blob: d799d896d74aae5f19c604928eb6a52738676d66 [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>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080016#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/init.h>
18#include <linux/file.h>
19#include <linux/fs.h>
20#include <linux/personality.h>
21#include <linux/security.h>
22#include <linux/hugetlb.h>
23#include <linux/profile.h>
24#include <linux/module.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
Kirill Korotaev3a459752006-09-07 14:17:04 +040033#ifndef arch_mmap_check
34#define arch_mmap_check(addr, len, flags) (0)
35#endif
36
Hugh Dickinse0da3822005-04-19 13:29:15 -070037static void unmap_region(struct mm_struct *mm,
38 struct vm_area_struct *vma, struct vm_area_struct *prev,
39 unsigned long start, unsigned long end);
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/*
42 * WARNING: the debugging will use recursive algorithms so never enable this
43 * unless you know what you are doing.
44 */
45#undef DEBUG_MM_RB
46
47/* description of effects of mapping type and prot in current implementation.
48 * this is due to the limited x86 page protection hardware. The expected
49 * behavior is in parens:
50 *
51 * map_type prot
52 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
53 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
54 * w: (no) no w: (no) no w: (yes) yes w: (no) no
55 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
56 *
57 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
58 * w: (no) no w: (no) no w: (copy) copy w: (no) no
59 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
60 *
61 */
62pgprot_t protection_map[16] = {
63 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
64 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
65};
66
Hugh Dickins804af2c2006-07-26 21:39:49 +010067pgprot_t vm_get_page_prot(unsigned long vm_flags)
68{
69 return protection_map[vm_flags &
70 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
71}
72EXPORT_SYMBOL(vm_get_page_prot);
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
75int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070076int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077atomic_t vm_committed_space = ATOMIC_INIT(0);
78
79/*
80 * Check that a process has enough memory to allocate a new virtual
81 * mapping. 0 means there is enough memory for the allocation to
82 * succeed and -ENOMEM implies there is not.
83 *
84 * We currently support three overcommit policies, which are set via the
85 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
86 *
87 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
88 * Additional code 2002 Jul 20 by Robert Love.
89 *
90 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
91 *
92 * Note this is a helper function intended to be used by LSMs which
93 * wish to use this logic.
94 */
95int __vm_enough_memory(long pages, int cap_sys_admin)
96{
97 unsigned long free, allowed;
98
99 vm_acct_memory(pages);
100
101 /*
102 * Sometimes we want to use more memory than we have
103 */
104 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
105 return 0;
106
107 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
108 unsigned long n;
109
Christoph Lameter347ce432006-06-30 01:55:35 -0700110 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 free += nr_swap_pages;
112
113 /*
114 * Any slabs which are created with the
115 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
116 * which are reclaimable, under pressure. The dentry
117 * cache and most inode caches should fall into this
118 */
119 free += atomic_read(&slab_reclaim_pages);
120
121 /*
122 * Leave the last 3% for root
123 */
124 if (!cap_sys_admin)
125 free -= free / 32;
126
127 if (free > pages)
128 return 0;
129
130 /*
131 * nr_free_pages() is very expensive on large systems,
132 * only call if we're about to fail.
133 */
134 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700135
136 /*
137 * Leave reserved pages. The pages are not for anonymous pages.
138 */
139 if (n <= totalreserve_pages)
140 goto error;
141 else
142 n -= totalreserve_pages;
143
144 /*
145 * Leave the last 3% for root
146 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 if (!cap_sys_admin)
148 n -= n / 32;
149 free += n;
150
151 if (free > pages)
152 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700153
154 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 }
156
157 allowed = (totalram_pages - hugetlb_total_pages())
158 * sysctl_overcommit_ratio / 100;
159 /*
160 * Leave the last 3% for root
161 */
162 if (!cap_sys_admin)
163 allowed -= allowed / 32;
164 allowed += total_swap_pages;
165
166 /* Don't let a single process grow too big:
167 leave 3% of the size of this process for other processes */
168 allowed -= current->mm->total_vm / 32;
169
Simon Derr2f60f8d2005-08-04 19:52:03 -0700170 /*
171 * cast `allowed' as a signed long because vm_committed_space
172 * sometimes has a negative value
173 */
174 if (atomic_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700176error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 vm_unacct_memory(pages);
178
179 return -ENOMEM;
180}
181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182EXPORT_SYMBOL(__vm_enough_memory);
183
184/*
185 * Requires inode->i_mapping->i_mmap_lock
186 */
187static void __remove_shared_vm_struct(struct vm_area_struct *vma,
188 struct file *file, struct address_space *mapping)
189{
190 if (vma->vm_flags & VM_DENYWRITE)
191 atomic_inc(&file->f_dentry->d_inode->i_writecount);
192 if (vma->vm_flags & VM_SHARED)
193 mapping->i_mmap_writable--;
194
195 flush_dcache_mmap_lock(mapping);
196 if (unlikely(vma->vm_flags & VM_NONLINEAR))
197 list_del_init(&vma->shared.vm_set.list);
198 else
199 vma_prio_tree_remove(vma, &mapping->i_mmap);
200 flush_dcache_mmap_unlock(mapping);
201}
202
203/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700204 * Unlink a file-based vm structure from its prio_tree, to hide
205 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700207void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208{
209 struct file *file = vma->vm_file;
210
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 if (file) {
212 struct address_space *mapping = file->f_mapping;
213 spin_lock(&mapping->i_mmap_lock);
214 __remove_shared_vm_struct(vma, file, mapping);
215 spin_unlock(&mapping->i_mmap_lock);
216 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700217}
218
219/*
220 * Close a vm structure and free it, returning the next.
221 */
222static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
223{
224 struct vm_area_struct *next = vma->vm_next;
225
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700226 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 if (vma->vm_ops && vma->vm_ops->close)
228 vma->vm_ops->close(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700229 if (vma->vm_file)
230 fput(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 mpol_free(vma_policy(vma));
232 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700233 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236asmlinkage unsigned long sys_brk(unsigned long brk)
237{
238 unsigned long rlim, retval;
239 unsigned long newbrk, oldbrk;
240 struct mm_struct *mm = current->mm;
241
242 down_write(&mm->mmap_sem);
243
244 if (brk < mm->end_code)
245 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700246
247 /*
248 * Check against rlimit here. If this check is done later after the test
249 * of oldbrk with newbrk then it can escape the test and let the data
250 * segment grow beyond its set limit the in case where the limit is
251 * not page aligned -Ram Gupta
252 */
253 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
254 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
255 goto out;
256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 newbrk = PAGE_ALIGN(brk);
258 oldbrk = PAGE_ALIGN(mm->brk);
259 if (oldbrk == newbrk)
260 goto set_brk;
261
262 /* Always allow shrinking brk. */
263 if (brk <= mm->brk) {
264 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
265 goto set_brk;
266 goto out;
267 }
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 /* Check against existing mmap mappings. */
270 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
271 goto out;
272
273 /* Ok, looks good - let it rip. */
274 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
275 goto out;
276set_brk:
277 mm->brk = brk;
278out:
279 retval = mm->brk;
280 up_write(&mm->mmap_sem);
281 return retval;
282}
283
284#ifdef DEBUG_MM_RB
285static int browse_rb(struct rb_root *root)
286{
287 int i = 0, j;
288 struct rb_node *nd, *pn = NULL;
289 unsigned long prev = 0, pend = 0;
290
291 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
292 struct vm_area_struct *vma;
293 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
294 if (vma->vm_start < prev)
295 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
296 if (vma->vm_start < pend)
297 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
298 if (vma->vm_start > vma->vm_end)
299 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
300 i++;
301 pn = nd;
302 }
303 j = 0;
304 for (nd = pn; nd; nd = rb_prev(nd)) {
305 j++;
306 }
307 if (i != j)
308 printk("backwards %d, forwards %d\n", j, i), i = 0;
309 return i;
310}
311
312void validate_mm(struct mm_struct *mm)
313{
314 int bug = 0;
315 int i = 0;
316 struct vm_area_struct *tmp = mm->mmap;
317 while (tmp) {
318 tmp = tmp->vm_next;
319 i++;
320 }
321 if (i != mm->map_count)
322 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
323 i = browse_rb(&mm->mm_rb);
324 if (i != mm->map_count)
325 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200326 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328#else
329#define validate_mm(mm) do { } while (0)
330#endif
331
332static struct vm_area_struct *
333find_vma_prepare(struct mm_struct *mm, unsigned long addr,
334 struct vm_area_struct **pprev, struct rb_node ***rb_link,
335 struct rb_node ** rb_parent)
336{
337 struct vm_area_struct * vma;
338 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
339
340 __rb_link = &mm->mm_rb.rb_node;
341 rb_prev = __rb_parent = NULL;
342 vma = NULL;
343
344 while (*__rb_link) {
345 struct vm_area_struct *vma_tmp;
346
347 __rb_parent = *__rb_link;
348 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
349
350 if (vma_tmp->vm_end > addr) {
351 vma = vma_tmp;
352 if (vma_tmp->vm_start <= addr)
353 return vma;
354 __rb_link = &__rb_parent->rb_left;
355 } else {
356 rb_prev = __rb_parent;
357 __rb_link = &__rb_parent->rb_right;
358 }
359 }
360
361 *pprev = NULL;
362 if (rb_prev)
363 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
364 *rb_link = __rb_link;
365 *rb_parent = __rb_parent;
366 return vma;
367}
368
369static inline void
370__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
371 struct vm_area_struct *prev, struct rb_node *rb_parent)
372{
373 if (prev) {
374 vma->vm_next = prev->vm_next;
375 prev->vm_next = vma;
376 } else {
377 mm->mmap = vma;
378 if (rb_parent)
379 vma->vm_next = rb_entry(rb_parent,
380 struct vm_area_struct, vm_rb);
381 else
382 vma->vm_next = NULL;
383 }
384}
385
386void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
387 struct rb_node **rb_link, struct rb_node *rb_parent)
388{
389 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
390 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
391}
392
393static inline void __vma_link_file(struct vm_area_struct *vma)
394{
395 struct file * file;
396
397 file = vma->vm_file;
398 if (file) {
399 struct address_space *mapping = file->f_mapping;
400
401 if (vma->vm_flags & VM_DENYWRITE)
402 atomic_dec(&file->f_dentry->d_inode->i_writecount);
403 if (vma->vm_flags & VM_SHARED)
404 mapping->i_mmap_writable++;
405
406 flush_dcache_mmap_lock(mapping);
407 if (unlikely(vma->vm_flags & VM_NONLINEAR))
408 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
409 else
410 vma_prio_tree_insert(vma, &mapping->i_mmap);
411 flush_dcache_mmap_unlock(mapping);
412 }
413}
414
415static void
416__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
417 struct vm_area_struct *prev, struct rb_node **rb_link,
418 struct rb_node *rb_parent)
419{
420 __vma_link_list(mm, vma, prev, rb_parent);
421 __vma_link_rb(mm, vma, rb_link, rb_parent);
422 __anon_vma_link(vma);
423}
424
425static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
426 struct vm_area_struct *prev, struct rb_node **rb_link,
427 struct rb_node *rb_parent)
428{
429 struct address_space *mapping = NULL;
430
431 if (vma->vm_file)
432 mapping = vma->vm_file->f_mapping;
433
434 if (mapping) {
435 spin_lock(&mapping->i_mmap_lock);
436 vma->vm_truncate_count = mapping->truncate_count;
437 }
438 anon_vma_lock(vma);
439
440 __vma_link(mm, vma, prev, rb_link, rb_parent);
441 __vma_link_file(vma);
442
443 anon_vma_unlock(vma);
444 if (mapping)
445 spin_unlock(&mapping->i_mmap_lock);
446
447 mm->map_count++;
448 validate_mm(mm);
449}
450
451/*
452 * Helper for vma_adjust in the split_vma insert case:
453 * insert vm structure into list and rbtree and anon_vma,
454 * but it has already been inserted into prio_tree earlier.
455 */
456static void
457__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
458{
459 struct vm_area_struct * __vma, * prev;
460 struct rb_node ** rb_link, * rb_parent;
461
462 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200463 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 __vma_link(mm, vma, prev, rb_link, rb_parent);
465 mm->map_count++;
466}
467
468static inline void
469__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
470 struct vm_area_struct *prev)
471{
472 prev->vm_next = vma->vm_next;
473 rb_erase(&vma->vm_rb, &mm->mm_rb);
474 if (mm->mmap_cache == vma)
475 mm->mmap_cache = prev;
476}
477
478/*
479 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
480 * is already present in an i_mmap tree without adjusting the tree.
481 * The following helper function should be used when such adjustments
482 * are necessary. The "insert" vma (if any) is to be inserted
483 * before we drop the necessary locks.
484 */
485void vma_adjust(struct vm_area_struct *vma, unsigned long start,
486 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
487{
488 struct mm_struct *mm = vma->vm_mm;
489 struct vm_area_struct *next = vma->vm_next;
490 struct vm_area_struct *importer = NULL;
491 struct address_space *mapping = NULL;
492 struct prio_tree_root *root = NULL;
493 struct file *file = vma->vm_file;
494 struct anon_vma *anon_vma = NULL;
495 long adjust_next = 0;
496 int remove_next = 0;
497
498 if (next && !insert) {
499 if (end >= next->vm_end) {
500 /*
501 * vma expands, overlapping all the next, and
502 * perhaps the one after too (mprotect case 6).
503 */
504again: remove_next = 1 + (end > next->vm_end);
505 end = next->vm_end;
506 anon_vma = next->anon_vma;
507 importer = vma;
508 } else if (end > next->vm_start) {
509 /*
510 * vma expands, overlapping part of the next:
511 * mprotect case 5 shifting the boundary up.
512 */
513 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
514 anon_vma = next->anon_vma;
515 importer = vma;
516 } else if (end < vma->vm_end) {
517 /*
518 * vma shrinks, and !insert tells it's not
519 * split_vma inserting another: so it must be
520 * mprotect case 4 shifting the boundary down.
521 */
522 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
523 anon_vma = next->anon_vma;
524 importer = next;
525 }
526 }
527
528 if (file) {
529 mapping = file->f_mapping;
530 if (!(vma->vm_flags & VM_NONLINEAR))
531 root = &mapping->i_mmap;
532 spin_lock(&mapping->i_mmap_lock);
533 if (importer &&
534 vma->vm_truncate_count != next->vm_truncate_count) {
535 /*
536 * unmap_mapping_range might be in progress:
537 * ensure that the expanding vma is rescanned.
538 */
539 importer->vm_truncate_count = 0;
540 }
541 if (insert) {
542 insert->vm_truncate_count = vma->vm_truncate_count;
543 /*
544 * Put into prio_tree now, so instantiated pages
545 * are visible to arm/parisc __flush_dcache_page
546 * throughout; but we cannot insert into address
547 * space until vma start or end is updated.
548 */
549 __vma_link_file(insert);
550 }
551 }
552
553 /*
554 * When changing only vma->vm_end, we don't really need
555 * anon_vma lock: but is that case worth optimizing out?
556 */
557 if (vma->anon_vma)
558 anon_vma = vma->anon_vma;
559 if (anon_vma) {
560 spin_lock(&anon_vma->lock);
561 /*
562 * Easily overlooked: when mprotect shifts the boundary,
563 * make sure the expanding vma has anon_vma set if the
564 * shrinking vma had, to cover any anon pages imported.
565 */
566 if (importer && !importer->anon_vma) {
567 importer->anon_vma = anon_vma;
568 __anon_vma_link(importer);
569 }
570 }
571
572 if (root) {
573 flush_dcache_mmap_lock(mapping);
574 vma_prio_tree_remove(vma, root);
575 if (adjust_next)
576 vma_prio_tree_remove(next, root);
577 }
578
579 vma->vm_start = start;
580 vma->vm_end = end;
581 vma->vm_pgoff = pgoff;
582 if (adjust_next) {
583 next->vm_start += adjust_next << PAGE_SHIFT;
584 next->vm_pgoff += adjust_next;
585 }
586
587 if (root) {
588 if (adjust_next)
589 vma_prio_tree_insert(next, root);
590 vma_prio_tree_insert(vma, root);
591 flush_dcache_mmap_unlock(mapping);
592 }
593
594 if (remove_next) {
595 /*
596 * vma_merge has merged next into vma, and needs
597 * us to remove next before dropping the locks.
598 */
599 __vma_unlink(mm, next, vma);
600 if (file)
601 __remove_shared_vm_struct(next, file, mapping);
602 if (next->anon_vma)
603 __anon_vma_merge(vma, next);
604 } else if (insert) {
605 /*
606 * split_vma has split insert from vma, and needs
607 * us to insert it before dropping the locks
608 * (it may either follow vma or precede it).
609 */
610 __insert_vm_struct(mm, insert);
611 }
612
613 if (anon_vma)
614 spin_unlock(&anon_vma->lock);
615 if (mapping)
616 spin_unlock(&mapping->i_mmap_lock);
617
618 if (remove_next) {
619 if (file)
620 fput(file);
621 mm->map_count--;
622 mpol_free(vma_policy(next));
623 kmem_cache_free(vm_area_cachep, next);
624 /*
625 * In mprotect's case 6 (see comments on vma_merge),
626 * we must remove another next too. It would clutter
627 * up the code too much to do both in one go.
628 */
629 if (remove_next == 2) {
630 next = vma->vm_next;
631 goto again;
632 }
633 }
634
635 validate_mm(mm);
636}
637
638/*
639 * If the vma has a ->close operation then the driver probably needs to release
640 * per-vma resources, so we don't attempt to merge those.
641 */
Hugh Dickinsa6f563d2006-03-22 00:08:29 -0800642#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
644static inline int is_mergeable_vma(struct vm_area_struct *vma,
645 struct file *file, unsigned long vm_flags)
646{
647 if (vma->vm_flags != vm_flags)
648 return 0;
649 if (vma->vm_file != file)
650 return 0;
651 if (vma->vm_ops && vma->vm_ops->close)
652 return 0;
653 return 1;
654}
655
656static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
657 struct anon_vma *anon_vma2)
658{
659 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
660}
661
662/*
663 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
664 * in front of (at a lower virtual address and file offset than) the vma.
665 *
666 * We cannot merge two vmas if they have differently assigned (non-NULL)
667 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
668 *
669 * We don't check here for the merged mmap wrapping around the end of pagecache
670 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
671 * wrap, nor mmaps which cover the final page at index -1UL.
672 */
673static int
674can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
675 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
676{
677 if (is_mergeable_vma(vma, file, vm_flags) &&
678 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
679 if (vma->vm_pgoff == vm_pgoff)
680 return 1;
681 }
682 return 0;
683}
684
685/*
686 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
687 * beyond (at a higher virtual address and file offset than) the vma.
688 *
689 * We cannot merge two vmas if they have differently assigned (non-NULL)
690 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
691 */
692static int
693can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
694 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
695{
696 if (is_mergeable_vma(vma, file, vm_flags) &&
697 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
698 pgoff_t vm_pglen;
699 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
700 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
701 return 1;
702 }
703 return 0;
704}
705
706/*
707 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
708 * whether that can be merged with its predecessor or its successor.
709 * Or both (it neatly fills a hole).
710 *
711 * In most cases - when called for mmap, brk or mremap - [addr,end) is
712 * certain not to be mapped by the time vma_merge is called; but when
713 * called for mprotect, it is certain to be already mapped (either at
714 * an offset within prev, or at the start of next), and the flags of
715 * this area are about to be changed to vm_flags - and the no-change
716 * case has already been eliminated.
717 *
718 * The following mprotect cases have to be considered, where AAAA is
719 * the area passed down from mprotect_fixup, never extending beyond one
720 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
721 *
722 * AAAA AAAA AAAA AAAA
723 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
724 * cannot merge might become might become might become
725 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
726 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
727 * mremap move: PPPPNNNNNNNN 8
728 * AAAA
729 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
730 * might become case 1 below case 2 below case 3 below
731 *
732 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
733 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
734 */
735struct vm_area_struct *vma_merge(struct mm_struct *mm,
736 struct vm_area_struct *prev, unsigned long addr,
737 unsigned long end, unsigned long vm_flags,
738 struct anon_vma *anon_vma, struct file *file,
739 pgoff_t pgoff, struct mempolicy *policy)
740{
741 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
742 struct vm_area_struct *area, *next;
743
744 /*
745 * We later require that vma->vm_flags == vm_flags,
746 * so this tests vma->vm_flags & VM_SPECIAL, too.
747 */
748 if (vm_flags & VM_SPECIAL)
749 return NULL;
750
751 if (prev)
752 next = prev->vm_next;
753 else
754 next = mm->mmap;
755 area = next;
756 if (next && next->vm_end == end) /* cases 6, 7, 8 */
757 next = next->vm_next;
758
759 /*
760 * Can it merge with the predecessor?
761 */
762 if (prev && prev->vm_end == addr &&
763 mpol_equal(vma_policy(prev), policy) &&
764 can_vma_merge_after(prev, vm_flags,
765 anon_vma, file, pgoff)) {
766 /*
767 * OK, it can. Can we now merge in the successor as well?
768 */
769 if (next && end == next->vm_start &&
770 mpol_equal(policy, vma_policy(next)) &&
771 can_vma_merge_before(next, vm_flags,
772 anon_vma, file, pgoff+pglen) &&
773 is_mergeable_anon_vma(prev->anon_vma,
774 next->anon_vma)) {
775 /* cases 1, 6 */
776 vma_adjust(prev, prev->vm_start,
777 next->vm_end, prev->vm_pgoff, NULL);
778 } else /* cases 2, 5, 7 */
779 vma_adjust(prev, prev->vm_start,
780 end, prev->vm_pgoff, NULL);
781 return prev;
782 }
783
784 /*
785 * Can this new request be merged in front of next?
786 */
787 if (next && end == next->vm_start &&
788 mpol_equal(policy, vma_policy(next)) &&
789 can_vma_merge_before(next, vm_flags,
790 anon_vma, file, pgoff+pglen)) {
791 if (prev && addr < prev->vm_end) /* case 4 */
792 vma_adjust(prev, prev->vm_start,
793 addr, prev->vm_pgoff, NULL);
794 else /* cases 3, 8 */
795 vma_adjust(area, addr, next->vm_end,
796 next->vm_pgoff - pglen, NULL);
797 return area;
798 }
799
800 return NULL;
801}
802
803/*
804 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
805 * neighbouring vmas for a suitable anon_vma, before it goes off
806 * to allocate a new anon_vma. It checks because a repetitive
807 * sequence of mprotects and faults may otherwise lead to distinct
808 * anon_vmas being allocated, preventing vma merge in subsequent
809 * mprotect.
810 */
811struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
812{
813 struct vm_area_struct *near;
814 unsigned long vm_flags;
815
816 near = vma->vm_next;
817 if (!near)
818 goto try_prev;
819
820 /*
821 * Since only mprotect tries to remerge vmas, match flags
822 * which might be mprotected into each other later on.
823 * Neither mlock nor madvise tries to remerge at present,
824 * so leave their flags as obstructing a merge.
825 */
826 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
827 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
828
829 if (near->anon_vma && vma->vm_end == near->vm_start &&
830 mpol_equal(vma_policy(vma), vma_policy(near)) &&
831 can_vma_merge_before(near, vm_flags,
832 NULL, vma->vm_file, vma->vm_pgoff +
833 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
834 return near->anon_vma;
835try_prev:
836 /*
837 * It is potentially slow to have to call find_vma_prev here.
838 * But it's only on the first write fault on the vma, not
839 * every time, and we could devise a way to avoid it later
840 * (e.g. stash info in next's anon_vma_node when assigning
841 * an anon_vma, or when trying vma_merge). Another time.
842 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200843 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (!near)
845 goto none;
846
847 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
848 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
849
850 if (near->anon_vma && near->vm_end == vma->vm_start &&
851 mpol_equal(vma_policy(near), vma_policy(vma)) &&
852 can_vma_merge_after(near, vm_flags,
853 NULL, vma->vm_file, vma->vm_pgoff))
854 return near->anon_vma;
855none:
856 /*
857 * There's no absolute need to look only at touching neighbours:
858 * we could search further afield for "compatible" anon_vmas.
859 * But it would probably just be a waste of time searching,
860 * or lead to too many vmas hanging off the same anon_vma.
861 * We're trying to allow mprotect remerging later on,
862 * not trying to minimize memory used for anon_vmas.
863 */
864 return NULL;
865}
866
867#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700868void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 struct file *file, long pages)
870{
871 const unsigned long stack_flags
872 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 if (file) {
875 mm->shared_vm += pages;
876 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
877 mm->exec_vm += pages;
878 } else if (flags & stack_flags)
879 mm->stack_vm += pages;
880 if (flags & (VM_RESERVED|VM_IO))
881 mm->reserved_vm += pages;
882}
883#endif /* CONFIG_PROC_FS */
884
885/*
886 * The caller must hold down_write(current->mm->mmap_sem).
887 */
888
889unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
890 unsigned long len, unsigned long prot,
891 unsigned long flags, unsigned long pgoff)
892{
893 struct mm_struct * mm = current->mm;
894 struct vm_area_struct * vma, * prev;
895 struct inode *inode;
896 unsigned int vm_flags;
897 int correct_wcount = 0;
898 int error;
899 struct rb_node ** rb_link, * rb_parent;
900 int accountable = 1;
901 unsigned long charged = 0, reqprot = prot;
902
903 if (file) {
904 if (is_file_hugepages(file))
905 accountable = 0;
906
907 if (!file->f_op || !file->f_op->mmap)
908 return -ENODEV;
909
910 if ((prot & PROT_EXEC) &&
911 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
912 return -EPERM;
913 }
914 /*
915 * Does the application expect PROT_READ to imply PROT_EXEC?
916 *
917 * (the exception is when the underlying filesystem is noexec
918 * mounted, in which case we dont add PROT_EXEC.)
919 */
920 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
921 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
922 prot |= PROT_EXEC;
923
924 if (!len)
925 return -EINVAL;
926
Kirill Korotaev3a459752006-09-07 14:17:04 +0400927 error = arch_mmap_check(addr, len, flags);
928 if (error)
929 return error;
930
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 /* Careful about overflows.. */
932 len = PAGE_ALIGN(len);
933 if (!len || len > TASK_SIZE)
934 return -ENOMEM;
935
936 /* offset overflow? */
937 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
938 return -EOVERFLOW;
939
940 /* Too many mappings? */
941 if (mm->map_count > sysctl_max_map_count)
942 return -ENOMEM;
943
944 /* Obtain the address to map to. we verify (or select) it and ensure
945 * that it represents a valid section of the address space.
946 */
947 addr = get_unmapped_area(file, addr, len, pgoff, flags);
948 if (addr & ~PAGE_MASK)
949 return addr;
950
951 /* Do simple checking here so the lower-level routines won't have
952 * to. we assume access permissions have been handled by the open
953 * of the memory object, so we don't do any here.
954 */
955 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
956 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
957
958 if (flags & MAP_LOCKED) {
959 if (!can_do_mlock())
960 return -EPERM;
961 vm_flags |= VM_LOCKED;
962 }
963 /* mlock MCL_FUTURE? */
964 if (vm_flags & VM_LOCKED) {
965 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700966 locked = len >> PAGE_SHIFT;
967 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700969 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
971 return -EAGAIN;
972 }
973
974 inode = file ? file->f_dentry->d_inode : NULL;
975
976 if (file) {
977 switch (flags & MAP_TYPE) {
978 case MAP_SHARED:
979 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
980 return -EACCES;
981
982 /*
983 * Make sure we don't allow writing to an append-only
984 * file..
985 */
986 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
987 return -EACCES;
988
989 /*
990 * Make sure there are no mandatory locks on the file.
991 */
992 if (locks_verify_locked(inode))
993 return -EAGAIN;
994
995 vm_flags |= VM_SHARED | VM_MAYSHARE;
996 if (!(file->f_mode & FMODE_WRITE))
997 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
998
999 /* fall through */
1000 case MAP_PRIVATE:
1001 if (!(file->f_mode & FMODE_READ))
1002 return -EACCES;
1003 break;
1004
1005 default:
1006 return -EINVAL;
1007 }
1008 } else {
1009 switch (flags & MAP_TYPE) {
1010 case MAP_SHARED:
1011 vm_flags |= VM_SHARED | VM_MAYSHARE;
1012 break;
1013 case MAP_PRIVATE:
1014 /*
1015 * Set pgoff according to addr for anon_vma.
1016 */
1017 pgoff = addr >> PAGE_SHIFT;
1018 break;
1019 default:
1020 return -EINVAL;
1021 }
1022 }
1023
1024 error = security_file_mmap(file, reqprot, prot, flags);
1025 if (error)
1026 return error;
1027
1028 /* Clear old maps */
1029 error = -ENOMEM;
1030munmap_back:
1031 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1032 if (vma && vma->vm_start < addr + len) {
1033 if (do_munmap(mm, addr, len))
1034 return -ENOMEM;
1035 goto munmap_back;
1036 }
1037
1038 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001039 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 return -ENOMEM;
1041
1042 if (accountable && (!(flags & MAP_NORESERVE) ||
1043 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1044 if (vm_flags & VM_SHARED) {
1045 /* Check memory availability in shmem_file_setup? */
1046 vm_flags |= VM_ACCOUNT;
1047 } else if (vm_flags & VM_WRITE) {
1048 /*
1049 * Private writable mapping: check memory availability
1050 */
1051 charged = len >> PAGE_SHIFT;
1052 if (security_vm_enough_memory(charged))
1053 return -ENOMEM;
1054 vm_flags |= VM_ACCOUNT;
1055 }
1056 }
1057
1058 /*
1059 * Can we just expand an old private anonymous mapping?
1060 * The VM_SHARED test is necessary because shmem_zero_setup
1061 * will create the file object for a shared anonymous map below.
1062 */
1063 if (!file && !(vm_flags & VM_SHARED) &&
1064 vma_merge(mm, prev, addr, addr + len, vm_flags,
1065 NULL, NULL, pgoff, NULL))
1066 goto out;
1067
1068 /*
1069 * Determine the object being mapped and call the appropriate
1070 * specific mapper. the address has already been validated, but
1071 * not unmapped, but the maps are removed from the list.
1072 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001073 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (!vma) {
1075 error = -ENOMEM;
1076 goto unacct_error;
1077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 vma->vm_mm = mm;
1080 vma->vm_start = addr;
1081 vma->vm_end = addr + len;
1082 vma->vm_flags = vm_flags;
David Howells9637a5e2006-06-23 02:03:43 -07001083 vma->vm_page_prot = protection_map[vm_flags &
1084 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 vma->vm_pgoff = pgoff;
1086
1087 if (file) {
1088 error = -EINVAL;
1089 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1090 goto free_vma;
1091 if (vm_flags & VM_DENYWRITE) {
1092 error = deny_write_access(file);
1093 if (error)
1094 goto free_vma;
1095 correct_wcount = 1;
1096 }
1097 vma->vm_file = file;
1098 get_file(file);
1099 error = file->f_op->mmap(file, vma);
1100 if (error)
1101 goto unmap_and_free_vma;
1102 } else if (vm_flags & VM_SHARED) {
1103 error = shmem_zero_setup(vma);
1104 if (error)
1105 goto free_vma;
1106 }
1107
David Howells9637a5e2006-06-23 02:03:43 -07001108 /* Don't make the VMA automatically writable if it's shared, but the
1109 * backer wishes to know when pages are first written to */
1110 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1111 vma->vm_page_prot =
1112 protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC)];
1113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1115 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1116 * that memory reservation must be checked; but that reservation
1117 * belongs to shared memory object, not to vma: so now clear it.
1118 */
1119 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1120 vma->vm_flags &= ~VM_ACCOUNT;
1121
1122 /* Can addr have changed??
1123 *
1124 * Answer: Yes, several device drivers can do it in their
1125 * f_op->mmap method. -DaveM
1126 */
1127 addr = vma->vm_start;
1128 pgoff = vma->vm_pgoff;
1129 vm_flags = vma->vm_flags;
1130
1131 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1132 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1133 file = vma->vm_file;
1134 vma_link(mm, vma, prev, rb_link, rb_parent);
1135 if (correct_wcount)
1136 atomic_inc(&inode->i_writecount);
1137 } else {
1138 if (file) {
1139 if (correct_wcount)
1140 atomic_inc(&inode->i_writecount);
1141 fput(file);
1142 }
1143 mpol_free(vma_policy(vma));
1144 kmem_cache_free(vm_area_cachep, vma);
1145 }
1146out:
1147 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001148 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 if (vm_flags & VM_LOCKED) {
1150 mm->locked_vm += len >> PAGE_SHIFT;
1151 make_pages_present(addr, addr + len);
1152 }
1153 if (flags & MAP_POPULATE) {
1154 up_write(&mm->mmap_sem);
1155 sys_remap_file_pages(addr, len, 0,
1156 pgoff, flags & MAP_NONBLOCK);
1157 down_write(&mm->mmap_sem);
1158 }
1159 return addr;
1160
1161unmap_and_free_vma:
1162 if (correct_wcount)
1163 atomic_inc(&inode->i_writecount);
1164 vma->vm_file = NULL;
1165 fput(file);
1166
1167 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001168 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1169 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170free_vma:
1171 kmem_cache_free(vm_area_cachep, vma);
1172unacct_error:
1173 if (charged)
1174 vm_unacct_memory(charged);
1175 return error;
1176}
1177
1178EXPORT_SYMBOL(do_mmap_pgoff);
1179
1180/* Get an address range which is currently unmapped.
1181 * For shmat() with addr=0.
1182 *
1183 * Ugly calling convention alert:
1184 * Return value with the low bits set means error value,
1185 * ie
1186 * if (ret & ~PAGE_MASK)
1187 * error = ret;
1188 *
1189 * This function "knows" that -ENOMEM has the bits set.
1190 */
1191#ifndef HAVE_ARCH_UNMAPPED_AREA
1192unsigned long
1193arch_get_unmapped_area(struct file *filp, unsigned long addr,
1194 unsigned long len, unsigned long pgoff, unsigned long flags)
1195{
1196 struct mm_struct *mm = current->mm;
1197 struct vm_area_struct *vma;
1198 unsigned long start_addr;
1199
1200 if (len > TASK_SIZE)
1201 return -ENOMEM;
1202
1203 if (addr) {
1204 addr = PAGE_ALIGN(addr);
1205 vma = find_vma(mm, addr);
1206 if (TASK_SIZE - len >= addr &&
1207 (!vma || addr + len <= vma->vm_start))
1208 return addr;
1209 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001210 if (len > mm->cached_hole_size) {
1211 start_addr = addr = mm->free_area_cache;
1212 } else {
1213 start_addr = addr = TASK_UNMAPPED_BASE;
1214 mm->cached_hole_size = 0;
1215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
1217full_search:
1218 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1219 /* At this point: (!vma || addr < vma->vm_end). */
1220 if (TASK_SIZE - len < addr) {
1221 /*
1222 * Start a new search - just in case we missed
1223 * some holes.
1224 */
1225 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001226 addr = TASK_UNMAPPED_BASE;
1227 start_addr = addr;
1228 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 goto full_search;
1230 }
1231 return -ENOMEM;
1232 }
1233 if (!vma || addr + len <= vma->vm_start) {
1234 /*
1235 * Remember the place where we stopped the search:
1236 */
1237 mm->free_area_cache = addr + len;
1238 return addr;
1239 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001240 if (addr + mm->cached_hole_size < vma->vm_start)
1241 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 addr = vma->vm_end;
1243 }
1244}
1245#endif
1246
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001247void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248{
1249 /*
1250 * Is this a new hole at the lowest possible address?
1251 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001252 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1253 mm->free_area_cache = addr;
1254 mm->cached_hole_size = ~0UL;
1255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
1257
1258/*
1259 * This mmap-allocator allocates new areas top-down from below the
1260 * stack's low limit (the base):
1261 */
1262#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1263unsigned long
1264arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1265 const unsigned long len, const unsigned long pgoff,
1266 const unsigned long flags)
1267{
1268 struct vm_area_struct *vma;
1269 struct mm_struct *mm = current->mm;
1270 unsigned long addr = addr0;
1271
1272 /* requested length too big for entire address space */
1273 if (len > TASK_SIZE)
1274 return -ENOMEM;
1275
1276 /* requesting a specific address */
1277 if (addr) {
1278 addr = PAGE_ALIGN(addr);
1279 vma = find_vma(mm, addr);
1280 if (TASK_SIZE - len >= addr &&
1281 (!vma || addr + len <= vma->vm_start))
1282 return addr;
1283 }
1284
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001285 /* check if free_area_cache is useful for us */
1286 if (len <= mm->cached_hole_size) {
1287 mm->cached_hole_size = 0;
1288 mm->free_area_cache = mm->mmap_base;
1289 }
1290
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 /* either no address requested or can't fit in requested address hole */
1292 addr = mm->free_area_cache;
1293
1294 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001295 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 vma = find_vma(mm, addr-len);
1297 if (!vma || addr <= vma->vm_start)
1298 /* remember the address as a hint for next time */
1299 return (mm->free_area_cache = addr-len);
1300 }
1301
Chris Wright73219d12005-06-21 17:14:52 -07001302 if (mm->mmap_base < len)
1303 goto bottomup;
1304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 addr = mm->mmap_base-len;
1306
1307 do {
1308 /*
1309 * Lookup failure means no vma is above this address,
1310 * else if new region fits below vma->vm_start,
1311 * return with success:
1312 */
1313 vma = find_vma(mm, addr);
1314 if (!vma || addr+len <= vma->vm_start)
1315 /* remember the address as a hint for next time */
1316 return (mm->free_area_cache = addr);
1317
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001318 /* remember the largest hole we saw so far */
1319 if (addr + mm->cached_hole_size < vma->vm_start)
1320 mm->cached_hole_size = vma->vm_start - addr;
1321
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 /* try just below the current vma->vm_start */
1323 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001324 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Chris Wright73219d12005-06-21 17:14:52 -07001326bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 /*
1328 * A failed mmap() very likely causes application failure,
1329 * so fall back to the bottom-up function here. This scenario
1330 * can happen with large stack limits and large mmap()
1331 * allocations.
1332 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001333 mm->cached_hole_size = ~0UL;
1334 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1336 /*
1337 * Restore the topdown base:
1338 */
1339 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001340 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
1342 return addr;
1343}
1344#endif
1345
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001346void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 /*
1349 * Is this a new hole at the highest possible address?
1350 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001351 if (addr > mm->free_area_cache)
1352 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
1354 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001355 if (mm->free_area_cache > mm->mmap_base)
1356 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357}
1358
1359unsigned long
1360get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1361 unsigned long pgoff, unsigned long flags)
1362{
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001363 unsigned long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001365 if (!(flags & MAP_FIXED)) {
1366 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1367
1368 get_area = current->mm->get_unmapped_area;
1369 if (file && file->f_op && file->f_op->get_unmapped_area)
1370 get_area = file->f_op->get_unmapped_area;
1371 addr = get_area(file, addr, len, pgoff, flags);
1372 if (IS_ERR_VALUE(addr))
1373 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 }
1375
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001376 if (addr > TASK_SIZE - len)
1377 return -ENOMEM;
1378 if (addr & ~PAGE_MASK)
1379 return -EINVAL;
1380 if (file && is_file_hugepages(file)) {
1381 /*
1382 * Check if the given range is hugepage aligned, and
1383 * can be made suitable for hugepages.
1384 */
1385 ret = prepare_hugepage_range(addr, len);
1386 } else {
1387 /*
1388 * Ensure that a normal request is not falling in a
1389 * reserved hugepage range. For some archs like IA-64,
1390 * there is a separate region for hugepages.
1391 */
1392 ret = is_hugepage_only_range(current->mm, addr, len);
1393 }
1394 if (ret)
1395 return -EINVAL;
1396 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397}
1398
1399EXPORT_SYMBOL(get_unmapped_area);
1400
1401/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1402struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1403{
1404 struct vm_area_struct *vma = NULL;
1405
1406 if (mm) {
1407 /* Check the cache first. */
1408 /* (Cache hit rate is typically around 35%.) */
1409 vma = mm->mmap_cache;
1410 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1411 struct rb_node * rb_node;
1412
1413 rb_node = mm->mm_rb.rb_node;
1414 vma = NULL;
1415
1416 while (rb_node) {
1417 struct vm_area_struct * vma_tmp;
1418
1419 vma_tmp = rb_entry(rb_node,
1420 struct vm_area_struct, vm_rb);
1421
1422 if (vma_tmp->vm_end > addr) {
1423 vma = vma_tmp;
1424 if (vma_tmp->vm_start <= addr)
1425 break;
1426 rb_node = rb_node->rb_left;
1427 } else
1428 rb_node = rb_node->rb_right;
1429 }
1430 if (vma)
1431 mm->mmap_cache = vma;
1432 }
1433 }
1434 return vma;
1435}
1436
1437EXPORT_SYMBOL(find_vma);
1438
1439/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1440struct vm_area_struct *
1441find_vma_prev(struct mm_struct *mm, unsigned long addr,
1442 struct vm_area_struct **pprev)
1443{
1444 struct vm_area_struct *vma = NULL, *prev = NULL;
1445 struct rb_node * rb_node;
1446 if (!mm)
1447 goto out;
1448
1449 /* Guard against addr being lower than the first VMA */
1450 vma = mm->mmap;
1451
1452 /* Go through the RB tree quickly. */
1453 rb_node = mm->mm_rb.rb_node;
1454
1455 while (rb_node) {
1456 struct vm_area_struct *vma_tmp;
1457 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1458
1459 if (addr < vma_tmp->vm_end) {
1460 rb_node = rb_node->rb_left;
1461 } else {
1462 prev = vma_tmp;
1463 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1464 break;
1465 rb_node = rb_node->rb_right;
1466 }
1467 }
1468
1469out:
1470 *pprev = prev;
1471 return prev ? prev->vm_next : vma;
1472}
1473
1474/*
1475 * Verify that the stack growth is acceptable and
1476 * update accounting. This is shared with both the
1477 * grow-up and grow-down cases.
1478 */
1479static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1480{
1481 struct mm_struct *mm = vma->vm_mm;
1482 struct rlimit *rlim = current->signal->rlim;
1483
1484 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001485 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return -ENOMEM;
1487
1488 /* Stack limit test */
1489 if (size > rlim[RLIMIT_STACK].rlim_cur)
1490 return -ENOMEM;
1491
1492 /* mlock limit tests */
1493 if (vma->vm_flags & VM_LOCKED) {
1494 unsigned long locked;
1495 unsigned long limit;
1496 locked = mm->locked_vm + grow;
1497 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1498 if (locked > limit && !capable(CAP_IPC_LOCK))
1499 return -ENOMEM;
1500 }
1501
1502 /*
1503 * Overcommit.. This must be the final test, as it will
1504 * update security statistics.
1505 */
1506 if (security_vm_enough_memory(grow))
1507 return -ENOMEM;
1508
1509 /* Ok, everything looks good - let it rip */
1510 mm->total_vm += grow;
1511 if (vma->vm_flags & VM_LOCKED)
1512 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001513 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 return 0;
1515}
1516
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001517#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001519 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1520 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001522#ifndef CONFIG_IA64
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001523static inline
1524#endif
1525int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
1527 int error;
1528
1529 if (!(vma->vm_flags & VM_GROWSUP))
1530 return -EFAULT;
1531
1532 /*
1533 * We must make sure the anon_vma is allocated
1534 * so that the anon_vma locking is not a noop.
1535 */
1536 if (unlikely(anon_vma_prepare(vma)))
1537 return -ENOMEM;
1538 anon_vma_lock(vma);
1539
1540 /*
1541 * vma->vm_start/vm_end cannot change under us because the caller
1542 * is required to hold the mmap_sem in read mode. We need the
1543 * anon_vma lock to serialize against concurrent expand_stacks.
1544 */
1545 address += 4 + PAGE_SIZE - 1;
1546 address &= PAGE_MASK;
1547 error = 0;
1548
1549 /* Somebody else might have raced and expanded it already */
1550 if (address > vma->vm_end) {
1551 unsigned long size, grow;
1552
1553 size = address - vma->vm_start;
1554 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1555
1556 error = acct_stack_growth(vma, size, grow);
1557 if (!error)
1558 vma->vm_end = address;
1559 }
1560 anon_vma_unlock(vma);
1561 return error;
1562}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001563#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1564
1565#ifdef CONFIG_STACK_GROWSUP
1566int expand_stack(struct vm_area_struct *vma, unsigned long address)
1567{
1568 return expand_upwards(vma, address);
1569}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
1571struct vm_area_struct *
1572find_extend_vma(struct mm_struct *mm, unsigned long addr)
1573{
1574 struct vm_area_struct *vma, *prev;
1575
1576 addr &= PAGE_MASK;
1577 vma = find_vma_prev(mm, addr, &prev);
1578 if (vma && (vma->vm_start <= addr))
1579 return vma;
1580 if (!prev || expand_stack(prev, addr))
1581 return NULL;
1582 if (prev->vm_flags & VM_LOCKED) {
1583 make_pages_present(addr, prev->vm_end);
1584 }
1585 return prev;
1586}
1587#else
1588/*
1589 * vma is the first one with address < vma->vm_start. Have to extend vma.
1590 */
1591int expand_stack(struct vm_area_struct *vma, unsigned long address)
1592{
1593 int error;
1594
1595 /*
1596 * We must make sure the anon_vma is allocated
1597 * so that the anon_vma locking is not a noop.
1598 */
1599 if (unlikely(anon_vma_prepare(vma)))
1600 return -ENOMEM;
1601 anon_vma_lock(vma);
1602
1603 /*
1604 * vma->vm_start/vm_end cannot change under us because the caller
1605 * is required to hold the mmap_sem in read mode. We need the
1606 * anon_vma lock to serialize against concurrent expand_stacks.
1607 */
1608 address &= PAGE_MASK;
1609 error = 0;
1610
1611 /* Somebody else might have raced and expanded it already */
1612 if (address < vma->vm_start) {
1613 unsigned long size, grow;
1614
1615 size = vma->vm_end - address;
1616 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1617
1618 error = acct_stack_growth(vma, size, grow);
1619 if (!error) {
1620 vma->vm_start = address;
1621 vma->vm_pgoff -= grow;
1622 }
1623 }
1624 anon_vma_unlock(vma);
1625 return error;
1626}
1627
1628struct vm_area_struct *
1629find_extend_vma(struct mm_struct * mm, unsigned long addr)
1630{
1631 struct vm_area_struct * vma;
1632 unsigned long start;
1633
1634 addr &= PAGE_MASK;
1635 vma = find_vma(mm,addr);
1636 if (!vma)
1637 return NULL;
1638 if (vma->vm_start <= addr)
1639 return vma;
1640 if (!(vma->vm_flags & VM_GROWSDOWN))
1641 return NULL;
1642 start = vma->vm_start;
1643 if (expand_stack(vma, addr))
1644 return NULL;
1645 if (vma->vm_flags & VM_LOCKED) {
1646 make_pages_present(addr, start);
1647 }
1648 return vma;
1649}
1650#endif
1651
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001653 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001655 *
1656 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001658static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001660 /* Update high watermark before we lower total_vm */
1661 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001663 long nrpages = vma_pages(vma);
1664
1665 mm->total_vm -= nrpages;
1666 if (vma->vm_flags & VM_LOCKED)
1667 mm->locked_vm -= nrpages;
1668 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001669 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001670 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 validate_mm(mm);
1672}
1673
1674/*
1675 * Get rid of page table information in the indicated region.
1676 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001677 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 */
1679static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001680 struct vm_area_struct *vma, struct vm_area_struct *prev,
1681 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001683 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 struct mmu_gather *tlb;
1685 unsigned long nr_accounted = 0;
1686
1687 lru_add_drain();
1688 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001689 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001690 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001692 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001693 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 tlb_finish_mmu(tlb, start, end);
1695}
1696
1697/*
1698 * Create a list of vma's touched by the unmap, removing them from the mm's
1699 * vma list as we go..
1700 */
1701static void
1702detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1703 struct vm_area_struct *prev, unsigned long end)
1704{
1705 struct vm_area_struct **insertion_point;
1706 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001707 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1710 do {
1711 rb_erase(&vma->vm_rb, &mm->mm_rb);
1712 mm->map_count--;
1713 tail_vma = vma;
1714 vma = vma->vm_next;
1715 } while (vma && vma->vm_start < end);
1716 *insertion_point = vma;
1717 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001718 if (mm->unmap_area == arch_unmap_area)
1719 addr = prev ? prev->vm_end : mm->mmap_base;
1720 else
1721 addr = vma ? vma->vm_start : mm->mmap_base;
1722 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 mm->mmap_cache = NULL; /* Kill the cache. */
1724}
1725
1726/*
1727 * Split a vma into two pieces at address 'addr', a new vma is allocated
1728 * either for the first part or the the tail.
1729 */
1730int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1731 unsigned long addr, int new_below)
1732{
1733 struct mempolicy *pol;
1734 struct vm_area_struct *new;
1735
1736 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1737 return -EINVAL;
1738
1739 if (mm->map_count >= sysctl_max_map_count)
1740 return -ENOMEM;
1741
1742 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1743 if (!new)
1744 return -ENOMEM;
1745
1746 /* most fields are the same, copy all, and then fixup */
1747 *new = *vma;
1748
1749 if (new_below)
1750 new->vm_end = addr;
1751 else {
1752 new->vm_start = addr;
1753 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1754 }
1755
1756 pol = mpol_copy(vma_policy(vma));
1757 if (IS_ERR(pol)) {
1758 kmem_cache_free(vm_area_cachep, new);
1759 return PTR_ERR(pol);
1760 }
1761 vma_set_policy(new, pol);
1762
1763 if (new->vm_file)
1764 get_file(new->vm_file);
1765
1766 if (new->vm_ops && new->vm_ops->open)
1767 new->vm_ops->open(new);
1768
1769 if (new_below)
1770 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1771 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1772 else
1773 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1774
1775 return 0;
1776}
1777
1778/* Munmap is split into 2 main parts -- this part which finds
1779 * what needs doing, and the areas themselves, which do the
1780 * work. This now handles partial unmappings.
1781 * Jeremy Fitzhardinge <jeremy@goop.org>
1782 */
1783int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1784{
1785 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001786 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
1788 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1789 return -EINVAL;
1790
1791 if ((len = PAGE_ALIGN(len)) == 0)
1792 return -EINVAL;
1793
1794 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001795 vma = find_vma_prev(mm, start, &prev);
1796 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001798 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
1800 /* if it doesn't overlap, we have nothing.. */
1801 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001802 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 return 0;
1804
1805 /*
1806 * If we need to split any vma, do it now to save pain later.
1807 *
1808 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1809 * unmapped vm_area_struct will remain in use: so lower split_vma
1810 * places tmp vma above, and higher split_vma places tmp vma below.
1811 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001812 if (start > vma->vm_start) {
1813 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 if (error)
1815 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001816 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 }
1818
1819 /* Does it split the last one? */
1820 last = find_vma(mm, end);
1821 if (last && end > last->vm_start) {
1822 int error = split_vma(mm, last, end, 1);
1823 if (error)
1824 return error;
1825 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001826 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
1828 /*
1829 * Remove the vma's, and unmap the actual pages
1830 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001831 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1832 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
1834 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001835 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
1837 return 0;
1838}
1839
1840EXPORT_SYMBOL(do_munmap);
1841
1842asmlinkage long sys_munmap(unsigned long addr, size_t len)
1843{
1844 int ret;
1845 struct mm_struct *mm = current->mm;
1846
1847 profile_munmap(addr);
1848
1849 down_write(&mm->mmap_sem);
1850 ret = do_munmap(mm, addr, len);
1851 up_write(&mm->mmap_sem);
1852 return ret;
1853}
1854
1855static inline void verify_mm_writelocked(struct mm_struct *mm)
1856{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001857#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1859 WARN_ON(1);
1860 up_read(&mm->mmap_sem);
1861 }
1862#endif
1863}
1864
1865/*
1866 * this is really a simplified "do_mmap". it only handles
1867 * anonymous maps. eventually we may be able to do some
1868 * brk-specific accounting here.
1869 */
1870unsigned long do_brk(unsigned long addr, unsigned long len)
1871{
1872 struct mm_struct * mm = current->mm;
1873 struct vm_area_struct * vma, * prev;
1874 unsigned long flags;
1875 struct rb_node ** rb_link, * rb_parent;
1876 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04001877 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
1879 len = PAGE_ALIGN(len);
1880 if (!len)
1881 return addr;
1882
1883 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1884 return -EINVAL;
1885
Kirill Korotaev3a459752006-09-07 14:17:04 +04001886 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1887
1888 error = arch_mmap_check(addr, len, flags);
1889 if (error)
1890 return error;
1891
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 /*
1893 * mlock MCL_FUTURE?
1894 */
1895 if (mm->def_flags & VM_LOCKED) {
1896 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001897 locked = len >> PAGE_SHIFT;
1898 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07001900 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1902 return -EAGAIN;
1903 }
1904
1905 /*
1906 * mm->mmap_sem is required to protect against another thread
1907 * changing the mappings in case we sleep.
1908 */
1909 verify_mm_writelocked(mm);
1910
1911 /*
1912 * Clear old maps. this also does some error checking for us
1913 */
1914 munmap_back:
1915 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1916 if (vma && vma->vm_start < addr + len) {
1917 if (do_munmap(mm, addr, len))
1918 return -ENOMEM;
1919 goto munmap_back;
1920 }
1921
1922 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001923 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 return -ENOMEM;
1925
1926 if (mm->map_count > sysctl_max_map_count)
1927 return -ENOMEM;
1928
1929 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1930 return -ENOMEM;
1931
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 /* Can we just expand an old private anonymous mapping? */
1933 if (vma_merge(mm, prev, addr, addr + len, flags,
1934 NULL, NULL, pgoff, NULL))
1935 goto out;
1936
1937 /*
1938 * create a vma struct for an anonymous mapping
1939 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001940 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 if (!vma) {
1942 vm_unacct_memory(len >> PAGE_SHIFT);
1943 return -ENOMEM;
1944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946 vma->vm_mm = mm;
1947 vma->vm_start = addr;
1948 vma->vm_end = addr + len;
1949 vma->vm_pgoff = pgoff;
1950 vma->vm_flags = flags;
David Howells9637a5e2006-06-23 02:03:43 -07001951 vma->vm_page_prot = protection_map[flags &
1952 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 vma_link(mm, vma, prev, rb_link, rb_parent);
1954out:
1955 mm->total_vm += len >> PAGE_SHIFT;
1956 if (flags & VM_LOCKED) {
1957 mm->locked_vm += len >> PAGE_SHIFT;
1958 make_pages_present(addr, addr + len);
1959 }
1960 return addr;
1961}
1962
1963EXPORT_SYMBOL(do_brk);
1964
1965/* Release all mmaps. */
1966void exit_mmap(struct mm_struct *mm)
1967{
1968 struct mmu_gather *tlb;
Hugh Dickinse0da3822005-04-19 13:29:15 -07001969 struct vm_area_struct *vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001971 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
1973 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001975 tlb = tlb_gather_mmu(mm, 1);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001976 /* Don't update_hiwater_rss(mm) here, do_exit already did */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001977 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07001978 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001980 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001981 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07001984 * Walk the list again, actually closing and freeing it,
1985 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001987 while (vma)
1988 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001989
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001990 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991}
1992
1993/* Insert vm structure into process list sorted by address
1994 * and into the inode's i_mmap tree. If vm_file is non-NULL
1995 * then i_mmap_lock is taken here.
1996 */
1997int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1998{
1999 struct vm_area_struct * __vma, * prev;
2000 struct rb_node ** rb_link, * rb_parent;
2001
2002 /*
2003 * The vm_pgoff of a purely anonymous vma should be irrelevant
2004 * until its first write fault, when page's anon_vma and index
2005 * are set. But now set the vm_pgoff it will almost certainly
2006 * end up with (unless mremap moves it elsewhere before that
2007 * first wfault), so /proc/pid/maps tells a consistent story.
2008 *
2009 * By setting it to reflect the virtual start address of the
2010 * vma, merges and splits can happen in a seamless way, just
2011 * using the existing file pgoff checks and manipulations.
2012 * Similarly in do_mmap_pgoff and in do_brk.
2013 */
2014 if (!vma->vm_file) {
2015 BUG_ON(vma->anon_vma);
2016 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2017 }
2018 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2019 if (__vma && __vma->vm_start < vma->vm_end)
2020 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002021 if ((vma->vm_flags & VM_ACCOUNT) &&
2022 security_vm_enough_memory(vma_pages(vma)))
2023 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 vma_link(mm, vma, prev, rb_link, rb_parent);
2025 return 0;
2026}
2027
2028/*
2029 * Copy the vma structure to a new location in the same mm,
2030 * prior to moving page table entries, to effect an mremap move.
2031 */
2032struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2033 unsigned long addr, unsigned long len, pgoff_t pgoff)
2034{
2035 struct vm_area_struct *vma = *vmap;
2036 unsigned long vma_start = vma->vm_start;
2037 struct mm_struct *mm = vma->vm_mm;
2038 struct vm_area_struct *new_vma, *prev;
2039 struct rb_node **rb_link, *rb_parent;
2040 struct mempolicy *pol;
2041
2042 /*
2043 * If anonymous vma has not yet been faulted, update new pgoff
2044 * to match new location, to increase its chance of merging.
2045 */
2046 if (!vma->vm_file && !vma->anon_vma)
2047 pgoff = addr >> PAGE_SHIFT;
2048
2049 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2050 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2051 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2052 if (new_vma) {
2053 /*
2054 * Source vma may have been merged into new_vma
2055 */
2056 if (vma_start >= new_vma->vm_start &&
2057 vma_start < new_vma->vm_end)
2058 *vmap = new_vma;
2059 } else {
2060 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2061 if (new_vma) {
2062 *new_vma = *vma;
2063 pol = mpol_copy(vma_policy(vma));
2064 if (IS_ERR(pol)) {
2065 kmem_cache_free(vm_area_cachep, new_vma);
2066 return NULL;
2067 }
2068 vma_set_policy(new_vma, pol);
2069 new_vma->vm_start = addr;
2070 new_vma->vm_end = addr + len;
2071 new_vma->vm_pgoff = pgoff;
2072 if (new_vma->vm_file)
2073 get_file(new_vma->vm_file);
2074 if (new_vma->vm_ops && new_vma->vm_ops->open)
2075 new_vma->vm_ops->open(new_vma);
2076 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2077 }
2078 }
2079 return new_vma;
2080}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002081
2082/*
2083 * Return true if the calling process may expand its vm space by the passed
2084 * number of pages
2085 */
2086int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2087{
2088 unsigned long cur = mm->total_vm; /* pages */
2089 unsigned long lim;
2090
2091 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2092
2093 if (cur + npages > lim)
2094 return 0;
2095 return 1;
2096}