address as a hint for next time */
1298 return (mm->free_area_cache = addr);
1299
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001300 /* remember the largest hole we saw so far */
1301 if (addr + mm->cached_hole_size < vma->vm_start)
1302 mm->cached_hole_size = vma->vm_start - addr;
1303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 /* try just below the current vma->vm_start */
1305 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001306 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Chris Wright73219d12005-06-21 17:14:52 -07001308bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 /*
1310 * A failed mmap() very likely causes application failure,
1311 * so fall back to the bottom-up function here. This scenario
1312 * can happen with large stack limits and large mmap()
1313 * allocations.
1314 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001315 mm->cached_hole_size = ~0UL;
1316 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1318 /*
1319 * Restore the topdown base:
1320 */
1321 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001322 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
1324 return addr;
1325}
1326#endif
1327
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001328void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
1330 /*
1331 * Is this a new hole at the highest possible address?
1332 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001333 if (addr > mm->free_area_cache)
1334 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
1336 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001337 if (mm->free_area_cache > mm->mmap_base)
1338 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339}
1340
1341unsigned long
1342get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1343 unsigned long pgoff, unsigned long flags)
1344{
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001345 unsigned long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001347 if (!(flags & MAP_FIXED)) {
1348 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1349
1350 get_area = current->mm->get_unmapped_area;
1351 if (file && file->f_op && file->f_op->get_unmapped_area)
1352 get_area = file->f_op->get_unmapped_area;
1353 addr = get_area(file, addr, len, pgoff, flags);
1354 if (IS_ERR_VALUE(addr))
1355 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
1357
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001358 if (addr > TASK_SIZE - len)
1359 return -ENOMEM;
1360 if (addr & ~PAGE_MASK)
1361 return -EINVAL;
1362 if (file && is_file_hugepages(file)) {
1363 /*
1364 * Check if the given range is hugepage aligned, and
1365 * can be made suitable for hugepages.
1366 */
1367 ret = prepare_hugepage_range(addr, len);
1368 } else {
1369 /*
1370 * Ensure that a normal request is not falling in a
1371 * reserved hugepage range. For some archs like IA-64,
1372 * there is a separate region for hugepages.
1373 */
1374 ret = is_hugepage_only_range(current->mm, addr, len);
1375 }
1376 if (ret)
1377 return -EINVAL;
1378 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379}
1380
1381EXPORT_SYMBOL(get_unmapped_area);
1382
1383/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1384struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1385{
1386 struct vm_area_struct *vma = NULL;
1387
1388 if (mm) {
1389 /* Check the cache first. */
1390 /* (Cache hit rate is typically around 35%.) */
1391 vma = mm->mmap_cache;
1392 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1393 struct rb_node * rb_node;
1394
1395 rb_node = mm->mm_rb.rb_node;
1396 vma = NULL;
1397
1398 while (rb_node) {
1399 struct vm_area_struct * vma_tmp;
1400
1401 vma_tmp = rb_entry(rb_node,
1402 struct vm_area_struct, vm_rb);
1403
1404 if (vma_tmp->vm_end > addr) {
1405 vma = vma_tmp;
1406 if (vma_tmp->vm_start <= addr)
1407 break;
1408 rb_node = rb_node->rb_left;
1409 } else
1410 rb_node = rb_node->rb_right;
1411 }
1412 if (vma)
1413 mm->mmap_cache = vma;
1414 }
1415 }
1416 return vma;
1417}
1418
1419EXPORT_SYMBOL(find_vma);
1420
1421/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1422struct vm_area_struct *
1423find_vma_prev(struct mm_struct *mm, unsigned long addr,
1424 struct vm_area_struct **pprev)
1425{
1426 struct vm_area_struct *vma = NULL, *prev = NULL;
1427 struct rb_node * rb_node;
1428 if (!mm)
1429 goto out;
1430
1431 /* Guard against addr being lower than the first VMA */
1432 vma = mm->mmap;
1433
1434 /* Go through the RB tree quickly. */
1435 rb_node = mm->mm_rb.rb_node;
1436
1437 while (rb_node) {
1438 struct vm_area_struct *vma_tmp;
1439 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1440
1441 if (addr < vma_tmp->vm_end) {
1442 rb_node = rb_node->rb_left;
1443 } else {
1444 prev = vma_tmp;
1445 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1446 break;
1447 rb_node = rb_node->rb_right;
1448 }
1449 }
1450
1451out:
1452 *pprev = prev;
1453 return prev ? prev->vm_next : vma;
1454}
1455
1456/*
1457 * Verify that the stack growth is acceptable and
1458 * update accounting. This is shared with both the
1459 * grow-up and grow-down cases.
1460 */
1461static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1462{
1463 struct mm_struct *mm = vma->vm_mm;
1464 struct rlimit *rlim = current->signal->rlim;
1465
1466 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001467 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return -ENOMEM;
1469
1470 /* Stack limit test */
1471 if (size > rlim[RLIMIT_STACK].rlim_cur)
1472 return -ENOMEM;
1473
1474 /* mlock limit tests */
1475 if (vma->vm_flags & VM_LOCKED) {
1476 unsigned long locked;
1477 unsigned long limit;
1478 locked = mm->locked_vm + grow;
1479 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1480 if (locked > limit && !capable(CAP_IPC_LOCK))
1481 return -ENOMEM;
1482 }
1483
1484 /*
1485 * Overcommit.. This must be the final test, as it will
1486 * update security statistics.
1487 */
1488 if (security_vm_enough_memory(grow))
1489 return -ENOMEM;
1490
1491 /* Ok, everything looks good - let it rip */
1492 mm->total_vm += grow;
1493 if (vma->vm_flags & VM_LOCKED)
1494 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001495 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 return 0;
1497}
1498
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001499#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001501 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1502 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001504#ifndef CONFIG_IA64
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001505static inline
1506#endif
1507int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
1509 int error;
1510
1511 if (!(vma->vm_flags & VM_GROWSUP))
1512 return -EFAULT;
1513
1514 /*
1515 * We must make sure the anon_vma is allocated
1516 * so that the anon_vma locking is not a noop.
1517 */
1518 if (unlikely(anon_vma_prepare(vma)))
1519 return -ENOMEM;
1520 anon_vma_lock(vma);
1521
1522 /*
1523 * vma->vm_start/vm_end cannot change under us because the caller
1524 * is required to hold the mmap_sem in read mode. We need the
1525 * anon_vma lock to serialize against concurrent expand_stacks.
1526 */
1527 address += 4 + PAGE_SIZE - 1;
1528 address &= PAGE_MASK;
1529 error = 0;
1530
1531 /* Somebody else might have raced and expanded it already */
1532 if (address > vma->vm_end) {
1533 unsigned long size, grow;
1534
1535 size = address - vma->vm_start;
1536 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1537
1538 error = acct_stack_growth(vma, size, grow);
1539 if (!error)
1540 vma->vm_end = address;
1541 }
1542 anon_vma_unlock(vma);
1543 return error;
1544}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001545#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1546
1547#ifdef CONFIG_STACK_GROWSUP
1548int expand_stack(struct vm_area_struct *vma, unsigned long address)
1549{
1550 return expand_upwards(vma, address);
1551}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
1553struct vm_area_struct *
1554find_extend_vma(struct mm_struct *mm, unsigned long addr)
1555{
1556 struct vm_area_struct *vma, *prev;
1557
1558 addr &= PAGE_MASK;
1559 vma = find_vma_prev(mm, addr, &prev);
1560 if (vma && (vma->vm_start <= addr))
1561 return vma;
1562 if (!prev || expand_stack(prev, addr))
1563 return NULL;
1564 if (prev->vm_flags & VM_LOCKED) {
1565 make_pages_present(addr, prev->vm_end);
1566 }
1567 return prev;
1568}
1569#else
1570/*
1571 * vma is the first one with address < vma->vm_start. Have to extend vma.
1572 */
1573int expand_stack(struct vm_area_struct *vma, unsigned long address)
1574{
1575 int error;
1576
1577 /*
1578 * We must make sure the anon_vma is allocated
1579 * so that the anon_vma locking is not a noop.
1580 */
1581 if (unlikely(anon_vma_prepare(vma)))
1582 return -ENOMEM;
1583 anon_vma_lock(vma);
1584
1585 /*
1586 * vma->vm_start/vm_end cannot change under us because the caller
1587 * is required to hold the mmap_sem in read mode. We need the
1588 * anon_vma lock to serialize against concurrent expand_stacks.
1589 */
1590 address &= PAGE_MASK;
1591 error = 0;
1592
1593 /* Somebody else might have raced and expanded it already */
1594 if (address < vma->vm_start) {
1595 unsigned long size, grow;
1596
1597 size = vma->vm_end - address;
1598 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1599
1600 error = acct_stack_growth(vma, size, grow);
1601 if (!error) {
1602 vma->vm_start = address;
1603 vma->vm_pgoff -= grow;
1604 }
1605 }
1606 anon_vma_unlock(vma);
1607 return error;
1608}
1609
1610struct vm_area_struct *
1611find_extend_vma(struct mm_struct * mm, unsigned long addr)
1612{
1613 struct vm_area_struct * vma;
1614 unsigned long start;
1615
1616 addr &= PAGE_MASK;
1617 vma = find_vma(mm,addr);
1618 if (!vma)
1619 return NULL;
1620 if (vma->vm_start <= addr)
1621 return vma;
1622 if (!(vma->vm_flags & VM_GROWSDOWN))
1623 return NULL;
1624 start = vma->vm_start;
1625 if (expand_stack(vma, addr))
1626 return NULL;
1627 if (vma->vm_flags & VM_LOCKED) {
1628 make_pages_present(addr, start);
1629 }
1630 return vma;
1631}
1632#endif
1633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001635 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001637 *
1638 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001640static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001642 /* Update high watermark before we lower total_vm */
1643 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001645 long nrpages = vma_pages(vma);
1646
1647 mm->total_vm -= nrpages;
1648 if (vma->vm_flags & VM_LOCKED)
1649 mm->locked_vm -= nrpages;
1650 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001651 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001652 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 validate_mm(mm);
1654}
1655
1656/*
1657 * Get rid of page table information in the indicated region.
1658 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001659 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 */
1661static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001662 struct vm_area_struct *vma, struct vm_area_struct *prev,
1663 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001665 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 struct mmu_gather *tlb;
1667 unsigned long nr_accounted = 0;
1668
1669 lru_add_drain();
1670 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001671 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001672 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001674 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001675 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 tlb_finish_mmu(tlb, start, end);
1677}
1678
1679/*
1680 * Create a list of vma's touched by the unmap, removing them from the mm's
1681 * vma list as we go..
1682 */
1683static void
1684detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1685 struct vm_area_struct *prev, unsigned long end)
1686{
1687 struct vm_area_struct **insertion_point;
1688 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001689 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1692 do {
1693 rb_erase(&vma->vm_rb, &mm->mm_rb);
1694 mm->map_count--;
1695 tail_vma = vma;
1696 vma = vma->vm_next;
1697 } while (vma && vma->vm_start < end);
1698 *insertion_point = vma;
1699 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001700 if (mm->unmap_area == arch_unmap_area)
1701 addr = prev ? prev->vm_end : mm->mmap_base;
1702 else
1703 addr = vma ? vma->vm_start : mm->mmap_base;
1704 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 mm->mmap_cache = NULL; /* Kill the cache. */
1706}
1707
1708/*
1709 * Split a vma into two pieces at address 'addr', a new vma is allocated
1710 * either for the first part or the the tail.
1711 */
1712int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1713 unsigned long addr, int new_below)
1714{
1715 struct mempolicy *pol;
1716 struct vm_area_struct *new;
1717
1718 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1719 return -EINVAL;
1720
1721 if (mm->map_count >= sysctl_max_map_count)
1722 return -ENOMEM;
1723
1724 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1725 if (!new)
1726 return -ENOMEM;
1727
1728 /* most fields are the same, copy all, and then fixup */
1729 *new = *vma;
1730
1731 if (new_below)
1732 new->vm_end = addr;
1733 else {
1734 new->vm_start = addr;
1735 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1736 }
1737
1738 pol = mpol_copy(vma_policy(vma));
1739 if (IS_ERR(pol)) {
1740 kmem_cache_free(vm_area_cachep, new);
1741 return PTR_ERR(pol);
1742 }
1743 vma_set_policy(new, pol);
1744
1745 if (new->vm_file)
1746 get_file(new->vm_file);
1747
1748 if (new->vm_ops && new->vm_ops->open)
1749 new->vm_ops->open(new);
1750
1751 if (new_below)
1752 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1753 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1754 else
1755 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1756
1757 return 0;
1758}
1759
1760/* Munmap is split into 2 main parts -- this part which finds
1761 * what needs doing, and the areas themselves, which do the
1762 * work. This now handles partial unmappings.
1763 * Jeremy Fitzhardinge <jeremy@goop.org>
1764 */
1765int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1766{
1767 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001768 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1771 return -EINVAL;
1772
1773 if ((len = PAGE_ALIGN(len)) == 0)
1774 return -EINVAL;
1775
1776 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001777 vma = find_vma_prev(mm, start, &prev);
1778 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001780 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
1782 /* if it doesn't overlap, we have nothing.. */
1783 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001784 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return 0;
1786
1787 /*
1788 * If we need to split any vma, do it now to save pain later.
1789 *
1790 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1791 * unmapped vm_area_struct will remain in use: so lower split_vma
1792 * places tmp vma above, and higher split_vma places tmp vma below.
1793 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001794 if (start > vma->vm_start) {
1795 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 if (error)
1797 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001798 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 }
1800
1801 /* Does it split the last one? */
1802 last = find_vma(mm, end);
1803 if (last && end > last->vm_start) {
1804 int error = split_vma(mm, last, end, 1);
1805 if (error)
1806 return error;
1807 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001808 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
1810 /*
1811 * Remove the vma's, and unmap the actual pages
1812 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001813 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1814 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
1816 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001817 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
1819 return 0;
1820}
1821
1822EXPORT_SYMBOL(do_munmap);
1823
1824asmlinkage long sys_munmap(unsigned long addr, size_t len)
1825{
1826 int ret;
1827 struct mm_struct *mm = current->mm;
1828
1829 profile_munmap(addr);
1830
1831 down_write(&mm->mmap_sem);
1832 ret = do_munmap(mm, addr, len);
1833 up_write(&mm->mmap_sem);
1834 return ret;
1835}
1836
1837static inline void verify_mm_writelocked(struct mm_struct *mm)
1838{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001839#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1841 WARN_ON(1);
1842 up_read(&mm->mmap_sem);
1843 }
1844#endif
1845}
1846
1847/*
1848 * this is really a simplified "do_mmap". it only handles
1849 * anonymous maps. eventually we may be able to do some
1850 * brk-specific accounting here.
1851 */
1852unsigned long do_brk(unsigned long addr, unsigned long len)
1853{
1854 struct mm_struct * mm = current->mm;
1855 struct vm_area_struct * vma, * prev;
1856 unsigned long flags;
1857 struct rb_node ** rb_link, * rb_parent;
1858 pgoff_t pgoff = addr >> PAGE_SHIFT;
1859
1860 len = PAGE_ALIGN(len);
1861 if (!len)
1862 return addr;
1863
1864 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1865 return -EINVAL;
1866
1867 /*
1868 * mlock MCL_FUTURE?
1869 */
1870 if (mm->def_flags & VM_LOCKED) {
1871 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001872 locked = len >> PAGE_SHIFT;
1873 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07001875 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1877 return -EAGAIN;
1878 }
1879
1880 /*
1881 * mm->mmap_sem is required to protect against another thread
1882 * changing the mappings in case we sleep.
1883 */
1884 verify_mm_writelocked(mm);
1885
1886 /*
1887 * Clear old maps. this also does some error checking for us
1888 */
1889 munmap_back:
1890 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1891 if (vma && vma->vm_start < addr + len) {
1892 if (do_munmap(mm, addr, len))
1893 return -ENOMEM;
1894 goto munmap_back;
1895 }
1896
1897 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001898 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 return -ENOMEM;
1900
1901 if (mm->map_count > sysctl_max_map_count)
1902 return -ENOMEM;
1903
1904 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1905 return -ENOMEM;
1906
1907 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1908
1909 /* Can we just expand an old private anonymous mapping? */
1910 if (vma_merge(mm, prev, addr, addr + len, flags,
1911 NULL, NULL, pgoff, NULL))
1912 goto out;
1913
1914 /*
1915 * create a vma struct for an anonymous mapping
1916 */
1917 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1918 if (!vma) {
1919 vm_unacct_memory(len >> PAGE_SHIFT);
1920 return -ENOMEM;
1921 }
1922 memset(vma, 0, sizeof(*vma));
1923
1924 vma->vm_mm = mm;
1925 vma->vm_start = addr;
1926 vma->vm_end = addr + len;
1927 vma->vm_pgoff = pgoff;
1928 vma->vm_flags = flags;
1929 vma->vm_page_prot = protection_map[flags & 0x0f];
1930 vma_link(mm, vma, prev, rb_link, rb_parent);
1931out:
1932 mm->total_vm += len >> PAGE_SHIFT;
1933 if (flags & VM_LOCKED) {
1934 mm->locked_vm += len >> PAGE_SHIFT;
1935 make_pages_present(addr, addr + len);
1936 }
1937 return addr;
1938}
1939
1940EXPORT_SYMBOL(do_brk);
1941
1942/* Release all mmaps. */
1943void exit_mmap(struct mm_struct *mm)
1944{
1945 struct mmu_gather *tlb;
Hugh Dickinse0da3822005-04-19 13:29:15 -07001946 struct vm_area_struct *vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001948 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
1950 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001952 tlb = tlb_gather_mmu(mm, 1);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001953 /* Don't update_hiwater_rss(mm) here, do_exit already did */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001954 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07001955 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001957 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001958 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07001961 * Walk the list again, actually closing and freeing it,
1962 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001964 while (vma)
1965 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001966
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001967 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968}
1969
1970/* Insert vm structure into process list sorted by address
1971 * and into the inode's i_mmap tree. If vm_file is non-NULL
1972 * then i_mmap_lock is taken here.
1973 */
1974int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1975{
1976 struct vm_area_struct * __vma, * prev;
1977 struct rb_node ** rb_link, * rb_parent;
1978
1979 /*
1980 * The vm_pgoff of a purely anonymous vma should be irrelevant
1981 * until its first write fault, when page's anon_vma and index
1982 * are set. But now set the vm_pgoff it will almost certainly
1983 * end up with (unless mremap moves it elsewhere before that
1984 * first wfault), so /proc/pid/maps tells a consistent story.
1985 *
1986 * By setting it to reflect the virtual start address of the
1987 * vma, merges and splits can happen in a seamless way, just
1988 * using the existing file pgoff checks and manipulations.
1989 * Similarly in do_mmap_pgoff and in do_brk.
1990 */
1991 if (!vma->vm_file) {
1992 BUG_ON(vma->anon_vma);
1993 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1994 }
1995 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1996 if (__vma && __vma->vm_start < vma->vm_end)
1997 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01001998 if ((vma->vm_flags & VM_ACCOUNT) &&
1999 security_vm_enough_memory(vma_pages(vma)))
2000 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 vma_link(mm, vma, prev, rb_link, rb_parent);
2002 return 0;
2003}
2004
2005/*
2006 * Copy the vma structure to a new location in the same mm,
2007 * prior to moving page table entries, to effect an mremap move.
2008 */
2009struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2010 unsigned long addr, unsigned long len, pgoff_t pgoff)
2011{
2012 struct vm_area_struct *vma = *vmap;
2013 unsigned long vma_start = vma->vm_start;
2014 struct mm_struct *mm = vma->vm_mm;
2015 struct vm_area_struct *new_vma, *prev;
2016 struct rb_node **rb_link, *rb_parent;
2017 struct mempolicy *pol;
2018
2019 /*
2020 * If anonymous vma has not yet been faulted, update new pgoff
2021 * to match new location, to increase its chance of merging.
2022 */
2023 if (!vma->vm_file && !vma->anon_vma)
2024 pgoff = addr >> PAGE_SHIFT;
2025
2026 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2027 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2028 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2029 if (new_vma) {
2030 /*
2031 * Source vma may have been merged into new_vma
2032 */
2033 if (vma_start >= new_vma->vm_start &&
2034 vma_start < new_vma->vm_end)
2035 *vmap = new_vma;
2036 } else {
2037 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2038 if (new_vma) {
2039 *new_vma = *vma;
2040 pol = mpol_copy(vma_policy(vma));
2041 if (IS_ERR(pol)) {
2042 kmem_cache_free(vm_area_cachep, new_vma);
2043 return NULL;
2044 }
2045 vma_set_policy(new_vma, pol);
2046 new_vma->vm_start = addr;
2047 new_vma->vm_end = addr + len;
2048 new_vma->vm_pgoff = pgoff;
2049 if (new_vma->vm_file)
2050 get_file(new_vma->vm_file);
2051 if (new_vma->vm_ops && new_vma->vm_ops->open)
2052 new_vma->vm_ops->open(new_vma);
2053 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2054 }
2055 }
2056 return new_vma;
2057}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002058
2059/*
2060 * Return true if the calling process may expand its vm space by the passed
2061 * number of pages
2062 */
2063int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2064{
2065 unsigned long cur = mm->total_vm; /* pages */
2066 unsigned long lim;
2067
2068 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2069
2070 if (cur + npages > lim)
2071 return 0;
2072 return 1;
2073}