blob: 0d40e66c841b2d2cad8e03ebc6d5e16491b0774b [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>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020032#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Kirill Korotaev3a459752006-09-07 14:17:04 +040034#ifndef arch_mmap_check
35#define arch_mmap_check(addr, len, flags) (0)
36#endif
37
Hugh Dickinse0da3822005-04-19 13:29:15 -070038static void unmap_region(struct mm_struct *mm,
39 struct vm_area_struct *vma, struct vm_area_struct *prev,
40 unsigned long start, unsigned long end);
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * WARNING: the debugging will use recursive algorithms so never enable this
44 * unless you know what you are doing.
45 */
46#undef DEBUG_MM_RB
47
48/* description of effects of mapping type and prot in current implementation.
49 * this is due to the limited x86 page protection hardware. The expected
50 * behavior is in parens:
51 *
52 * map_type prot
53 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
54 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
55 * w: (no) no w: (no) no w: (yes) yes w: (no) no
56 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
57 *
58 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
59 * w: (no) no w: (no) no w: (copy) copy w: (no) no
60 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
61 *
62 */
63pgprot_t protection_map[16] = {
64 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
65 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
66};
67
Hugh Dickins804af2c2006-07-26 21:39:49 +010068pgprot_t vm_get_page_prot(unsigned long vm_flags)
69{
70 return protection_map[vm_flags &
71 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
72}
73EXPORT_SYMBOL(vm_get_page_prot);
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
76int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070077int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078atomic_t vm_committed_space = ATOMIC_INIT(0);
79
80/*
81 * Check that a process has enough memory to allocate a new virtual
82 * mapping. 0 means there is enough memory for the allocation to
83 * succeed and -ENOMEM implies there is not.
84 *
85 * We currently support three overcommit policies, which are set via the
86 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
87 *
88 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
89 * Additional code 2002 Jul 20 by Robert Love.
90 *
91 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
92 *
93 * Note this is a helper function intended to be used by LSMs which
94 * wish to use this logic.
95 */
Alan Cox34b4e4a2007-08-22 14:01:28 -070096int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097{
98 unsigned long free, allowed;
99
100 vm_acct_memory(pages);
101
102 /*
103 * Sometimes we want to use more memory than we have
104 */
105 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
106 return 0;
107
108 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
109 unsigned long n;
110
Christoph Lameter347ce432006-06-30 01:55:35 -0700111 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 free += nr_swap_pages;
113
114 /*
115 * Any slabs which are created with the
116 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
117 * which are reclaimable, under pressure. The dentry
118 * cache and most inode caches should fall into this
119 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700120 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
122 /*
123 * Leave the last 3% for root
124 */
125 if (!cap_sys_admin)
126 free -= free / 32;
127
128 if (free > pages)
129 return 0;
130
131 /*
132 * nr_free_pages() is very expensive on large systems,
133 * only call if we're about to fail.
134 */
135 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700136
137 /*
138 * Leave reserved pages. The pages are not for anonymous pages.
139 */
140 if (n <= totalreserve_pages)
141 goto error;
142 else
143 n -= totalreserve_pages;
144
145 /*
146 * Leave the last 3% for root
147 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 if (!cap_sys_admin)
149 n -= n / 32;
150 free += n;
151
152 if (free > pages)
153 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700154
155 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 }
157
158 allowed = (totalram_pages - hugetlb_total_pages())
159 * sysctl_overcommit_ratio / 100;
160 /*
161 * Leave the last 3% for root
162 */
163 if (!cap_sys_admin)
164 allowed -= allowed / 32;
165 allowed += total_swap_pages;
166
167 /* Don't let a single process grow too big:
168 leave 3% of the size of this process for other processes */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700169 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Simon Derr2f60f8d2005-08-04 19:52:03 -0700171 /*
172 * cast `allowed' as a signed long because vm_committed_space
173 * sometimes has a negative value
174 */
175 if (atomic_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700177error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 vm_unacct_memory(pages);
179
180 return -ENOMEM;
181}
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183EXPORT_SYMBOL(__vm_enough_memory);
184
185/*
186 * Requires inode->i_mapping->i_mmap_lock
187 */
188static void __remove_shared_vm_struct(struct vm_area_struct *vma,
189 struct file *file, struct address_space *mapping)
190{
191 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800192 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (vma->vm_flags & VM_SHARED)
194 mapping->i_mmap_writable--;
195
196 flush_dcache_mmap_lock(mapping);
197 if (unlikely(vma->vm_flags & VM_NONLINEAR))
198 list_del_init(&vma->shared.vm_set.list);
199 else
200 vma_prio_tree_remove(vma, &mapping->i_mmap);
201 flush_dcache_mmap_unlock(mapping);
202}
203
204/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700205 * Unlink a file-based vm structure from its prio_tree, to hide
206 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700208void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 struct file *file = vma->vm_file;
211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 if (file) {
213 struct address_space *mapping = file->f_mapping;
214 spin_lock(&mapping->i_mmap_lock);
215 __remove_shared_vm_struct(vma, file, mapping);
216 spin_unlock(&mapping->i_mmap_lock);
217 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700218}
219
220/*
221 * Close a vm structure and free it, returning the next.
222 */
223static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
224{
225 struct vm_area_struct *next = vma->vm_next;
226
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700227 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 if (vma->vm_ops && vma->vm_ops->close)
229 vma->vm_ops->close(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700230 if (vma->vm_file)
231 fput(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 mpol_free(vma_policy(vma));
233 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700234 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237asmlinkage unsigned long sys_brk(unsigned long brk)
238{
239 unsigned long rlim, retval;
240 unsigned long newbrk, oldbrk;
241 struct mm_struct *mm = current->mm;
242
243 down_write(&mm->mmap_sem);
244
245 if (brk < mm->end_code)
246 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700247
248 /*
249 * Check against rlimit here. If this check is done later after the test
250 * of oldbrk with newbrk then it can escape the test and let the data
251 * segment grow beyond its set limit the in case where the limit is
252 * not page aligned -Ram Gupta
253 */
254 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
255 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
256 goto out;
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 newbrk = PAGE_ALIGN(brk);
259 oldbrk = PAGE_ALIGN(mm->brk);
260 if (oldbrk == newbrk)
261 goto set_brk;
262
263 /* Always allow shrinking brk. */
264 if (brk <= mm->brk) {
265 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
266 goto set_brk;
267 goto out;
268 }
269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 /* Check against existing mmap mappings. */
271 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
272 goto out;
273
274 /* Ok, looks good - let it rip. */
275 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
276 goto out;
277set_brk:
278 mm->brk = brk;
279out:
280 retval = mm->brk;
281 up_write(&mm->mmap_sem);
282 return retval;
283}
284
285#ifdef DEBUG_MM_RB
286static int browse_rb(struct rb_root *root)
287{
288 int i = 0, j;
289 struct rb_node *nd, *pn = NULL;
290 unsigned long prev = 0, pend = 0;
291
292 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
293 struct vm_area_struct *vma;
294 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
295 if (vma->vm_start < prev)
296 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
297 if (vma->vm_start < pend)
298 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
299 if (vma->vm_start > vma->vm_end)
300 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
301 i++;
302 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800303 prev = vma->vm_start;
304 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 }
306 j = 0;
307 for (nd = pn; nd; nd = rb_prev(nd)) {
308 j++;
309 }
310 if (i != j)
311 printk("backwards %d, forwards %d\n", j, i), i = 0;
312 return i;
313}
314
315void validate_mm(struct mm_struct *mm)
316{
317 int bug = 0;
318 int i = 0;
319 struct vm_area_struct *tmp = mm->mmap;
320 while (tmp) {
321 tmp = tmp->vm_next;
322 i++;
323 }
324 if (i != mm->map_count)
325 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
326 i = browse_rb(&mm->mm_rb);
327 if (i != mm->map_count)
328 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200329 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331#else
332#define validate_mm(mm) do { } while (0)
333#endif
334
335static struct vm_area_struct *
336find_vma_prepare(struct mm_struct *mm, unsigned long addr,
337 struct vm_area_struct **pprev, struct rb_node ***rb_link,
338 struct rb_node ** rb_parent)
339{
340 struct vm_area_struct * vma;
341 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
342
343 __rb_link = &mm->mm_rb.rb_node;
344 rb_prev = __rb_parent = NULL;
345 vma = NULL;
346
347 while (*__rb_link) {
348 struct vm_area_struct *vma_tmp;
349
350 __rb_parent = *__rb_link;
351 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
352
353 if (vma_tmp->vm_end > addr) {
354 vma = vma_tmp;
355 if (vma_tmp->vm_start <= addr)
356 return vma;
357 __rb_link = &__rb_parent->rb_left;
358 } else {
359 rb_prev = __rb_parent;
360 __rb_link = &__rb_parent->rb_right;
361 }
362 }
363
364 *pprev = NULL;
365 if (rb_prev)
366 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
367 *rb_link = __rb_link;
368 *rb_parent = __rb_parent;
369 return vma;
370}
371
372static inline void
373__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
374 struct vm_area_struct *prev, struct rb_node *rb_parent)
375{
376 if (prev) {
377 vma->vm_next = prev->vm_next;
378 prev->vm_next = vma;
379 } else {
380 mm->mmap = vma;
381 if (rb_parent)
382 vma->vm_next = rb_entry(rb_parent,
383 struct vm_area_struct, vm_rb);
384 else
385 vma->vm_next = NULL;
386 }
387}
388
389void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
390 struct rb_node **rb_link, struct rb_node *rb_parent)
391{
392 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
393 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
394}
395
396static inline void __vma_link_file(struct vm_area_struct *vma)
397{
398 struct file * file;
399
400 file = vma->vm_file;
401 if (file) {
402 struct address_space *mapping = file->f_mapping;
403
404 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800405 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 if (vma->vm_flags & VM_SHARED)
407 mapping->i_mmap_writable++;
408
409 flush_dcache_mmap_lock(mapping);
410 if (unlikely(vma->vm_flags & VM_NONLINEAR))
411 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
412 else
413 vma_prio_tree_insert(vma, &mapping->i_mmap);
414 flush_dcache_mmap_unlock(mapping);
415 }
416}
417
418static void
419__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
420 struct vm_area_struct *prev, struct rb_node **rb_link,
421 struct rb_node *rb_parent)
422{
423 __vma_link_list(mm, vma, prev, rb_parent);
424 __vma_link_rb(mm, vma, rb_link, rb_parent);
425 __anon_vma_link(vma);
426}
427
428static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
429 struct vm_area_struct *prev, struct rb_node **rb_link,
430 struct rb_node *rb_parent)
431{
432 struct address_space *mapping = NULL;
433
434 if (vma->vm_file)
435 mapping = vma->vm_file->f_mapping;
436
437 if (mapping) {
438 spin_lock(&mapping->i_mmap_lock);
439 vma->vm_truncate_count = mapping->truncate_count;
440 }
441 anon_vma_lock(vma);
442
443 __vma_link(mm, vma, prev, rb_link, rb_parent);
444 __vma_link_file(vma);
445
446 anon_vma_unlock(vma);
447 if (mapping)
448 spin_unlock(&mapping->i_mmap_lock);
449
450 mm->map_count++;
451 validate_mm(mm);
452}
453
454/*
455 * Helper for vma_adjust in the split_vma insert case:
456 * insert vm structure into list and rbtree and anon_vma,
457 * but it has already been inserted into prio_tree earlier.
458 */
459static void
460__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
461{
462 struct vm_area_struct * __vma, * prev;
463 struct rb_node ** rb_link, * rb_parent;
464
465 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200466 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 __vma_link(mm, vma, prev, rb_link, rb_parent);
468 mm->map_count++;
469}
470
471static inline void
472__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
473 struct vm_area_struct *prev)
474{
475 prev->vm_next = vma->vm_next;
476 rb_erase(&vma->vm_rb, &mm->mm_rb);
477 if (mm->mmap_cache == vma)
478 mm->mmap_cache = prev;
479}
480
481/*
482 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
483 * is already present in an i_mmap tree without adjusting the tree.
484 * The following helper function should be used when such adjustments
485 * are necessary. The "insert" vma (if any) is to be inserted
486 * before we drop the necessary locks.
487 */
488void vma_adjust(struct vm_area_struct *vma, unsigned long start,
489 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
490{
491 struct mm_struct *mm = vma->vm_mm;
492 struct vm_area_struct *next = vma->vm_next;
493 struct vm_area_struct *importer = NULL;
494 struct address_space *mapping = NULL;
495 struct prio_tree_root *root = NULL;
496 struct file *file = vma->vm_file;
497 struct anon_vma *anon_vma = NULL;
498 long adjust_next = 0;
499 int remove_next = 0;
500
501 if (next && !insert) {
502 if (end >= next->vm_end) {
503 /*
504 * vma expands, overlapping all the next, and
505 * perhaps the one after too (mprotect case 6).
506 */
507again: remove_next = 1 + (end > next->vm_end);
508 end = next->vm_end;
509 anon_vma = next->anon_vma;
510 importer = vma;
511 } else if (end > next->vm_start) {
512 /*
513 * vma expands, overlapping part of the next:
514 * mprotect case 5 shifting the boundary up.
515 */
516 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
517 anon_vma = next->anon_vma;
518 importer = vma;
519 } else if (end < vma->vm_end) {
520 /*
521 * vma shrinks, and !insert tells it's not
522 * split_vma inserting another: so it must be
523 * mprotect case 4 shifting the boundary down.
524 */
525 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
526 anon_vma = next->anon_vma;
527 importer = next;
528 }
529 }
530
531 if (file) {
532 mapping = file->f_mapping;
533 if (!(vma->vm_flags & VM_NONLINEAR))
534 root = &mapping->i_mmap;
535 spin_lock(&mapping->i_mmap_lock);
536 if (importer &&
537 vma->vm_truncate_count != next->vm_truncate_count) {
538 /*
539 * unmap_mapping_range might be in progress:
540 * ensure that the expanding vma is rescanned.
541 */
542 importer->vm_truncate_count = 0;
543 }
544 if (insert) {
545 insert->vm_truncate_count = vma->vm_truncate_count;
546 /*
547 * Put into prio_tree now, so instantiated pages
548 * are visible to arm/parisc __flush_dcache_page
549 * throughout; but we cannot insert into address
550 * space until vma start or end is updated.
551 */
552 __vma_link_file(insert);
553 }
554 }
555
556 /*
557 * When changing only vma->vm_end, we don't really need
558 * anon_vma lock: but is that case worth optimizing out?
559 */
560 if (vma->anon_vma)
561 anon_vma = vma->anon_vma;
562 if (anon_vma) {
563 spin_lock(&anon_vma->lock);
564 /*
565 * Easily overlooked: when mprotect shifts the boundary,
566 * make sure the expanding vma has anon_vma set if the
567 * shrinking vma had, to cover any anon pages imported.
568 */
569 if (importer && !importer->anon_vma) {
570 importer->anon_vma = anon_vma;
571 __anon_vma_link(importer);
572 }
573 }
574
575 if (root) {
576 flush_dcache_mmap_lock(mapping);
577 vma_prio_tree_remove(vma, root);
578 if (adjust_next)
579 vma_prio_tree_remove(next, root);
580 }
581
582 vma->vm_start = start;
583 vma->vm_end = end;
584 vma->vm_pgoff = pgoff;
585 if (adjust_next) {
586 next->vm_start += adjust_next << PAGE_SHIFT;
587 next->vm_pgoff += adjust_next;
588 }
589
590 if (root) {
591 if (adjust_next)
592 vma_prio_tree_insert(next, root);
593 vma_prio_tree_insert(vma, root);
594 flush_dcache_mmap_unlock(mapping);
595 }
596
597 if (remove_next) {
598 /*
599 * vma_merge has merged next into vma, and needs
600 * us to remove next before dropping the locks.
601 */
602 __vma_unlink(mm, next, vma);
603 if (file)
604 __remove_shared_vm_struct(next, file, mapping);
605 if (next->anon_vma)
606 __anon_vma_merge(vma, next);
607 } else if (insert) {
608 /*
609 * split_vma has split insert from vma, and needs
610 * us to insert it before dropping the locks
611 * (it may either follow vma or precede it).
612 */
613 __insert_vm_struct(mm, insert);
614 }
615
616 if (anon_vma)
617 spin_unlock(&anon_vma->lock);
618 if (mapping)
619 spin_unlock(&mapping->i_mmap_lock);
620
621 if (remove_next) {
622 if (file)
623 fput(file);
624 mm->map_count--;
625 mpol_free(vma_policy(next));
626 kmem_cache_free(vm_area_cachep, next);
627 /*
628 * In mprotect's case 6 (see comments on vma_merge),
629 * we must remove another next too. It would clutter
630 * up the code too much to do both in one go.
631 */
632 if (remove_next == 2) {
633 next = vma->vm_next;
634 goto again;
635 }
636 }
637
638 validate_mm(mm);
639}
640
641/*
642 * If the vma has a ->close operation then the driver probably needs to release
643 * per-vma resources, so we don't attempt to merge those.
644 */
Hugh Dickinsa6f563d2006-03-22 00:08:29 -0800645#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647static inline int is_mergeable_vma(struct vm_area_struct *vma,
648 struct file *file, unsigned long vm_flags)
649{
650 if (vma->vm_flags != vm_flags)
651 return 0;
652 if (vma->vm_file != file)
653 return 0;
654 if (vma->vm_ops && vma->vm_ops->close)
655 return 0;
656 return 1;
657}
658
659static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
660 struct anon_vma *anon_vma2)
661{
662 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
663}
664
665/*
666 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
667 * in front of (at a lower virtual address and file offset than) the vma.
668 *
669 * We cannot merge two vmas if they have differently assigned (non-NULL)
670 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
671 *
672 * We don't check here for the merged mmap wrapping around the end of pagecache
673 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
674 * wrap, nor mmaps which cover the final page at index -1UL.
675 */
676static int
677can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
678 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
679{
680 if (is_mergeable_vma(vma, file, vm_flags) &&
681 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
682 if (vma->vm_pgoff == vm_pgoff)
683 return 1;
684 }
685 return 0;
686}
687
688/*
689 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
690 * beyond (at a higher virtual address and file offset than) the vma.
691 *
692 * We cannot merge two vmas if they have differently assigned (non-NULL)
693 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
694 */
695static int
696can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
697 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
698{
699 if (is_mergeable_vma(vma, file, vm_flags) &&
700 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
701 pgoff_t vm_pglen;
702 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
703 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
704 return 1;
705 }
706 return 0;
707}
708
709/*
710 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
711 * whether that can be merged with its predecessor or its successor.
712 * Or both (it neatly fills a hole).
713 *
714 * In most cases - when called for mmap, brk or mremap - [addr,end) is
715 * certain not to be mapped by the time vma_merge is called; but when
716 * called for mprotect, it is certain to be already mapped (either at
717 * an offset within prev, or at the start of next), and the flags of
718 * this area are about to be changed to vm_flags - and the no-change
719 * case has already been eliminated.
720 *
721 * The following mprotect cases have to be considered, where AAAA is
722 * the area passed down from mprotect_fixup, never extending beyond one
723 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
724 *
725 * AAAA AAAA AAAA AAAA
726 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
727 * cannot merge might become might become might become
728 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
729 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
730 * mremap move: PPPPNNNNNNNN 8
731 * AAAA
732 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
733 * might become case 1 below case 2 below case 3 below
734 *
735 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
736 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
737 */
738struct vm_area_struct *vma_merge(struct mm_struct *mm,
739 struct vm_area_struct *prev, unsigned long addr,
740 unsigned long end, unsigned long vm_flags,
741 struct anon_vma *anon_vma, struct file *file,
742 pgoff_t pgoff, struct mempolicy *policy)
743{
744 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
745 struct vm_area_struct *area, *next;
746
747 /*
748 * We later require that vma->vm_flags == vm_flags,
749 * so this tests vma->vm_flags & VM_SPECIAL, too.
750 */
751 if (vm_flags & VM_SPECIAL)
752 return NULL;
753
754 if (prev)
755 next = prev->vm_next;
756 else
757 next = mm->mmap;
758 area = next;
759 if (next && next->vm_end == end) /* cases 6, 7, 8 */
760 next = next->vm_next;
761
762 /*
763 * Can it merge with the predecessor?
764 */
765 if (prev && prev->vm_end == addr &&
766 mpol_equal(vma_policy(prev), policy) &&
767 can_vma_merge_after(prev, vm_flags,
768 anon_vma, file, pgoff)) {
769 /*
770 * OK, it can. Can we now merge in the successor as well?
771 */
772 if (next && end == next->vm_start &&
773 mpol_equal(policy, vma_policy(next)) &&
774 can_vma_merge_before(next, vm_flags,
775 anon_vma, file, pgoff+pglen) &&
776 is_mergeable_anon_vma(prev->anon_vma,
777 next->anon_vma)) {
778 /* cases 1, 6 */
779 vma_adjust(prev, prev->vm_start,
780 next->vm_end, prev->vm_pgoff, NULL);
781 } else /* cases 2, 5, 7 */
782 vma_adjust(prev, prev->vm_start,
783 end, prev->vm_pgoff, NULL);
784 return prev;
785 }
786
787 /*
788 * Can this new request be merged in front of next?
789 */
790 if (next && end == next->vm_start &&
791 mpol_equal(policy, vma_policy(next)) &&
792 can_vma_merge_before(next, vm_flags,
793 anon_vma, file, pgoff+pglen)) {
794 if (prev && addr < prev->vm_end) /* case 4 */
795 vma_adjust(prev, prev->vm_start,
796 addr, prev->vm_pgoff, NULL);
797 else /* cases 3, 8 */
798 vma_adjust(area, addr, next->vm_end,
799 next->vm_pgoff - pglen, NULL);
800 return area;
801 }
802
803 return NULL;
804}
805
806/*
807 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
808 * neighbouring vmas for a suitable anon_vma, before it goes off
809 * to allocate a new anon_vma. It checks because a repetitive
810 * sequence of mprotects and faults may otherwise lead to distinct
811 * anon_vmas being allocated, preventing vma merge in subsequent
812 * mprotect.
813 */
814struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
815{
816 struct vm_area_struct *near;
817 unsigned long vm_flags;
818
819 near = vma->vm_next;
820 if (!near)
821 goto try_prev;
822
823 /*
824 * Since only mprotect tries to remerge vmas, match flags
825 * which might be mprotected into each other later on.
826 * Neither mlock nor madvise tries to remerge at present,
827 * so leave their flags as obstructing a merge.
828 */
829 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
830 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
831
832 if (near->anon_vma && vma->vm_end == near->vm_start &&
833 mpol_equal(vma_policy(vma), vma_policy(near)) &&
834 can_vma_merge_before(near, vm_flags,
835 NULL, vma->vm_file, vma->vm_pgoff +
836 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
837 return near->anon_vma;
838try_prev:
839 /*
840 * It is potentially slow to have to call find_vma_prev here.
841 * But it's only on the first write fault on the vma, not
842 * every time, and we could devise a way to avoid it later
843 * (e.g. stash info in next's anon_vma_node when assigning
844 * an anon_vma, or when trying vma_merge). Another time.
845 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200846 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 if (!near)
848 goto none;
849
850 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
851 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
852
853 if (near->anon_vma && near->vm_end == vma->vm_start &&
854 mpol_equal(vma_policy(near), vma_policy(vma)) &&
855 can_vma_merge_after(near, vm_flags,
856 NULL, vma->vm_file, vma->vm_pgoff))
857 return near->anon_vma;
858none:
859 /*
860 * There's no absolute need to look only at touching neighbours:
861 * we could search further afield for "compatible" anon_vmas.
862 * But it would probably just be a waste of time searching,
863 * or lead to too many vmas hanging off the same anon_vma.
864 * We're trying to allow mprotect remerging later on,
865 * not trying to minimize memory used for anon_vmas.
866 */
867 return NULL;
868}
869
870#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700871void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 struct file *file, long pages)
873{
874 const unsigned long stack_flags
875 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 if (file) {
878 mm->shared_vm += pages;
879 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
880 mm->exec_vm += pages;
881 } else if (flags & stack_flags)
882 mm->stack_vm += pages;
883 if (flags & (VM_RESERVED|VM_IO))
884 mm->reserved_vm += pages;
885}
886#endif /* CONFIG_PROC_FS */
887
888/*
889 * The caller must hold down_write(current->mm->mmap_sem).
890 */
891
892unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
893 unsigned long len, unsigned long prot,
894 unsigned long flags, unsigned long pgoff)
895{
896 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 struct inode *inode;
898 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 int accountable = 1;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700901 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 /*
904 * Does the application expect PROT_READ to imply PROT_EXEC?
905 *
906 * (the exception is when the underlying filesystem is noexec
907 * mounted, in which case we dont add PROT_EXEC.)
908 */
909 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800910 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 prot |= PROT_EXEC;
912
913 if (!len)
914 return -EINVAL;
915
Kirill Korotaev3a459752006-09-07 14:17:04 +0400916 error = arch_mmap_check(addr, len, flags);
917 if (error)
918 return error;
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 /* Careful about overflows.. */
921 len = PAGE_ALIGN(len);
922 if (!len || len > TASK_SIZE)
923 return -ENOMEM;
924
925 /* offset overflow? */
926 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
927 return -EOVERFLOW;
928
929 /* Too many mappings? */
930 if (mm->map_count > sysctl_max_map_count)
931 return -ENOMEM;
932
933 /* Obtain the address to map to. we verify (or select) it and ensure
934 * that it represents a valid section of the address space.
935 */
936 addr = get_unmapped_area(file, addr, len, pgoff, flags);
937 if (addr & ~PAGE_MASK)
938 return addr;
939
940 /* Do simple checking here so the lower-level routines won't have
941 * to. we assume access permissions have been handled by the open
942 * of the memory object, so we don't do any here.
943 */
944 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
945 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
946
947 if (flags & MAP_LOCKED) {
948 if (!can_do_mlock())
949 return -EPERM;
950 vm_flags |= VM_LOCKED;
951 }
952 /* mlock MCL_FUTURE? */
953 if (vm_flags & VM_LOCKED) {
954 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700955 locked = len >> PAGE_SHIFT;
956 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700958 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
960 return -EAGAIN;
961 }
962
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800963 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
965 if (file) {
966 switch (flags & MAP_TYPE) {
967 case MAP_SHARED:
968 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
969 return -EACCES;
970
971 /*
972 * Make sure we don't allow writing to an append-only
973 * file..
974 */
975 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
976 return -EACCES;
977
978 /*
979 * Make sure there are no mandatory locks on the file.
980 */
981 if (locks_verify_locked(inode))
982 return -EAGAIN;
983
984 vm_flags |= VM_SHARED | VM_MAYSHARE;
985 if (!(file->f_mode & FMODE_WRITE))
986 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
987
988 /* fall through */
989 case MAP_PRIVATE:
990 if (!(file->f_mode & FMODE_READ))
991 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800992 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -0700993 if (vm_flags & VM_EXEC)
994 return -EPERM;
995 vm_flags &= ~VM_MAYEXEC;
996 }
997 if (is_file_hugepages(file))
998 accountable = 0;
999
1000 if (!file->f_op || !file->f_op->mmap)
1001 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 break;
1003
1004 default:
1005 return -EINVAL;
1006 }
1007 } else {
1008 switch (flags & MAP_TYPE) {
1009 case MAP_SHARED:
1010 vm_flags |= VM_SHARED | VM_MAYSHARE;
1011 break;
1012 case MAP_PRIVATE:
1013 /*
1014 * Set pgoff according to addr for anon_vma.
1015 */
1016 pgoff = addr >> PAGE_SHIFT;
1017 break;
1018 default:
1019 return -EINVAL;
1020 }
1021 }
1022
Eric Parised032182007-06-28 15:55:21 -04001023 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 if (error)
1025 return error;
Eric Parised032182007-06-28 15:55:21 -04001026
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001027 return mmap_region(file, addr, len, flags, vm_flags, pgoff,
1028 accountable);
1029}
1030EXPORT_SYMBOL(do_mmap_pgoff);
1031
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001032/*
1033 * Some shared mappigns will want the pages marked read-only
1034 * to track write events. If so, we'll downgrade vm_page_prot
1035 * to the private version (using protection_map[] without the
1036 * VM_SHARED bit).
1037 */
1038int vma_wants_writenotify(struct vm_area_struct *vma)
1039{
1040 unsigned int vm_flags = vma->vm_flags;
1041
1042 /* If it was private or non-writable, the write bit is already clear */
1043 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1044 return 0;
1045
1046 /* The backer wishes to know when pages are first written to? */
1047 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1048 return 1;
1049
1050 /* The open routine did something to the protections already? */
1051 if (pgprot_val(vma->vm_page_prot) !=
1052 pgprot_val(protection_map[vm_flags &
1053 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]))
1054 return 0;
1055
1056 /* Specialty mapping? */
1057 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1058 return 0;
1059
1060 /* Can the mapping track the dirty pages? */
1061 return vma->vm_file && vma->vm_file->f_mapping &&
1062 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1063}
1064
1065
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001066unsigned long mmap_region(struct file *file, unsigned long addr,
1067 unsigned long len, unsigned long flags,
1068 unsigned int vm_flags, unsigned long pgoff,
1069 int accountable)
1070{
1071 struct mm_struct *mm = current->mm;
1072 struct vm_area_struct *vma, *prev;
1073 int correct_wcount = 0;
1074 int error;
1075 struct rb_node **rb_link, *rb_parent;
1076 unsigned long charged = 0;
1077 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 /* Clear old maps */
1080 error = -ENOMEM;
1081munmap_back:
1082 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1083 if (vma && vma->vm_start < addr + len) {
1084 if (do_munmap(mm, addr, len))
1085 return -ENOMEM;
1086 goto munmap_back;
1087 }
1088
1089 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001090 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 return -ENOMEM;
1092
1093 if (accountable && (!(flags & MAP_NORESERVE) ||
1094 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1095 if (vm_flags & VM_SHARED) {
1096 /* Check memory availability in shmem_file_setup? */
1097 vm_flags |= VM_ACCOUNT;
1098 } else if (vm_flags & VM_WRITE) {
1099 /*
1100 * Private writable mapping: check memory availability
1101 */
1102 charged = len >> PAGE_SHIFT;
1103 if (security_vm_enough_memory(charged))
1104 return -ENOMEM;
1105 vm_flags |= VM_ACCOUNT;
1106 }
1107 }
1108
1109 /*
1110 * Can we just expand an old private anonymous mapping?
1111 * The VM_SHARED test is necessary because shmem_zero_setup
1112 * will create the file object for a shared anonymous map below.
1113 */
1114 if (!file && !(vm_flags & VM_SHARED) &&
1115 vma_merge(mm, prev, addr, addr + len, vm_flags,
1116 NULL, NULL, pgoff, NULL))
1117 goto out;
1118
1119 /*
1120 * Determine the object being mapped and call the appropriate
1121 * specific mapper. the address has already been validated, but
1122 * not unmapped, but the maps are removed from the list.
1123 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001124 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 if (!vma) {
1126 error = -ENOMEM;
1127 goto unacct_error;
1128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129
1130 vma->vm_mm = mm;
1131 vma->vm_start = addr;
1132 vma->vm_end = addr + len;
1133 vma->vm_flags = vm_flags;
David Howells9637a5e2006-06-23 02:03:43 -07001134 vma->vm_page_prot = protection_map[vm_flags &
1135 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 vma->vm_pgoff = pgoff;
1137
1138 if (file) {
1139 error = -EINVAL;
1140 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1141 goto free_vma;
1142 if (vm_flags & VM_DENYWRITE) {
1143 error = deny_write_access(file);
1144 if (error)
1145 goto free_vma;
1146 correct_wcount = 1;
1147 }
1148 vma->vm_file = file;
1149 get_file(file);
1150 error = file->f_op->mmap(file, vma);
1151 if (error)
1152 goto unmap_and_free_vma;
1153 } else if (vm_flags & VM_SHARED) {
1154 error = shmem_zero_setup(vma);
1155 if (error)
1156 goto free_vma;
1157 }
1158
1159 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1160 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1161 * that memory reservation must be checked; but that reservation
1162 * belongs to shared memory object, not to vma: so now clear it.
1163 */
1164 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1165 vma->vm_flags &= ~VM_ACCOUNT;
1166
1167 /* Can addr have changed??
1168 *
1169 * Answer: Yes, several device drivers can do it in their
1170 * f_op->mmap method. -DaveM
1171 */
1172 addr = vma->vm_start;
1173 pgoff = vma->vm_pgoff;
1174 vm_flags = vma->vm_flags;
1175
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001176 if (vma_wants_writenotify(vma))
1177 vma->vm_page_prot =
1178 protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC)];
1179
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1181 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1182 file = vma->vm_file;
1183 vma_link(mm, vma, prev, rb_link, rb_parent);
1184 if (correct_wcount)
1185 atomic_inc(&inode->i_writecount);
1186 } else {
1187 if (file) {
1188 if (correct_wcount)
1189 atomic_inc(&inode->i_writecount);
1190 fput(file);
1191 }
1192 mpol_free(vma_policy(vma));
1193 kmem_cache_free(vm_area_cachep, vma);
1194 }
1195out:
1196 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001197 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 if (vm_flags & VM_LOCKED) {
1199 mm->locked_vm += len >> PAGE_SHIFT;
1200 make_pages_present(addr, addr + len);
1201 }
Nick Piggin54cb8822007-07-19 01:46:59 -07001202 if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1203 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 return addr;
1205
1206unmap_and_free_vma:
1207 if (correct_wcount)
1208 atomic_inc(&inode->i_writecount);
1209 vma->vm_file = NULL;
1210 fput(file);
1211
1212 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001213 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1214 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215free_vma:
1216 kmem_cache_free(vm_area_cachep, vma);
1217unacct_error:
1218 if (charged)
1219 vm_unacct_memory(charged);
1220 return error;
1221}
1222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223/* Get an address range which is currently unmapped.
1224 * For shmat() with addr=0.
1225 *
1226 * Ugly calling convention alert:
1227 * Return value with the low bits set means error value,
1228 * ie
1229 * if (ret & ~PAGE_MASK)
1230 * error = ret;
1231 *
1232 * This function "knows" that -ENOMEM has the bits set.
1233 */
1234#ifndef HAVE_ARCH_UNMAPPED_AREA
1235unsigned long
1236arch_get_unmapped_area(struct file *filp, unsigned long addr,
1237 unsigned long len, unsigned long pgoff, unsigned long flags)
1238{
1239 struct mm_struct *mm = current->mm;
1240 struct vm_area_struct *vma;
1241 unsigned long start_addr;
1242
1243 if (len > TASK_SIZE)
1244 return -ENOMEM;
1245
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001246 if (flags & MAP_FIXED)
1247 return addr;
1248
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 if (addr) {
1250 addr = PAGE_ALIGN(addr);
1251 vma = find_vma(mm, addr);
1252 if (TASK_SIZE - len >= addr &&
1253 (!vma || addr + len <= vma->vm_start))
1254 return addr;
1255 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001256 if (len > mm->cached_hole_size) {
1257 start_addr = addr = mm->free_area_cache;
1258 } else {
1259 start_addr = addr = TASK_UNMAPPED_BASE;
1260 mm->cached_hole_size = 0;
1261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
1263full_search:
1264 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1265 /* At this point: (!vma || addr < vma->vm_end). */
1266 if (TASK_SIZE - len < addr) {
1267 /*
1268 * Start a new search - just in case we missed
1269 * some holes.
1270 */
1271 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001272 addr = TASK_UNMAPPED_BASE;
1273 start_addr = addr;
1274 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 goto full_search;
1276 }
1277 return -ENOMEM;
1278 }
1279 if (!vma || addr + len <= vma->vm_start) {
1280 /*
1281 * Remember the place where we stopped the search:
1282 */
1283 mm->free_area_cache = addr + len;
1284 return addr;
1285 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001286 if (addr + mm->cached_hole_size < vma->vm_start)
1287 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 addr = vma->vm_end;
1289 }
1290}
1291#endif
1292
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001293void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
1295 /*
1296 * Is this a new hole at the lowest possible address?
1297 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001298 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1299 mm->free_area_cache = addr;
1300 mm->cached_hole_size = ~0UL;
1301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
1304/*
1305 * This mmap-allocator allocates new areas top-down from below the
1306 * stack's low limit (the base):
1307 */
1308#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1309unsigned long
1310arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1311 const unsigned long len, const unsigned long pgoff,
1312 const unsigned long flags)
1313{
1314 struct vm_area_struct *vma;
1315 struct mm_struct *mm = current->mm;
1316 unsigned long addr = addr0;
1317
1318 /* requested length too big for entire address space */
1319 if (len > TASK_SIZE)
1320 return -ENOMEM;
1321
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001322 if (flags & MAP_FIXED)
1323 return addr;
1324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 /* requesting a specific address */
1326 if (addr) {
1327 addr = PAGE_ALIGN(addr);
1328 vma = find_vma(mm, addr);
1329 if (TASK_SIZE - len >= addr &&
1330 (!vma || addr + len <= vma->vm_start))
1331 return addr;
1332 }
1333
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001334 /* check if free_area_cache is useful for us */
1335 if (len <= mm->cached_hole_size) {
1336 mm->cached_hole_size = 0;
1337 mm->free_area_cache = mm->mmap_base;
1338 }
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 /* either no address requested or can't fit in requested address hole */
1341 addr = mm->free_area_cache;
1342
1343 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001344 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 vma = find_vma(mm, addr-len);
1346 if (!vma || addr <= vma->vm_start)
1347 /* remember the address as a hint for next time */
1348 return (mm->free_area_cache = addr-len);
1349 }
1350
Chris Wright73219d12005-06-21 17:14:52 -07001351 if (mm->mmap_base < len)
1352 goto bottomup;
1353
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 addr = mm->mmap_base-len;
1355
1356 do {
1357 /*
1358 * Lookup failure means no vma is above this address,
1359 * else if new region fits below vma->vm_start,
1360 * return with success:
1361 */
1362 vma = find_vma(mm, addr);
1363 if (!vma || addr+len <= vma->vm_start)
1364 /* remember the address as a hint for next time */
1365 return (mm->free_area_cache = addr);
1366
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001367 /* remember the largest hole we saw so far */
1368 if (addr + mm->cached_hole_size < vma->vm_start)
1369 mm->cached_hole_size = vma->vm_start - addr;
1370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 /* try just below the current vma->vm_start */
1372 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001373 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Chris Wright73219d12005-06-21 17:14:52 -07001375bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 /*
1377 * A failed mmap() very likely causes application failure,
1378 * so fall back to the bottom-up function here. This scenario
1379 * can happen with large stack limits and large mmap()
1380 * allocations.
1381 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001382 mm->cached_hole_size = ~0UL;
1383 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1385 /*
1386 * Restore the topdown base:
1387 */
1388 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001389 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391 return addr;
1392}
1393#endif
1394
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001395void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
1397 /*
1398 * Is this a new hole at the highest possible address?
1399 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001400 if (addr > mm->free_area_cache)
1401 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001404 if (mm->free_area_cache > mm->mmap_base)
1405 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406}
1407
1408unsigned long
1409get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1410 unsigned long pgoff, unsigned long flags)
1411{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001412 unsigned long (*get_area)(struct file *, unsigned long,
1413 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001415 get_area = current->mm->get_unmapped_area;
1416 if (file && file->f_op && file->f_op->get_unmapped_area)
1417 get_area = file->f_op->get_unmapped_area;
1418 addr = get_area(file, addr, len, pgoff, flags);
1419 if (IS_ERR_VALUE(addr))
1420 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001422 if (addr > TASK_SIZE - len)
1423 return -ENOMEM;
1424 if (addr & ~PAGE_MASK)
1425 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001426
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001427 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
1429
1430EXPORT_SYMBOL(get_unmapped_area);
1431
1432/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1433struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1434{
1435 struct vm_area_struct *vma = NULL;
1436
1437 if (mm) {
1438 /* Check the cache first. */
1439 /* (Cache hit rate is typically around 35%.) */
1440 vma = mm->mmap_cache;
1441 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1442 struct rb_node * rb_node;
1443
1444 rb_node = mm->mm_rb.rb_node;
1445 vma = NULL;
1446
1447 while (rb_node) {
1448 struct vm_area_struct * vma_tmp;
1449
1450 vma_tmp = rb_entry(rb_node,
1451 struct vm_area_struct, vm_rb);
1452
1453 if (vma_tmp->vm_end > addr) {
1454 vma = vma_tmp;
1455 if (vma_tmp->vm_start <= addr)
1456 break;
1457 rb_node = rb_node->rb_left;
1458 } else
1459 rb_node = rb_node->rb_right;
1460 }
1461 if (vma)
1462 mm->mmap_cache = vma;
1463 }
1464 }
1465 return vma;
1466}
1467
1468EXPORT_SYMBOL(find_vma);
1469
1470/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1471struct vm_area_struct *
1472find_vma_prev(struct mm_struct *mm, unsigned long addr,
1473 struct vm_area_struct **pprev)
1474{
1475 struct vm_area_struct *vma = NULL, *prev = NULL;
1476 struct rb_node * rb_node;
1477 if (!mm)
1478 goto out;
1479
1480 /* Guard against addr being lower than the first VMA */
1481 vma = mm->mmap;
1482
1483 /* Go through the RB tree quickly. */
1484 rb_node = mm->mm_rb.rb_node;
1485
1486 while (rb_node) {
1487 struct vm_area_struct *vma_tmp;
1488 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1489
1490 if (addr < vma_tmp->vm_end) {
1491 rb_node = rb_node->rb_left;
1492 } else {
1493 prev = vma_tmp;
1494 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1495 break;
1496 rb_node = rb_node->rb_right;
1497 }
1498 }
1499
1500out:
1501 *pprev = prev;
1502 return prev ? prev->vm_next : vma;
1503}
1504
1505/*
1506 * Verify that the stack growth is acceptable and
1507 * update accounting. This is shared with both the
1508 * grow-up and grow-down cases.
1509 */
1510static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1511{
1512 struct mm_struct *mm = vma->vm_mm;
1513 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001514 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001517 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 return -ENOMEM;
1519
1520 /* Stack limit test */
1521 if (size > rlim[RLIMIT_STACK].rlim_cur)
1522 return -ENOMEM;
1523
1524 /* mlock limit tests */
1525 if (vma->vm_flags & VM_LOCKED) {
1526 unsigned long locked;
1527 unsigned long limit;
1528 locked = mm->locked_vm + grow;
1529 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1530 if (locked > limit && !capable(CAP_IPC_LOCK))
1531 return -ENOMEM;
1532 }
1533
Adam Litke0d59a012007-01-30 14:35:39 -08001534 /* Check to ensure the stack will not grow into a hugetlb-only region */
1535 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1536 vma->vm_end - size;
1537 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1538 return -EFAULT;
1539
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 /*
1541 * Overcommit.. This must be the final test, as it will
1542 * update security statistics.
1543 */
1544 if (security_vm_enough_memory(grow))
1545 return -ENOMEM;
1546
1547 /* Ok, everything looks good - let it rip */
1548 mm->total_vm += grow;
1549 if (vma->vm_flags & VM_LOCKED)
1550 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001551 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 return 0;
1553}
1554
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001555#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001557 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1558 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001560#ifndef CONFIG_IA64
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001561static inline
1562#endif
1563int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564{
1565 int error;
1566
1567 if (!(vma->vm_flags & VM_GROWSUP))
1568 return -EFAULT;
1569
1570 /*
1571 * We must make sure the anon_vma is allocated
1572 * so that the anon_vma locking is not a noop.
1573 */
1574 if (unlikely(anon_vma_prepare(vma)))
1575 return -ENOMEM;
1576 anon_vma_lock(vma);
1577
1578 /*
1579 * vma->vm_start/vm_end cannot change under us because the caller
1580 * is required to hold the mmap_sem in read mode. We need the
1581 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001582 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 */
Helge Deller06b32f32006-12-19 19:28:33 +01001584 if (address < PAGE_ALIGN(address+4))
1585 address = PAGE_ALIGN(address+4);
1586 else {
1587 anon_vma_unlock(vma);
1588 return -ENOMEM;
1589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 error = 0;
1591
1592 /* Somebody else might have raced and expanded it already */
1593 if (address > vma->vm_end) {
1594 unsigned long size, grow;
1595
1596 size = address - vma->vm_start;
1597 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1598
1599 error = acct_stack_growth(vma, size, grow);
1600 if (!error)
1601 vma->vm_end = address;
1602 }
1603 anon_vma_unlock(vma);
1604 return error;
1605}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001606#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1607
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608/*
1609 * vma is the first one with address < vma->vm_start. Have to extend vma.
1610 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001611static inline int expand_downwards(struct vm_area_struct *vma,
1612 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
1614 int error;
1615
1616 /*
1617 * We must make sure the anon_vma is allocated
1618 * so that the anon_vma locking is not a noop.
1619 */
1620 if (unlikely(anon_vma_prepare(vma)))
1621 return -ENOMEM;
1622 anon_vma_lock(vma);
1623
1624 /*
1625 * vma->vm_start/vm_end cannot change under us because the caller
1626 * is required to hold the mmap_sem in read mode. We need the
1627 * anon_vma lock to serialize against concurrent expand_stacks.
1628 */
1629 address &= PAGE_MASK;
1630 error = 0;
1631
1632 /* Somebody else might have raced and expanded it already */
1633 if (address < vma->vm_start) {
1634 unsigned long size, grow;
1635
1636 size = vma->vm_end - address;
1637 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1638
1639 error = acct_stack_growth(vma, size, grow);
1640 if (!error) {
1641 vma->vm_start = address;
1642 vma->vm_pgoff -= grow;
1643 }
1644 }
1645 anon_vma_unlock(vma);
1646 return error;
1647}
1648
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001649int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1650{
1651 return expand_downwards(vma, address);
1652}
1653
1654#ifdef CONFIG_STACK_GROWSUP
1655int expand_stack(struct vm_area_struct *vma, unsigned long address)
1656{
1657 return expand_upwards(vma, address);
1658}
1659
1660struct vm_area_struct *
1661find_extend_vma(struct mm_struct *mm, unsigned long addr)
1662{
1663 struct vm_area_struct *vma, *prev;
1664
1665 addr &= PAGE_MASK;
1666 vma = find_vma_prev(mm, addr, &prev);
1667 if (vma && (vma->vm_start <= addr))
1668 return vma;
1669 if (!prev || expand_stack(prev, addr))
1670 return NULL;
1671 if (prev->vm_flags & VM_LOCKED)
1672 make_pages_present(addr, prev->vm_end);
1673 return prev;
1674}
1675#else
1676int expand_stack(struct vm_area_struct *vma, unsigned long address)
1677{
1678 return expand_downwards(vma, address);
1679}
1680
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681struct vm_area_struct *
1682find_extend_vma(struct mm_struct * mm, unsigned long addr)
1683{
1684 struct vm_area_struct * vma;
1685 unsigned long start;
1686
1687 addr &= PAGE_MASK;
1688 vma = find_vma(mm,addr);
1689 if (!vma)
1690 return NULL;
1691 if (vma->vm_start <= addr)
1692 return vma;
1693 if (!(vma->vm_flags & VM_GROWSDOWN))
1694 return NULL;
1695 start = vma->vm_start;
1696 if (expand_stack(vma, addr))
1697 return NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001698 if (vma->vm_flags & VM_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 make_pages_present(addr, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 return vma;
1701}
1702#endif
1703
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001705 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001707 *
1708 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001710static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001712 /* Update high watermark before we lower total_vm */
1713 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001715 long nrpages = vma_pages(vma);
1716
1717 mm->total_vm -= nrpages;
1718 if (vma->vm_flags & VM_LOCKED)
1719 mm->locked_vm -= nrpages;
1720 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001721 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001722 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 validate_mm(mm);
1724}
1725
1726/*
1727 * Get rid of page table information in the indicated region.
1728 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001729 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 */
1731static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001732 struct vm_area_struct *vma, struct vm_area_struct *prev,
1733 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001735 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 struct mmu_gather *tlb;
1737 unsigned long nr_accounted = 0;
1738
1739 lru_add_drain();
1740 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001741 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001742 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001744 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001745 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 tlb_finish_mmu(tlb, start, end);
1747}
1748
1749/*
1750 * Create a list of vma's touched by the unmap, removing them from the mm's
1751 * vma list as we go..
1752 */
1753static void
1754detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1755 struct vm_area_struct *prev, unsigned long end)
1756{
1757 struct vm_area_struct **insertion_point;
1758 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001759 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
1761 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1762 do {
1763 rb_erase(&vma->vm_rb, &mm->mm_rb);
1764 mm->map_count--;
1765 tail_vma = vma;
1766 vma = vma->vm_next;
1767 } while (vma && vma->vm_start < end);
1768 *insertion_point = vma;
1769 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001770 if (mm->unmap_area == arch_unmap_area)
1771 addr = prev ? prev->vm_end : mm->mmap_base;
1772 else
1773 addr = vma ? vma->vm_start : mm->mmap_base;
1774 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 mm->mmap_cache = NULL; /* Kill the cache. */
1776}
1777
1778/*
1779 * Split a vma into two pieces at address 'addr', a new vma is allocated
Michael Opdenacker59c51592007-05-09 08:57:56 +02001780 * either for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 */
1782int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1783 unsigned long addr, int new_below)
1784{
1785 struct mempolicy *pol;
1786 struct vm_area_struct *new;
1787
1788 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1789 return -EINVAL;
1790
1791 if (mm->map_count >= sysctl_max_map_count)
1792 return -ENOMEM;
1793
Christoph Lametere94b1762006-12-06 20:33:17 -08001794 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 if (!new)
1796 return -ENOMEM;
1797
1798 /* most fields are the same, copy all, and then fixup */
1799 *new = *vma;
1800
1801 if (new_below)
1802 new->vm_end = addr;
1803 else {
1804 new->vm_start = addr;
1805 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1806 }
1807
1808 pol = mpol_copy(vma_policy(vma));
1809 if (IS_ERR(pol)) {
1810 kmem_cache_free(vm_area_cachep, new);
1811 return PTR_ERR(pol);
1812 }
1813 vma_set_policy(new, pol);
1814
1815 if (new->vm_file)
1816 get_file(new->vm_file);
1817
1818 if (new->vm_ops && new->vm_ops->open)
1819 new->vm_ops->open(new);
1820
1821 if (new_below)
1822 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1823 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1824 else
1825 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1826
1827 return 0;
1828}
1829
1830/* Munmap is split into 2 main parts -- this part which finds
1831 * what needs doing, and the areas themselves, which do the
1832 * work. This now handles partial unmappings.
1833 * Jeremy Fitzhardinge <jeremy@goop.org>
1834 */
1835int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1836{
1837 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001838 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
1840 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1841 return -EINVAL;
1842
1843 if ((len = PAGE_ALIGN(len)) == 0)
1844 return -EINVAL;
1845
1846 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001847 vma = find_vma_prev(mm, start, &prev);
1848 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001850 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
1852 /* if it doesn't overlap, we have nothing.. */
1853 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001854 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 return 0;
1856
1857 /*
1858 * If we need to split any vma, do it now to save pain later.
1859 *
1860 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1861 * unmapped vm_area_struct will remain in use: so lower split_vma
1862 * places tmp vma above, and higher split_vma places tmp vma below.
1863 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001864 if (start > vma->vm_start) {
1865 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 if (error)
1867 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001868 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 }
1870
1871 /* Does it split the last one? */
1872 last = find_vma(mm, end);
1873 if (last && end > last->vm_start) {
1874 int error = split_vma(mm, last, end, 1);
1875 if (error)
1876 return error;
1877 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001878 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
1880 /*
1881 * Remove the vma's, and unmap the actual pages
1882 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001883 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1884 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
1886 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001887 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
1889 return 0;
1890}
1891
1892EXPORT_SYMBOL(do_munmap);
1893
1894asmlinkage long sys_munmap(unsigned long addr, size_t len)
1895{
1896 int ret;
1897 struct mm_struct *mm = current->mm;
1898
1899 profile_munmap(addr);
1900
1901 down_write(&mm->mmap_sem);
1902 ret = do_munmap(mm, addr, len);
1903 up_write(&mm->mmap_sem);
1904 return ret;
1905}
1906
1907static inline void verify_mm_writelocked(struct mm_struct *mm)
1908{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001909#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1911 WARN_ON(1);
1912 up_read(&mm->mmap_sem);
1913 }
1914#endif
1915}
1916
1917/*
1918 * this is really a simplified "do_mmap". it only handles
1919 * anonymous maps. eventually we may be able to do some
1920 * brk-specific accounting here.
1921 */
1922unsigned long do_brk(unsigned long addr, unsigned long len)
1923{
1924 struct mm_struct * mm = current->mm;
1925 struct vm_area_struct * vma, * prev;
1926 unsigned long flags;
1927 struct rb_node ** rb_link, * rb_parent;
1928 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04001929 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
1931 len = PAGE_ALIGN(len);
1932 if (!len)
1933 return addr;
1934
1935 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1936 return -EINVAL;
1937
Hugh Dickinscd2579d2006-11-14 13:43:38 +00001938 if (is_hugepage_only_range(mm, addr, len))
1939 return -EINVAL;
David Gibsoncb07c9a2006-11-14 02:03:38 -08001940
Kirill Korotaev3a459752006-09-07 14:17:04 +04001941 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1942
1943 error = arch_mmap_check(addr, len, flags);
1944 if (error)
1945 return error;
1946
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 /*
1948 * mlock MCL_FUTURE?
1949 */
1950 if (mm->def_flags & VM_LOCKED) {
1951 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001952 locked = len >> PAGE_SHIFT;
1953 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07001955 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1957 return -EAGAIN;
1958 }
1959
1960 /*
1961 * mm->mmap_sem is required to protect against another thread
1962 * changing the mappings in case we sleep.
1963 */
1964 verify_mm_writelocked(mm);
1965
1966 /*
1967 * Clear old maps. this also does some error checking for us
1968 */
1969 munmap_back:
1970 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1971 if (vma && vma->vm_start < addr + len) {
1972 if (do_munmap(mm, addr, len))
1973 return -ENOMEM;
1974 goto munmap_back;
1975 }
1976
1977 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001978 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 return -ENOMEM;
1980
1981 if (mm->map_count > sysctl_max_map_count)
1982 return -ENOMEM;
1983
1984 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1985 return -ENOMEM;
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 /* Can we just expand an old private anonymous mapping? */
1988 if (vma_merge(mm, prev, addr, addr + len, flags,
1989 NULL, NULL, pgoff, NULL))
1990 goto out;
1991
1992 /*
1993 * create a vma struct for an anonymous mapping
1994 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001995 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 if (!vma) {
1997 vm_unacct_memory(len >> PAGE_SHIFT);
1998 return -ENOMEM;
1999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
2001 vma->vm_mm = mm;
2002 vma->vm_start = addr;
2003 vma->vm_end = addr + len;
2004 vma->vm_pgoff = pgoff;
2005 vma->vm_flags = flags;
David Howells9637a5e2006-06-23 02:03:43 -07002006 vma->vm_page_prot = protection_map[flags &
2007 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 vma_link(mm, vma, prev, rb_link, rb_parent);
2009out:
2010 mm->total_vm += len >> PAGE_SHIFT;
2011 if (flags & VM_LOCKED) {
2012 mm->locked_vm += len >> PAGE_SHIFT;
2013 make_pages_present(addr, addr + len);
2014 }
2015 return addr;
2016}
2017
2018EXPORT_SYMBOL(do_brk);
2019
2020/* Release all mmaps. */
2021void exit_mmap(struct mm_struct *mm)
2022{
2023 struct mmu_gather *tlb;
Hugh Dickinse0da3822005-04-19 13:29:15 -07002024 struct vm_area_struct *vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002026 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002028 /* mm's last user has gone, and its about to be pulled down */
2029 arch_exit_mmap(mm);
2030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002033 tlb = tlb_gather_mmu(mm, 1);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002034 /* Don't update_hiwater_rss(mm) here, do_exit already did */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002035 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002036 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002038 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07002039 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002042 * Walk the list again, actually closing and freeing it,
2043 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002045 while (vma)
2046 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002047
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002048 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049}
2050
2051/* Insert vm structure into process list sorted by address
2052 * and into the inode's i_mmap tree. If vm_file is non-NULL
2053 * then i_mmap_lock is taken here.
2054 */
2055int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2056{
2057 struct vm_area_struct * __vma, * prev;
2058 struct rb_node ** rb_link, * rb_parent;
2059
2060 /*
2061 * The vm_pgoff of a purely anonymous vma should be irrelevant
2062 * until its first write fault, when page's anon_vma and index
2063 * are set. But now set the vm_pgoff it will almost certainly
2064 * end up with (unless mremap moves it elsewhere before that
2065 * first wfault), so /proc/pid/maps tells a consistent story.
2066 *
2067 * By setting it to reflect the virtual start address of the
2068 * vma, merges and splits can happen in a seamless way, just
2069 * using the existing file pgoff checks and manipulations.
2070 * Similarly in do_mmap_pgoff and in do_brk.
2071 */
2072 if (!vma->vm_file) {
2073 BUG_ON(vma->anon_vma);
2074 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2075 }
2076 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2077 if (__vma && __vma->vm_start < vma->vm_end)
2078 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002079 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002080 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002081 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 vma_link(mm, vma, prev, rb_link, rb_parent);
2083 return 0;
2084}
2085
2086/*
2087 * Copy the vma structure to a new location in the same mm,
2088 * prior to moving page table entries, to effect an mremap move.
2089 */
2090struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2091 unsigned long addr, unsigned long len, pgoff_t pgoff)
2092{
2093 struct vm_area_struct *vma = *vmap;
2094 unsigned long vma_start = vma->vm_start;
2095 struct mm_struct *mm = vma->vm_mm;
2096 struct vm_area_struct *new_vma, *prev;
2097 struct rb_node **rb_link, *rb_parent;
2098 struct mempolicy *pol;
2099
2100 /*
2101 * If anonymous vma has not yet been faulted, update new pgoff
2102 * to match new location, to increase its chance of merging.
2103 */
2104 if (!vma->vm_file && !vma->anon_vma)
2105 pgoff = addr >> PAGE_SHIFT;
2106
2107 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2108 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2109 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2110 if (new_vma) {
2111 /*
2112 * Source vma may have been merged into new_vma
2113 */
2114 if (vma_start >= new_vma->vm_start &&
2115 vma_start < new_vma->vm_end)
2116 *vmap = new_vma;
2117 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002118 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 if (new_vma) {
2120 *new_vma = *vma;
2121 pol = mpol_copy(vma_policy(vma));
2122 if (IS_ERR(pol)) {
2123 kmem_cache_free(vm_area_cachep, new_vma);
2124 return NULL;
2125 }
2126 vma_set_policy(new_vma, pol);
2127 new_vma->vm_start = addr;
2128 new_vma->vm_end = addr + len;
2129 new_vma->vm_pgoff = pgoff;
2130 if (new_vma->vm_file)
2131 get_file(new_vma->vm_file);
2132 if (new_vma->vm_ops && new_vma->vm_ops->open)
2133 new_vma->vm_ops->open(new_vma);
2134 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2135 }
2136 }
2137 return new_vma;
2138}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002139
2140/*
2141 * Return true if the calling process may expand its vm space by the passed
2142 * number of pages
2143 */
2144int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2145{
2146 unsigned long cur = mm->total_vm; /* pages */
2147 unsigned long lim;
2148
2149 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2150
2151 if (cur + npages > lim)
2152 return 0;
2153 return 1;
2154}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002155
2156
2157static struct page *special_mapping_nopage(struct vm_area_struct *vma,
2158 unsigned long address, int *type)
2159{
2160 struct page **pages;
2161
2162 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
2163
2164 address -= vma->vm_start;
2165 for (pages = vma->vm_private_data; address > 0 && *pages; ++pages)
2166 address -= PAGE_SIZE;
2167
2168 if (*pages) {
2169 struct page *page = *pages;
2170 get_page(page);
2171 return page;
2172 }
2173
2174 return NOPAGE_SIGBUS;
2175}
2176
2177/*
2178 * Having a close hook prevents vma merging regardless of flags.
2179 */
2180static void special_mapping_close(struct vm_area_struct *vma)
2181{
2182}
2183
2184static struct vm_operations_struct special_mapping_vmops = {
2185 .close = special_mapping_close,
2186 .nopage = special_mapping_nopage,
2187};
2188
2189/*
2190 * Called with mm->mmap_sem held for writing.
2191 * Insert a new vma covering the given region, with the given flags.
2192 * Its pages are supplied by the given array of struct page *.
2193 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2194 * The region past the last page supplied will always produce SIGBUS.
2195 * The array pointer and the pages it points to are assumed to stay alive
2196 * for as long as this mapping might exist.
2197 */
2198int install_special_mapping(struct mm_struct *mm,
2199 unsigned long addr, unsigned long len,
2200 unsigned long vm_flags, struct page **pages)
2201{
2202 struct vm_area_struct *vma;
2203
2204 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2205 if (unlikely(vma == NULL))
2206 return -ENOMEM;
2207
2208 vma->vm_mm = mm;
2209 vma->vm_start = addr;
2210 vma->vm_end = addr + len;
2211
2212 vma->vm_flags = vm_flags | mm->def_flags;
2213 vma->vm_page_prot = protection_map[vma->vm_flags & 7];
2214
2215 vma->vm_ops = &special_mapping_vmops;
2216 vma->vm_private_data = pages;
2217
2218 if (unlikely(insert_vm_struct(mm, vma))) {
2219 kmem_cache_free(vm_area_cachep, vma);
2220 return -ENOMEM;
2221 }
2222
2223 mm->total_vm += len >> PAGE_SHIFT;
2224
2225 return 0;
2226}