blob: 497e502dfd6b6e54fc2bf902ea55fac36e70ac72 [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 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700119 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
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
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))
910 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
911 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
963 inode = file ? file->f_dentry->d_inode : NULL;
964
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;
Linus Torvalds80c56062006-10-15 14:09:55 -0700992 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
993 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
1023 error = security_file_mmap(file, reqprot, prot, flags);
1024 if (error)
1025 return error;
1026
1027 /* Clear old maps */
1028 error = -ENOMEM;
1029munmap_back:
1030 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1031 if (vma && vma->vm_start < addr + len) {
1032 if (do_munmap(mm, addr, len))
1033 return -ENOMEM;
1034 goto munmap_back;
1035 }
1036
1037 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001038 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 return -ENOMEM;
1040
1041 if (accountable && (!(flags & MAP_NORESERVE) ||
1042 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1043 if (vm_flags & VM_SHARED) {
1044 /* Check memory availability in shmem_file_setup? */
1045 vm_flags |= VM_ACCOUNT;
1046 } else if (vm_flags & VM_WRITE) {
1047 /*
1048 * Private writable mapping: check memory availability
1049 */
1050 charged = len >> PAGE_SHIFT;
1051 if (security_vm_enough_memory(charged))
1052 return -ENOMEM;
1053 vm_flags |= VM_ACCOUNT;
1054 }
1055 }
1056
1057 /*
1058 * Can we just expand an old private anonymous mapping?
1059 * The VM_SHARED test is necessary because shmem_zero_setup
1060 * will create the file object for a shared anonymous map below.
1061 */
1062 if (!file && !(vm_flags & VM_SHARED) &&
1063 vma_merge(mm, prev, addr, addr + len, vm_flags,
1064 NULL, NULL, pgoff, NULL))
1065 goto out;
1066
1067 /*
1068 * Determine the object being mapped and call the appropriate
1069 * specific mapper. the address has already been validated, but
1070 * not unmapped, but the maps are removed from the list.
1071 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001072 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (!vma) {
1074 error = -ENOMEM;
1075 goto unacct_error;
1076 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 vma->vm_mm = mm;
1079 vma->vm_start = addr;
1080 vma->vm_end = addr + len;
1081 vma->vm_flags = vm_flags;
David Howells9637a5e2006-06-23 02:03:43 -07001082 vma->vm_page_prot = protection_map[vm_flags &
1083 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 vma->vm_pgoff = pgoff;
1085
1086 if (file) {
1087 error = -EINVAL;
1088 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1089 goto free_vma;
1090 if (vm_flags & VM_DENYWRITE) {
1091 error = deny_write_access(file);
1092 if (error)
1093 goto free_vma;
1094 correct_wcount = 1;
1095 }
1096 vma->vm_file = file;
1097 get_file(file);
1098 error = file->f_op->mmap(file, vma);
1099 if (error)
1100 goto unmap_and_free_vma;
1101 } else if (vm_flags & VM_SHARED) {
1102 error = shmem_zero_setup(vma);
1103 if (error)
1104 goto free_vma;
1105 }
1106
1107 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1108 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1109 * that memory reservation must be checked; but that reservation
1110 * belongs to shared memory object, not to vma: so now clear it.
1111 */
1112 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1113 vma->vm_flags &= ~VM_ACCOUNT;
1114
1115 /* Can addr have changed??
1116 *
1117 * Answer: Yes, several device drivers can do it in their
1118 * f_op->mmap method. -DaveM
1119 */
1120 addr = vma->vm_start;
1121 pgoff = vma->vm_pgoff;
1122 vm_flags = vma->vm_flags;
1123
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001124 if (vma_wants_writenotify(vma))
1125 vma->vm_page_prot =
1126 protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC)];
1127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1129 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1130 file = vma->vm_file;
1131 vma_link(mm, vma, prev, rb_link, rb_parent);
1132 if (correct_wcount)
1133 atomic_inc(&inode->i_writecount);
1134 } else {
1135 if (file) {
1136 if (correct_wcount)
1137 atomic_inc(&inode->i_writecount);
1138 fput(file);
1139 }
1140 mpol_free(vma_policy(vma));
1141 kmem_cache_free(vm_area_cachep, vma);
1142 }
1143out:
1144 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001145 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 if (vm_flags & VM_LOCKED) {
1147 mm->locked_vm += len >> PAGE_SHIFT;
1148 make_pages_present(addr, addr + len);
1149 }
1150 if (flags & MAP_POPULATE) {
1151 up_write(&mm->mmap_sem);
1152 sys_remap_file_pages(addr, len, 0,
1153 pgoff, flags & MAP_NONBLOCK);
1154 down_write(&mm->mmap_sem);
1155 }
1156 return addr;
1157
1158unmap_and_free_vma:
1159 if (correct_wcount)
1160 atomic_inc(&inode->i_writecount);
1161 vma->vm_file = NULL;
1162 fput(file);
1163
1164 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001165 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1166 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167free_vma:
1168 kmem_cache_free(vm_area_cachep, vma);
1169unacct_error:
1170 if (charged)
1171 vm_unacct_memory(charged);
1172 return error;
1173}
1174
1175EXPORT_SYMBOL(do_mmap_pgoff);
1176
1177/* Get an address range which is currently unmapped.
1178 * For shmat() with addr=0.
1179 *
1180 * Ugly calling convention alert:
1181 * Return value with the low bits set means error value,
1182 * ie
1183 * if (ret & ~PAGE_MASK)
1184 * error = ret;
1185 *
1186 * This function "knows" that -ENOMEM has the bits set.
1187 */
1188#ifndef HAVE_ARCH_UNMAPPED_AREA
1189unsigned long
1190arch_get_unmapped_area(struct file *filp, unsigned long addr,
1191 unsigned long len, unsigned long pgoff, unsigned long flags)
1192{
1193 struct mm_struct *mm = current->mm;
1194 struct vm_area_struct *vma;
1195 unsigned long start_addr;
1196
1197 if (len > TASK_SIZE)
1198 return -ENOMEM;
1199
1200 if (addr) {
1201 addr = PAGE_ALIGN(addr);
1202 vma = find_vma(mm, addr);
1203 if (TASK_SIZE - len >= addr &&
1204 (!vma || addr + len <= vma->vm_start))
1205 return addr;
1206 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001207 if (len > mm->cached_hole_size) {
1208 start_addr = addr = mm->free_area_cache;
1209 } else {
1210 start_addr = addr = TASK_UNMAPPED_BASE;
1211 mm->cached_hole_size = 0;
1212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
1214full_search:
1215 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1216 /* At this point: (!vma || addr < vma->vm_end). */
1217 if (TASK_SIZE - len < addr) {
1218 /*
1219 * Start a new search - just in case we missed
1220 * some holes.
1221 */
1222 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001223 addr = TASK_UNMAPPED_BASE;
1224 start_addr = addr;
1225 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 goto full_search;
1227 }
1228 return -ENOMEM;
1229 }
1230 if (!vma || addr + len <= vma->vm_start) {
1231 /*
1232 * Remember the place where we stopped the search:
1233 */
1234 mm->free_area_cache = addr + len;
1235 return addr;
1236 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001237 if (addr + mm->cached_hole_size < vma->vm_start)
1238 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 addr = vma->vm_end;
1240 }
1241}
1242#endif
1243
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001244void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
1246 /*
1247 * Is this a new hole at the lowest possible address?
1248 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001249 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1250 mm->free_area_cache = addr;
1251 mm->cached_hole_size = ~0UL;
1252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
1254
1255/*
1256 * This mmap-allocator allocates new areas top-down from below the
1257 * stack's low limit (the base):
1258 */
1259#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1260unsigned long
1261arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1262 const unsigned long len, const unsigned long pgoff,
1263 const unsigned long flags)
1264{
1265 struct vm_area_struct *vma;
1266 struct mm_struct *mm = current->mm;
1267 unsigned long addr = addr0;
1268
1269 /* requested length too big for entire address space */
1270 if (len > TASK_SIZE)
1271 return -ENOMEM;
1272
1273 /* requesting a specific address */
1274 if (addr) {
1275 addr = PAGE_ALIGN(addr);
1276 vma = find_vma(mm, addr);
1277 if (TASK_SIZE - len >= addr &&
1278 (!vma || addr + len <= vma->vm_start))
1279 return addr;
1280 }
1281
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001282 /* check if free_area_cache is useful for us */
1283 if (len <= mm->cached_hole_size) {
1284 mm->cached_hole_size = 0;
1285 mm->free_area_cache = mm->mmap_base;
1286 }
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 /* either no address requested or can't fit in requested address hole */
1289 addr = mm->free_area_cache;
1290
1291 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001292 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 vma = find_vma(mm, addr-len);
1294 if (!vma || addr <= vma->vm_start)
1295 /* remember the address as a hint for next time */
1296 return (mm->free_area_cache = addr-len);
1297 }
1298
Chris Wright73219d12005-06-21 17:14:52 -07001299 if (mm->mmap_base < len)
1300 goto bottomup;
1301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 addr = mm->mmap_base-len;
1303
1304 do {
1305 /*
1306 * Lookup failure means no vma is above this address,
1307 * else if new region fits below vma->vm_start,
1308 * return with success:
1309 */
1310 vma = find_vma(mm, addr);
1311 if (!vma || addr+len <= vma->vm_start)
1312 /* remember the address as a hint for next time */
1313 return (mm->free_area_cache = addr);
1314
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001315 /* remember the largest hole we saw so far */
1316 if (addr + mm->cached_hole_size < vma->vm_start)
1317 mm->cached_hole_size = vma->vm_start - addr;
1318
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 /* try just below the current vma->vm_start */
1320 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001321 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Chris Wright73219d12005-06-21 17:14:52 -07001323bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 /*
1325 * A failed mmap() very likely causes application failure,
1326 * so fall back to the bottom-up function here. This scenario
1327 * can happen with large stack limits and large mmap()
1328 * allocations.
1329 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001330 mm->cached_hole_size = ~0UL;
1331 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1333 /*
1334 * Restore the topdown base:
1335 */
1336 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001337 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339 return addr;
1340}
1341#endif
1342
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001343void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344{
1345 /*
1346 * Is this a new hole at the highest possible address?
1347 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001348 if (addr > mm->free_area_cache)
1349 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
1351 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001352 if (mm->free_area_cache > mm->mmap_base)
1353 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354}
1355
1356unsigned long
1357get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1358 unsigned long pgoff, unsigned long flags)
1359{
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001360 unsigned long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001362 if (!(flags & MAP_FIXED)) {
1363 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1364
1365 get_area = current->mm->get_unmapped_area;
1366 if (file && file->f_op && file->f_op->get_unmapped_area)
1367 get_area = file->f_op->get_unmapped_area;
1368 addr = get_area(file, addr, len, pgoff, flags);
1369 if (IS_ERR_VALUE(addr))
1370 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001373 if (addr > TASK_SIZE - len)
1374 return -ENOMEM;
1375 if (addr & ~PAGE_MASK)
1376 return -EINVAL;
1377 if (file && is_file_hugepages(file)) {
1378 /*
1379 * Check if the given range is hugepage aligned, and
1380 * can be made suitable for hugepages.
1381 */
1382 ret = prepare_hugepage_range(addr, len);
1383 } else {
1384 /*
1385 * Ensure that a normal request is not falling in a
1386 * reserved hugepage range. For some archs like IA-64,
1387 * there is a separate region for hugepages.
1388 */
1389 ret = is_hugepage_only_range(current->mm, addr, len);
1390 }
1391 if (ret)
1392 return -EINVAL;
1393 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394}
1395
1396EXPORT_SYMBOL(get_unmapped_area);
1397
1398/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1399struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1400{
1401 struct vm_area_struct *vma = NULL;
1402
1403 if (mm) {
1404 /* Check the cache first. */
1405 /* (Cache hit rate is typically around 35%.) */
1406 vma = mm->mmap_cache;
1407 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1408 struct rb_node * rb_node;
1409
1410 rb_node = mm->mm_rb.rb_node;
1411 vma = NULL;
1412
1413 while (rb_node) {
1414 struct vm_area_struct * vma_tmp;
1415
1416 vma_tmp = rb_entry(rb_node,
1417 struct vm_area_struct, vm_rb);
1418
1419 if (vma_tmp->vm_end > addr) {
1420 vma = vma_tmp;
1421 if (vma_tmp->vm_start <= addr)
1422 break;
1423 rb_node = rb_node->rb_left;
1424 } else
1425 rb_node = rb_node->rb_right;
1426 }
1427 if (vma)
1428 mm->mmap_cache = vma;
1429 }
1430 }
1431 return vma;
1432}
1433
1434EXPORT_SYMBOL(find_vma);
1435
1436/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1437struct vm_area_struct *
1438find_vma_prev(struct mm_struct *mm, unsigned long addr,
1439 struct vm_area_struct **pprev)
1440{
1441 struct vm_area_struct *vma = NULL, *prev = NULL;
1442 struct rb_node * rb_node;
1443 if (!mm)
1444 goto out;
1445
1446 /* Guard against addr being lower than the first VMA */
1447 vma = mm->mmap;
1448
1449 /* Go through the RB tree quickly. */
1450 rb_node = mm->mm_rb.rb_node;
1451
1452 while (rb_node) {
1453 struct vm_area_struct *vma_tmp;
1454 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1455
1456 if (addr < vma_tmp->vm_end) {
1457 rb_node = rb_node->rb_left;
1458 } else {
1459 prev = vma_tmp;
1460 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1461 break;
1462 rb_node = rb_node->rb_right;
1463 }
1464 }
1465
1466out:
1467 *pprev = prev;
1468 return prev ? prev->vm_next : vma;
1469}
1470
1471/*
1472 * Verify that the stack growth is acceptable and
1473 * update accounting. This is shared with both the
1474 * grow-up and grow-down cases.
1475 */
1476static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1477{
1478 struct mm_struct *mm = vma->vm_mm;
1479 struct rlimit *rlim = current->signal->rlim;
1480
1481 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001482 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 return -ENOMEM;
1484
1485 /* Stack limit test */
1486 if (size > rlim[RLIMIT_STACK].rlim_cur)
1487 return -ENOMEM;
1488
1489 /* mlock limit tests */
1490 if (vma->vm_flags & VM_LOCKED) {
1491 unsigned long locked;
1492 unsigned long limit;
1493 locked = mm->locked_vm + grow;
1494 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1495 if (locked > limit && !capable(CAP_IPC_LOCK))
1496 return -ENOMEM;
1497 }
1498
1499 /*
1500 * Overcommit.. This must be the final test, as it will
1501 * update security statistics.
1502 */
1503 if (security_vm_enough_memory(grow))
1504 return -ENOMEM;
1505
1506 /* Ok, everything looks good - let it rip */
1507 mm->total_vm += grow;
1508 if (vma->vm_flags & VM_LOCKED)
1509 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001510 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return 0;
1512}
1513
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001514#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001516 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1517 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001519#ifndef CONFIG_IA64
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001520static inline
1521#endif
1522int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523{
1524 int error;
1525
1526 if (!(vma->vm_flags & VM_GROWSUP))
1527 return -EFAULT;
1528
1529 /*
1530 * We must make sure the anon_vma is allocated
1531 * so that the anon_vma locking is not a noop.
1532 */
1533 if (unlikely(anon_vma_prepare(vma)))
1534 return -ENOMEM;
1535 anon_vma_lock(vma);
1536
1537 /*
1538 * vma->vm_start/vm_end cannot change under us because the caller
1539 * is required to hold the mmap_sem in read mode. We need the
1540 * anon_vma lock to serialize against concurrent expand_stacks.
1541 */
1542 address += 4 + PAGE_SIZE - 1;
1543 address &= PAGE_MASK;
1544 error = 0;
1545
1546 /* Somebody else might have raced and expanded it already */
1547 if (address > vma->vm_end) {
1548 unsigned long size, grow;
1549
1550 size = address - vma->vm_start;
1551 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1552
1553 error = acct_stack_growth(vma, size, grow);
1554 if (!error)
1555 vma->vm_end = address;
1556 }
1557 anon_vma_unlock(vma);
1558 return error;
1559}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001560#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1561
1562#ifdef CONFIG_STACK_GROWSUP
1563int expand_stack(struct vm_area_struct *vma, unsigned long address)
1564{
1565 return expand_upwards(vma, address);
1566}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568struct vm_area_struct *
1569find_extend_vma(struct mm_struct *mm, unsigned long addr)
1570{
1571 struct vm_area_struct *vma, *prev;
1572
1573 addr &= PAGE_MASK;
1574 vma = find_vma_prev(mm, addr, &prev);
1575 if (vma && (vma->vm_start <= addr))
1576 return vma;
1577 if (!prev || expand_stack(prev, addr))
1578 return NULL;
1579 if (prev->vm_flags & VM_LOCKED) {
1580 make_pages_present(addr, prev->vm_end);
1581 }
1582 return prev;
1583}
1584#else
1585/*
1586 * vma is the first one with address < vma->vm_start. Have to extend vma.
1587 */
1588int expand_stack(struct vm_area_struct *vma, unsigned long address)
1589{
1590 int error;
1591
1592 /*
1593 * We must make sure the anon_vma is allocated
1594 * so that the anon_vma locking is not a noop.
1595 */
1596 if (unlikely(anon_vma_prepare(vma)))
1597 return -ENOMEM;
1598 anon_vma_lock(vma);
1599
1600 /*
1601 * vma->vm_start/vm_end cannot change under us because the caller
1602 * is required to hold the mmap_sem in read mode. We need the
1603 * anon_vma lock to serialize against concurrent expand_stacks.
1604 */
1605 address &= PAGE_MASK;
1606 error = 0;
1607
1608 /* Somebody else might have raced and expanded it already */
1609 if (address < vma->vm_start) {
1610 unsigned long size, grow;
1611
1612 size = vma->vm_end - address;
1613 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1614
1615 error = acct_stack_growth(vma, size, grow);
1616 if (!error) {
1617 vma->vm_start = address;
1618 vma->vm_pgoff -= grow;
1619 }
1620 }
1621 anon_vma_unlock(vma);
1622 return error;
1623}
1624
1625struct vm_area_struct *
1626find_extend_vma(struct mm_struct * mm, unsigned long addr)
1627{
1628 struct vm_area_struct * vma;
1629 unsigned long start;
1630
1631 addr &= PAGE_MASK;
1632 vma = find_vma(mm,addr);
1633 if (!vma)
1634 return NULL;
1635 if (vma->vm_start <= addr)
1636 return vma;
1637 if (!(vma->vm_flags & VM_GROWSDOWN))
1638 return NULL;
1639 start = vma->vm_start;
1640 if (expand_stack(vma, addr))
1641 return NULL;
1642 if (vma->vm_flags & VM_LOCKED) {
1643 make_pages_present(addr, start);
1644 }
1645 return vma;
1646}
1647#endif
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001650 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001652 *
1653 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001655static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001657 /* Update high watermark before we lower total_vm */
1658 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001660 long nrpages = vma_pages(vma);
1661
1662 mm->total_vm -= nrpages;
1663 if (vma->vm_flags & VM_LOCKED)
1664 mm->locked_vm -= nrpages;
1665 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001666 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001667 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 validate_mm(mm);
1669}
1670
1671/*
1672 * Get rid of page table information in the indicated region.
1673 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001674 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 */
1676static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001677 struct vm_area_struct *vma, struct vm_area_struct *prev,
1678 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001680 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 struct mmu_gather *tlb;
1682 unsigned long nr_accounted = 0;
1683
1684 lru_add_drain();
1685 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001686 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001687 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001689 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001690 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 tlb_finish_mmu(tlb, start, end);
1692}
1693
1694/*
1695 * Create a list of vma's touched by the unmap, removing them from the mm's
1696 * vma list as we go..
1697 */
1698static void
1699detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1700 struct vm_area_struct *prev, unsigned long end)
1701{
1702 struct vm_area_struct **insertion_point;
1703 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001704 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
1706 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1707 do {
1708 rb_erase(&vma->vm_rb, &mm->mm_rb);
1709 mm->map_count--;
1710 tail_vma = vma;
1711 vma = vma->vm_next;
1712 } while (vma && vma->vm_start < end);
1713 *insertion_point = vma;
1714 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001715 if (mm->unmap_area == arch_unmap_area)
1716 addr = prev ? prev->vm_end : mm->mmap_base;
1717 else
1718 addr = vma ? vma->vm_start : mm->mmap_base;
1719 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 mm->mmap_cache = NULL; /* Kill the cache. */
1721}
1722
1723/*
1724 * Split a vma into two pieces at address 'addr', a new vma is allocated
1725 * either for the first part or the the tail.
1726 */
1727int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1728 unsigned long addr, int new_below)
1729{
1730 struct mempolicy *pol;
1731 struct vm_area_struct *new;
1732
1733 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1734 return -EINVAL;
1735
1736 if (mm->map_count >= sysctl_max_map_count)
1737 return -ENOMEM;
1738
1739 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1740 if (!new)
1741 return -ENOMEM;
1742
1743 /* most fields are the same, copy all, and then fixup */
1744 *new = *vma;
1745
1746 if (new_below)
1747 new->vm_end = addr;
1748 else {
1749 new->vm_start = addr;
1750 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1751 }
1752
1753 pol = mpol_copy(vma_policy(vma));
1754 if (IS_ERR(pol)) {
1755 kmem_cache_free(vm_area_cachep, new);
1756 return PTR_ERR(pol);
1757 }
1758 vma_set_policy(new, pol);
1759
1760 if (new->vm_file)
1761 get_file(new->vm_file);
1762
1763 if (new->vm_ops && new->vm_ops->open)
1764 new->vm_ops->open(new);
1765
1766 if (new_below)
1767 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1768 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1769 else
1770 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1771
1772 return 0;
1773}
1774
1775/* Munmap is split into 2 main parts -- this part which finds
1776 * what needs doing, and the areas themselves, which do the
1777 * work. This now handles partial unmappings.
1778 * Jeremy Fitzhardinge <jeremy@goop.org>
1779 */
1780int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1781{
1782 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001783 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
1785 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1786 return -EINVAL;
1787
1788 if ((len = PAGE_ALIGN(len)) == 0)
1789 return -EINVAL;
1790
1791 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001792 vma = find_vma_prev(mm, start, &prev);
1793 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001795 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
1797 /* if it doesn't overlap, we have nothing.. */
1798 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001799 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 return 0;
1801
1802 /*
1803 * If we need to split any vma, do it now to save pain later.
1804 *
1805 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1806 * unmapped vm_area_struct will remain in use: so lower split_vma
1807 * places tmp vma above, and higher split_vma places tmp vma below.
1808 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001809 if (start > vma->vm_start) {
1810 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (error)
1812 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001813 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 }
1815
1816 /* Does it split the last one? */
1817 last = find_vma(mm, end);
1818 if (last && end > last->vm_start) {
1819 int error = split_vma(mm, last, end, 1);
1820 if (error)
1821 return error;
1822 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001823 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824
1825 /*
1826 * Remove the vma's, and unmap the actual pages
1827 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001828 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1829 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
1831 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001832 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
1834 return 0;
1835}
1836
1837EXPORT_SYMBOL(do_munmap);
1838
1839asmlinkage long sys_munmap(unsigned long addr, size_t len)
1840{
1841 int ret;
1842 struct mm_struct *mm = current->mm;
1843
1844 profile_munmap(addr);
1845
1846 down_write(&mm->mmap_sem);
1847 ret = do_munmap(mm, addr, len);
1848 up_write(&mm->mmap_sem);
1849 return ret;
1850}
1851
1852static inline void verify_mm_writelocked(struct mm_struct *mm)
1853{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001854#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1856 WARN_ON(1);
1857 up_read(&mm->mmap_sem);
1858 }
1859#endif
1860}
1861
1862/*
1863 * this is really a simplified "do_mmap". it only handles
1864 * anonymous maps. eventually we may be able to do some
1865 * brk-specific accounting here.
1866 */
1867unsigned long do_brk(unsigned long addr, unsigned long len)
1868{
1869 struct mm_struct * mm = current->mm;
1870 struct vm_area_struct * vma, * prev;
1871 unsigned long flags;
1872 struct rb_node ** rb_link, * rb_parent;
1873 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04001874 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
1876 len = PAGE_ALIGN(len);
1877 if (!len)
1878 return addr;
1879
1880 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1881 return -EINVAL;
1882
Kirill Korotaev3a459752006-09-07 14:17:04 +04001883 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1884
1885 error = arch_mmap_check(addr, len, flags);
1886 if (error)
1887 return error;
1888
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 /*
1890 * mlock MCL_FUTURE?
1891 */
1892 if (mm->def_flags & VM_LOCKED) {
1893 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001894 locked = len >> PAGE_SHIFT;
1895 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07001897 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1899 return -EAGAIN;
1900 }
1901
1902 /*
1903 * mm->mmap_sem is required to protect against another thread
1904 * changing the mappings in case we sleep.
1905 */
1906 verify_mm_writelocked(mm);
1907
1908 /*
1909 * Clear old maps. this also does some error checking for us
1910 */
1911 munmap_back:
1912 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1913 if (vma && vma->vm_start < addr + len) {
1914 if (do_munmap(mm, addr, len))
1915 return -ENOMEM;
1916 goto munmap_back;
1917 }
1918
1919 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001920 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 return -ENOMEM;
1922
1923 if (mm->map_count > sysctl_max_map_count)
1924 return -ENOMEM;
1925
1926 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1927 return -ENOMEM;
1928
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 /* Can we just expand an old private anonymous mapping? */
1930 if (vma_merge(mm, prev, addr, addr + len, flags,
1931 NULL, NULL, pgoff, NULL))
1932 goto out;
1933
1934 /*
1935 * create a vma struct for an anonymous mapping
1936 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001937 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 if (!vma) {
1939 vm_unacct_memory(len >> PAGE_SHIFT);
1940 return -ENOMEM;
1941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
1943 vma->vm_mm = mm;
1944 vma->vm_start = addr;
1945 vma->vm_end = addr + len;
1946 vma->vm_pgoff = pgoff;
1947 vma->vm_flags = flags;
David Howells9637a5e2006-06-23 02:03:43 -07001948 vma->vm_page_prot = protection_map[flags &
1949 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 vma_link(mm, vma, prev, rb_link, rb_parent);
1951out:
1952 mm->total_vm += len >> PAGE_SHIFT;
1953 if (flags & VM_LOCKED) {
1954 mm->locked_vm += len >> PAGE_SHIFT;
1955 make_pages_present(addr, addr + len);
1956 }
1957 return addr;
1958}
1959
1960EXPORT_SYMBOL(do_brk);
1961
1962/* Release all mmaps. */
1963void exit_mmap(struct mm_struct *mm)
1964{
1965 struct mmu_gather *tlb;
Hugh Dickinse0da3822005-04-19 13:29:15 -07001966 struct vm_area_struct *vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001968 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
1970 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001972 tlb = tlb_gather_mmu(mm, 1);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001973 /* Don't update_hiwater_rss(mm) here, do_exit already did */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001974 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07001975 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001977 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001978 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07001981 * Walk the list again, actually closing and freeing it,
1982 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001984 while (vma)
1985 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001986
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001987 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988}
1989
1990/* Insert vm structure into process list sorted by address
1991 * and into the inode's i_mmap tree. If vm_file is non-NULL
1992 * then i_mmap_lock is taken here.
1993 */
1994int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1995{
1996 struct vm_area_struct * __vma, * prev;
1997 struct rb_node ** rb_link, * rb_parent;
1998
1999 /*
2000 * The vm_pgoff of a purely anonymous vma should be irrelevant
2001 * until its first write fault, when page's anon_vma and index
2002 * are set. But now set the vm_pgoff it will almost certainly
2003 * end up with (unless mremap moves it elsewhere before that
2004 * first wfault), so /proc/pid/maps tells a consistent story.
2005 *
2006 * By setting it to reflect the virtual start address of the
2007 * vma, merges and splits can happen in a seamless way, just
2008 * using the existing file pgoff checks and manipulations.
2009 * Similarly in do_mmap_pgoff and in do_brk.
2010 */
2011 if (!vma->vm_file) {
2012 BUG_ON(vma->anon_vma);
2013 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2014 }
2015 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2016 if (__vma && __vma->vm_start < vma->vm_end)
2017 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002018 if ((vma->vm_flags & VM_ACCOUNT) &&
2019 security_vm_enough_memory(vma_pages(vma)))
2020 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 vma_link(mm, vma, prev, rb_link, rb_parent);
2022 return 0;
2023}
2024
2025/*
2026 * Copy the vma structure to a new location in the same mm,
2027 * prior to moving page table entries, to effect an mremap move.
2028 */
2029struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2030 unsigned long addr, unsigned long len, pgoff_t pgoff)
2031{
2032 struct vm_area_struct *vma = *vmap;
2033 unsigned long vma_start = vma->vm_start;
2034 struct mm_struct *mm = vma->vm_mm;
2035 struct vm_area_struct *new_vma, *prev;
2036 struct rb_node **rb_link, *rb_parent;
2037 struct mempolicy *pol;
2038
2039 /*
2040 * If anonymous vma has not yet been faulted, update new pgoff
2041 * to match new location, to increase its chance of merging.
2042 */
2043 if (!vma->vm_file && !vma->anon_vma)
2044 pgoff = addr >> PAGE_SHIFT;
2045
2046 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2047 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2048 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2049 if (new_vma) {
2050 /*
2051 * Source vma may have been merged into new_vma
2052 */
2053 if (vma_start >= new_vma->vm_start &&
2054 vma_start < new_vma->vm_end)
2055 *vmap = new_vma;
2056 } else {
2057 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2058 if (new_vma) {
2059 *new_vma = *vma;
2060 pol = mpol_copy(vma_policy(vma));
2061 if (IS_ERR(pol)) {
2062 kmem_cache_free(vm_area_cachep, new_vma);
2063 return NULL;
2064 }
2065 vma_set_policy(new_vma, pol);
2066 new_vma->vm_start = addr;
2067 new_vma->vm_end = addr + len;
2068 new_vma->vm_pgoff = pgoff;
2069 if (new_vma->vm_file)
2070 get_file(new_vma->vm_file);
2071 if (new_vma->vm_ops && new_vma->vm_ops->open)
2072 new_vma->vm_ops->open(new_vma);
2073 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2074 }
2075 }
2076 return new_vma;
2077}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002078
2079/*
2080 * Return true if the calling process may expand its vm space by the passed
2081 * number of pages
2082 */
2083int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2084{
2085 unsigned long cur = mm->total_vm; /* pages */
2086 unsigned long lim;
2087
2088 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2089
2090 if (cur + npages > lim)
2091 return 0;
2092 return 1;
2093}