blob: e6ee12344b139274325a1db834da1de6cf334da1 [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
Hugh Dickinse0da3822005-04-19 13:29:15 -070033static void unmap_region(struct mm_struct *mm,
34 struct vm_area_struct *vma, struct vm_area_struct *prev,
35 unsigned long start, unsigned long end);
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
38 * WARNING: the debugging will use recursive algorithms so never enable this
39 * unless you know what you are doing.
40 */
41#undef DEBUG_MM_RB
42
43/* description of effects of mapping type and prot in current implementation.
44 * this is due to the limited x86 page protection hardware. The expected
45 * behavior is in parens:
46 *
47 * map_type prot
48 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
49 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
50 * w: (no) no w: (no) no w: (yes) yes w: (no) no
51 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
52 *
53 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
54 * w: (no) no w: (no) no w: (copy) copy w: (no) no
55 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
56 *
57 */
58pgprot_t protection_map[16] = {
59 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
60 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
61};
62
63int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
64int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070065int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066atomic_t vm_committed_space = ATOMIC_INIT(0);
67
68/*
69 * Check that a process has enough memory to allocate a new virtual
70 * mapping. 0 means there is enough memory for the allocation to
71 * succeed and -ENOMEM implies there is not.
72 *
73 * We currently support three overcommit policies, which are set via the
74 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
75 *
76 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
77 * Additional code 2002 Jul 20 by Robert Love.
78 *
79 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
80 *
81 * Note this is a helper function intended to be used by LSMs which
82 * wish to use this logic.
83 */
84int __vm_enough_memory(long pages, int cap_sys_admin)
85{
86 unsigned long free, allowed;
87
88 vm_acct_memory(pages);
89
90 /*
91 * Sometimes we want to use more memory than we have
92 */
93 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
94 return 0;
95
96 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
97 unsigned long n;
98
99 free = get_page_cache_size();
100 free += nr_swap_pages;
101
102 /*
103 * Any slabs which are created with the
104 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
105 * which are reclaimable, under pressure. The dentry
106 * cache and most inode caches should fall into this
107 */
108 free += atomic_read(&slab_reclaim_pages);
109
110 /*
111 * Leave the last 3% for root
112 */
113 if (!cap_sys_admin)
114 free -= free / 32;
115
116 if (free > pages)
117 return 0;
118
119 /*
120 * nr_free_pages() is very expensive on large systems,
121 * only call if we're about to fail.
122 */
123 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700124
125 /*
126 * Leave reserved pages. The pages are not for anonymous pages.
127 */
128 if (n <= totalreserve_pages)
129 goto error;
130 else
131 n -= totalreserve_pages;
132
133 /*
134 * Leave the last 3% for root
135 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (!cap_sys_admin)
137 n -= n / 32;
138 free += n;
139
140 if (free > pages)
141 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700142
143 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 }
145
146 allowed = (totalram_pages - hugetlb_total_pages())
147 * sysctl_overcommit_ratio / 100;
148 /*
149 * Leave the last 3% for root
150 */
151 if (!cap_sys_admin)
152 allowed -= allowed / 32;
153 allowed += total_swap_pages;
154
155 /* Don't let a single process grow too big:
156 leave 3% of the size of this process for other processes */
157 allowed -= current->mm->total_vm / 32;
158
Simon Derr2f60f8d2005-08-04 19:52:03 -0700159 /*
160 * cast `allowed' as a signed long because vm_committed_space
161 * sometimes has a negative value
162 */
163 if (atomic_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700165error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 vm_unacct_memory(pages);
167
168 return -ENOMEM;
169}
170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171EXPORT_SYMBOL(__vm_enough_memory);
172
173/*
174 * Requires inode->i_mapping->i_mmap_lock
175 */
176static void __remove_shared_vm_struct(struct vm_area_struct *vma,
177 struct file *file, struct address_space *mapping)
178{
179 if (vma->vm_flags & VM_DENYWRITE)
180 atomic_inc(&file->f_dentry->d_inode->i_writecount);
181 if (vma->vm_flags & VM_SHARED)
182 mapping->i_mmap_writable--;
183
184 flush_dcache_mmap_lock(mapping);
185 if (unlikely(vma->vm_flags & VM_NONLINEAR))
186 list_del_init(&vma->shared.vm_set.list);
187 else
188 vma_prio_tree_remove(vma, &mapping->i_mmap);
189 flush_dcache_mmap_unlock(mapping);
190}
191
192/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700193 * Unlink a file-based vm structure from its prio_tree, to hide
194 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700196void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 struct file *file = vma->vm_file;
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (file) {
201 struct address_space *mapping = file->f_mapping;
202 spin_lock(&mapping->i_mmap_lock);
203 __remove_shared_vm_struct(vma, file, mapping);
204 spin_unlock(&mapping->i_mmap_lock);
205 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700206}
207
208/*
209 * Close a vm structure and free it, returning the next.
210 */
211static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
212{
213 struct vm_area_struct *next = vma->vm_next;
214
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700215 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (vma->vm_ops && vma->vm_ops->close)
217 vma->vm_ops->close(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700218 if (vma->vm_file)
219 fput(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 mpol_free(vma_policy(vma));
221 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700222 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225asmlinkage unsigned long sys_brk(unsigned long brk)
226{
227 unsigned long rlim, retval;
228 unsigned long newbrk, oldbrk;
229 struct mm_struct *mm = current->mm;
230
231 down_write(&mm->mmap_sem);
232
233 if (brk < mm->end_code)
234 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700235
236 /*
237 * Check against rlimit here. If this check is done later after the test
238 * of oldbrk with newbrk then it can escape the test and let the data
239 * segment grow beyond its set limit the in case where the limit is
240 * not page aligned -Ram Gupta
241 */
242 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
243 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
244 goto out;
245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 newbrk = PAGE_ALIGN(brk);
247 oldbrk = PAGE_ALIGN(mm->brk);
248 if (oldbrk == newbrk)
249 goto set_brk;
250
251 /* Always allow shrinking brk. */
252 if (brk <= mm->brk) {
253 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
254 goto set_brk;
255 goto out;
256 }
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 /* Check against existing mmap mappings. */
259 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
260 goto out;
261
262 /* Ok, looks good - let it rip. */
263 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
264 goto out;
265set_brk:
266 mm->brk = brk;
267out:
268 retval = mm->brk;
269 up_write(&mm->mmap_sem);
270 return retval;
271}
272
273#ifdef DEBUG_MM_RB
274static int browse_rb(struct rb_root *root)
275{
276 int i = 0, j;
277 struct rb_node *nd, *pn = NULL;
278 unsigned long prev = 0, pend = 0;
279
280 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
281 struct vm_area_struct *vma;
282 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
283 if (vma->vm_start < prev)
284 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
285 if (vma->vm_start < pend)
286 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
287 if (vma->vm_start > vma->vm_end)
288 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
289 i++;
290 pn = nd;
291 }
292 j = 0;
293 for (nd = pn; nd; nd = rb_prev(nd)) {
294 j++;
295 }
296 if (i != j)
297 printk("backwards %d, forwards %d\n", j, i), i = 0;
298 return i;
299}
300
301void validate_mm(struct mm_struct *mm)
302{
303 int bug = 0;
304 int i = 0;
305 struct vm_area_struct *tmp = mm->mmap;
306 while (tmp) {
307 tmp = tmp->vm_next;
308 i++;
309 }
310 if (i != mm->map_count)
311 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
312 i = browse_rb(&mm->mm_rb);
313 if (i != mm->map_count)
314 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200315 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317#else
318#define validate_mm(mm) do { } while (0)
319#endif
320
321static struct vm_area_struct *
322find_vma_prepare(struct mm_struct *mm, unsigned long addr,
323 struct vm_area_struct **pprev, struct rb_node ***rb_link,
324 struct rb_node ** rb_parent)
325{
326 struct vm_area_struct * vma;
327 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
328
329 __rb_link = &mm->mm_rb.rb_node;
330 rb_prev = __rb_parent = NULL;
331 vma = NULL;
332
333 while (*__rb_link) {
334 struct vm_area_struct *vma_tmp;
335
336 __rb_parent = *__rb_link;
337 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
338
339 if (vma_tmp->vm_end > addr) {
340 vma = vma_tmp;
341 if (vma_tmp->vm_start <= addr)
342 return vma;
343 __rb_link = &__rb_parent->rb_left;
344 } else {
345 rb_prev = __rb_parent;
346 __rb_link = &__rb_parent->rb_right;
347 }
348 }
349
350 *pprev = NULL;
351 if (rb_prev)
352 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
353 *rb_link = __rb_link;
354 *rb_parent = __rb_parent;
355 return vma;
356}
357
358static inline void
359__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
360 struct vm_area_struct *prev, struct rb_node *rb_parent)
361{
362 if (prev) {
363 vma->vm_next = prev->vm_next;
364 prev->vm_next = vma;
365 } else {
366 mm->mmap = vma;
367 if (rb_parent)
368 vma->vm_next = rb_entry(rb_parent,
369 struct vm_area_struct, vm_rb);
370 else
371 vma->vm_next = NULL;
372 }
373}
374
375void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
376 struct rb_node **rb_link, struct rb_node *rb_parent)
377{
378 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
379 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
380}
381
382static inline void __vma_link_file(struct vm_area_struct *vma)
383{
384 struct file * file;
385
386 file = vma->vm_file;
387 if (file) {
388 struct address_space *mapping = file->f_mapping;
389
390 if (vma->vm_flags & VM_DENYWRITE)
391 atomic_dec(&file->f_dentry->d_inode->i_writecount);
392 if (vma->vm_flags & VM_SHARED)
393 mapping->i_mmap_writable++;
394
395 flush_dcache_mmap_lock(mapping);
396 if (unlikely(vma->vm_flags & VM_NONLINEAR))
397 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
398 else
399 vma_prio_tree_insert(vma, &mapping->i_mmap);
400 flush_dcache_mmap_unlock(mapping);
401 }
402}
403
404static void
405__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
406 struct vm_area_struct *prev, struct rb_node **rb_link,
407 struct rb_node *rb_parent)
408{
409 __vma_link_list(mm, vma, prev, rb_parent);
410 __vma_link_rb(mm, vma, rb_link, rb_parent);
411 __anon_vma_link(vma);
412}
413
414static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
415 struct vm_area_struct *prev, struct rb_node **rb_link,
416 struct rb_node *rb_parent)
417{
418 struct address_space *mapping = NULL;
419
420 if (vma->vm_file)
421 mapping = vma->vm_file->f_mapping;
422
423 if (mapping) {
424 spin_lock(&mapping->i_mmap_lock);
425 vma->vm_truncate_count = mapping->truncate_count;
426 }
427 anon_vma_lock(vma);
428
429 __vma_link(mm, vma, prev, rb_link, rb_parent);
430 __vma_link_file(vma);
431
432 anon_vma_unlock(vma);
433 if (mapping)
434 spin_unlock(&mapping->i_mmap_lock);
435
436 mm->map_count++;
437 validate_mm(mm);
438}
439
440/*
441 * Helper for vma_adjust in the split_vma insert case:
442 * insert vm structure into list and rbtree and anon_vma,
443 * but it has already been inserted into prio_tree earlier.
444 */
445static void
446__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
447{
448 struct vm_area_struct * __vma, * prev;
449 struct rb_node ** rb_link, * rb_parent;
450
451 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200452 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 __vma_link(mm, vma, prev, rb_link, rb_parent);
454 mm->map_count++;
455}
456
457static inline void
458__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
459 struct vm_area_struct *prev)
460{
461 prev->vm_next = vma->vm_next;
462 rb_erase(&vma->vm_rb, &mm->mm_rb);
463 if (mm->mmap_cache == vma)
464 mm->mmap_cache = prev;
465}
466
467/*
468 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
469 * is already present in an i_mmap tree without adjusting the tree.
470 * The following helper function should be used when such adjustments
471 * are necessary. The "insert" vma (if any) is to be inserted
472 * before we drop the necessary locks.
473 */
474void vma_adjust(struct vm_area_struct *vma, unsigned long start,
475 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
476{
477 struct mm_struct *mm = vma->vm_mm;
478 struct vm_area_struct *next = vma->vm_next;
479 struct vm_area_struct *importer = NULL;
480 struct address_space *mapping = NULL;
481 struct prio_tree_root *root = NULL;
482 struct file *file = vma->vm_file;
483 struct anon_vma *anon_vma = NULL;
484 long adjust_next = 0;
485 int remove_next = 0;
486
487 if (next && !insert) {
488 if (end >= next->vm_end) {
489 /*
490 * vma expands, overlapping all the next, and
491 * perhaps the one after too (mprotect case 6).
492 */
493again: remove_next = 1 + (end > next->vm_end);
494 end = next->vm_end;
495 anon_vma = next->anon_vma;
496 importer = vma;
497 } else if (end > next->vm_start) {
498 /*
499 * vma expands, overlapping part of the next:
500 * mprotect case 5 shifting the boundary up.
501 */
502 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
503 anon_vma = next->anon_vma;
504 importer = vma;
505 } else if (end < vma->vm_end) {
506 /*
507 * vma shrinks, and !insert tells it's not
508 * split_vma inserting another: so it must be
509 * mprotect case 4 shifting the boundary down.
510 */
511 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
512 anon_vma = next->anon_vma;
513 importer = next;
514 }
515 }
516
517 if (file) {
518 mapping = file->f_mapping;
519 if (!(vma->vm_flags & VM_NONLINEAR))
520 root = &mapping->i_mmap;
521 spin_lock(&mapping->i_mmap_lock);
522 if (importer &&
523 vma->vm_truncate_count != next->vm_truncate_count) {
524 /*
525 * unmap_mapping_range might be in progress:
526 * ensure that the expanding vma is rescanned.
527 */
528 importer->vm_truncate_count = 0;
529 }
530 if (insert) {
531 insert->vm_truncate_count = vma->vm_truncate_count;
532 /*
533 * Put into prio_tree now, so instantiated pages
534 * are visible to arm/parisc __flush_dcache_page
535 * throughout; but we cannot insert into address
536 * space until vma start or end is updated.
537 */
538 __vma_link_file(insert);
539 }
540 }
541
542 /*
543 * When changing only vma->vm_end, we don't really need
544 * anon_vma lock: but is that case worth optimizing out?
545 */
546 if (vma->anon_vma)
547 anon_vma = vma->anon_vma;
548 if (anon_vma) {
549 spin_lock(&anon_vma->lock);
550 /*
551 * Easily overlooked: when mprotect shifts the boundary,
552 * make sure the expanding vma has anon_vma set if the
553 * shrinking vma had, to cover any anon pages imported.
554 */
555 if (importer && !importer->anon_vma) {
556 importer->anon_vma = anon_vma;
557 __anon_vma_link(importer);
558 }
559 }
560
561 if (root) {
562 flush_dcache_mmap_lock(mapping);
563 vma_prio_tree_remove(vma, root);
564 if (adjust_next)
565 vma_prio_tree_remove(next, root);
566 }
567
568 vma->vm_start = start;
569 vma->vm_end = end;
570 vma->vm_pgoff = pgoff;
571 if (adjust_next) {
572 next->vm_start += adjust_next << PAGE_SHIFT;
573 next->vm_pgoff += adjust_next;
574 }
575
576 if (root) {
577 if (adjust_next)
578 vma_prio_tree_insert(next, root);
579 vma_prio_tree_insert(vma, root);
580 flush_dcache_mmap_unlock(mapping);
581 }
582
583 if (remove_next) {
584 /*
585 * vma_merge has merged next into vma, and needs
586 * us to remove next before dropping the locks.
587 */
588 __vma_unlink(mm, next, vma);
589 if (file)
590 __remove_shared_vm_struct(next, file, mapping);
591 if (next->anon_vma)
592 __anon_vma_merge(vma, next);
593 } else if (insert) {
594 /*
595 * split_vma has split insert from vma, and needs
596 * us to insert it before dropping the locks
597 * (it may either follow vma or precede it).
598 */
599 __insert_vm_struct(mm, insert);
600 }
601
602 if (anon_vma)
603 spin_unlock(&anon_vma->lock);
604 if (mapping)
605 spin_unlock(&mapping->i_mmap_lock);
606
607 if (remove_next) {
608 if (file)
609 fput(file);
610 mm->map_count--;
611 mpol_free(vma_policy(next));
612 kmem_cache_free(vm_area_cachep, next);
613 /*
614 * In mprotect's case 6 (see comments on vma_merge),
615 * we must remove another next too. It would clutter
616 * up the code too much to do both in one go.
617 */
618 if (remove_next == 2) {
619 next = vma->vm_next;
620 goto again;
621 }
622 }
623
624 validate_mm(mm);
625}
626
627/*
628 * If the vma has a ->close operation then the driver probably needs to release
629 * per-vma resources, so we don't attempt to merge those.
630 */
Hugh Dickinsa6f563d2006-03-22 00:08:29 -0800631#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633static inline int is_mergeable_vma(struct vm_area_struct *vma,
634 struct file *file, unsigned long vm_flags)
635{
636 if (vma->vm_flags != vm_flags)
637 return 0;
638 if (vma->vm_file != file)
639 return 0;
640 if (vma->vm_ops && vma->vm_ops->close)
641 return 0;
642 return 1;
643}
644
645static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
646 struct anon_vma *anon_vma2)
647{
648 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
649}
650
651/*
652 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
653 * in front of (at a lower virtual address and file offset than) the vma.
654 *
655 * We cannot merge two vmas if they have differently assigned (non-NULL)
656 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
657 *
658 * We don't check here for the merged mmap wrapping around the end of pagecache
659 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
660 * wrap, nor mmaps which cover the final page at index -1UL.
661 */
662static int
663can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
664 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
665{
666 if (is_mergeable_vma(vma, file, vm_flags) &&
667 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
668 if (vma->vm_pgoff == vm_pgoff)
669 return 1;
670 }
671 return 0;
672}
673
674/*
675 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
676 * beyond (at a higher virtual address and file offset than) the vma.
677 *
678 * We cannot merge two vmas if they have differently assigned (non-NULL)
679 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
680 */
681static int
682can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
683 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
684{
685 if (is_mergeable_vma(vma, file, vm_flags) &&
686 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
687 pgoff_t vm_pglen;
688 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
689 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
690 return 1;
691 }
692 return 0;
693}
694
695/*
696 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
697 * whether that can be merged with its predecessor or its successor.
698 * Or both (it neatly fills a hole).
699 *
700 * In most cases - when called for mmap, brk or mremap - [addr,end) is
701 * certain not to be mapped by the time vma_merge is called; but when
702 * called for mprotect, it is certain to be already mapped (either at
703 * an offset within prev, or at the start of next), and the flags of
704 * this area are about to be changed to vm_flags - and the no-change
705 * case has already been eliminated.
706 *
707 * The following mprotect cases have to be considered, where AAAA is
708 * the area passed down from mprotect_fixup, never extending beyond one
709 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
710 *
711 * AAAA AAAA AAAA AAAA
712 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
713 * cannot merge might become might become might become
714 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
715 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
716 * mremap move: PPPPNNNNNNNN 8
717 * AAAA
718 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
719 * might become case 1 below case 2 below case 3 below
720 *
721 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
722 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
723 */
724struct vm_area_struct *vma_merge(struct mm_struct *mm,
725 struct vm_area_struct *prev, unsigned long addr,
726 unsigned long end, unsigned long vm_flags,
727 struct anon_vma *anon_vma, struct file *file,
728 pgoff_t pgoff, struct mempolicy *policy)
729{
730 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
731 struct vm_area_struct *area, *next;
732
733 /*
734 * We later require that vma->vm_flags == vm_flags,
735 * so this tests vma->vm_flags & VM_SPECIAL, too.
736 */
737 if (vm_flags & VM_SPECIAL)
738 return NULL;
739
740 if (prev)
741 next = prev->vm_next;
742 else
743 next = mm->mmap;
744 area = next;
745 if (next && next->vm_end == end) /* cases 6, 7, 8 */
746 next = next->vm_next;
747
748 /*
749 * Can it merge with the predecessor?
750 */
751 if (prev && prev->vm_end == addr &&
752 mpol_equal(vma_policy(prev), policy) &&
753 can_vma_merge_after(prev, vm_flags,
754 anon_vma, file, pgoff)) {
755 /*
756 * OK, it can. Can we now merge in the successor as well?
757 */
758 if (next && end == next->vm_start &&
759 mpol_equal(policy, vma_policy(next)) &&
760 can_vma_merge_before(next, vm_flags,
761 anon_vma, file, pgoff+pglen) &&
762 is_mergeable_anon_vma(prev->anon_vma,
763 next->anon_vma)) {
764 /* cases 1, 6 */
765 vma_adjust(prev, prev->vm_start,
766 next->vm_end, prev->vm_pgoff, NULL);
767 } else /* cases 2, 5, 7 */
768 vma_adjust(prev, prev->vm_start,
769 end, prev->vm_pgoff, NULL);
770 return prev;
771 }
772
773 /*
774 * Can this new request be merged in front of next?
775 */
776 if (next && end == next->vm_start &&
777 mpol_equal(policy, vma_policy(next)) &&
778 can_vma_merge_before(next, vm_flags,
779 anon_vma, file, pgoff+pglen)) {
780 if (prev && addr < prev->vm_end) /* case 4 */
781 vma_adjust(prev, prev->vm_start,
782 addr, prev->vm_pgoff, NULL);
783 else /* cases 3, 8 */
784 vma_adjust(area, addr, next->vm_end,
785 next->vm_pgoff - pglen, NULL);
786 return area;
787 }
788
789 return NULL;
790}
791
792/*
793 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
794 * neighbouring vmas for a suitable anon_vma, before it goes off
795 * to allocate a new anon_vma. It checks because a repetitive
796 * sequence of mprotects and faults may otherwise lead to distinct
797 * anon_vmas being allocated, preventing vma merge in subsequent
798 * mprotect.
799 */
800struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
801{
802 struct vm_area_struct *near;
803 unsigned long vm_flags;
804
805 near = vma->vm_next;
806 if (!near)
807 goto try_prev;
808
809 /*
810 * Since only mprotect tries to remerge vmas, match flags
811 * which might be mprotected into each other later on.
812 * Neither mlock nor madvise tries to remerge at present,
813 * so leave their flags as obstructing a merge.
814 */
815 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
816 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
817
818 if (near->anon_vma && vma->vm_end == near->vm_start &&
819 mpol_equal(vma_policy(vma), vma_policy(near)) &&
820 can_vma_merge_before(near, vm_flags,
821 NULL, vma->vm_file, vma->vm_pgoff +
822 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
823 return near->anon_vma;
824try_prev:
825 /*
826 * It is potentially slow to have to call find_vma_prev here.
827 * But it's only on the first write fault on the vma, not
828 * every time, and we could devise a way to avoid it later
829 * (e.g. stash info in next's anon_vma_node when assigning
830 * an anon_vma, or when trying vma_merge). Another time.
831 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200832 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (!near)
834 goto none;
835
836 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
837 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
838
839 if (near->anon_vma && near->vm_end == vma->vm_start &&
840 mpol_equal(vma_policy(near), vma_policy(vma)) &&
841 can_vma_merge_after(near, vm_flags,
842 NULL, vma->vm_file, vma->vm_pgoff))
843 return near->anon_vma;
844none:
845 /*
846 * There's no absolute need to look only at touching neighbours:
847 * we could search further afield for "compatible" anon_vmas.
848 * But it would probably just be a waste of time searching,
849 * or lead to too many vmas hanging off the same anon_vma.
850 * We're trying to allow mprotect remerging later on,
851 * not trying to minimize memory used for anon_vmas.
852 */
853 return NULL;
854}
855
856#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700857void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 struct file *file, long pages)
859{
860 const unsigned long stack_flags
861 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (file) {
864 mm->shared_vm += pages;
865 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
866 mm->exec_vm += pages;
867 } else if (flags & stack_flags)
868 mm->stack_vm += pages;
869 if (flags & (VM_RESERVED|VM_IO))
870 mm->reserved_vm += pages;
871}
872#endif /* CONFIG_PROC_FS */
873
874/*
875 * The caller must hold down_write(current->mm->mmap_sem).
876 */
877
878unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
879 unsigned long len, unsigned long prot,
880 unsigned long flags, unsigned long pgoff)
881{
882 struct mm_struct * mm = current->mm;
883 struct vm_area_struct * vma, * prev;
884 struct inode *inode;
885 unsigned int vm_flags;
886 int correct_wcount = 0;
887 int error;
888 struct rb_node ** rb_link, * rb_parent;
889 int accountable = 1;
890 unsigned long charged = 0, reqprot = prot;
891
892 if (file) {
893 if (is_file_hugepages(file))
894 accountable = 0;
895
896 if (!file->f_op || !file->f_op->mmap)
897 return -ENODEV;
898
899 if ((prot & PROT_EXEC) &&
900 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
901 return -EPERM;
902 }
903 /*
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
916 /* Careful about overflows.. */
917 len = PAGE_ALIGN(len);
918 if (!len || len > TASK_SIZE)
919 return -ENOMEM;
920
921 /* offset overflow? */
922 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
923 return -EOVERFLOW;
924
925 /* Too many mappings? */
926 if (mm->map_count > sysctl_max_map_count)
927 return -ENOMEM;
928
929 /* Obtain the address to map to. we verify (or select) it and ensure
930 * that it represents a valid section of the address space.
931 */
932 addr = get_unmapped_area(file, addr, len, pgoff, flags);
933 if (addr & ~PAGE_MASK)
934 return addr;
935
936 /* Do simple checking here so the lower-level routines won't have
937 * to. we assume access permissions have been handled by the open
938 * of the memory object, so we don't do any here.
939 */
940 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
941 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
942
943 if (flags & MAP_LOCKED) {
944 if (!can_do_mlock())
945 return -EPERM;
946 vm_flags |= VM_LOCKED;
947 }
948 /* mlock MCL_FUTURE? */
949 if (vm_flags & VM_LOCKED) {
950 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700951 locked = len >> PAGE_SHIFT;
952 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700954 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
956 return -EAGAIN;
957 }
958
959 inode = file ? file->f_dentry->d_inode : NULL;
960
961 if (file) {
962 switch (flags & MAP_TYPE) {
963 case MAP_SHARED:
964 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
965 return -EACCES;
966
967 /*
968 * Make sure we don't allow writing to an append-only
969 * file..
970 */
971 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
972 return -EACCES;
973
974 /*
975 * Make sure there are no mandatory locks on the file.
976 */
977 if (locks_verify_locked(inode))
978 return -EAGAIN;
979
980 vm_flags |= VM_SHARED | VM_MAYSHARE;
981 if (!(file->f_mode & FMODE_WRITE))
982 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
983
984 /* fall through */
985 case MAP_PRIVATE:
986 if (!(file->f_mode & FMODE_READ))
987 return -EACCES;
988 break;
989
990 default:
991 return -EINVAL;
992 }
993 } else {
994 switch (flags & MAP_TYPE) {
995 case MAP_SHARED:
996 vm_flags |= VM_SHARED | VM_MAYSHARE;
997 break;
998 case MAP_PRIVATE:
999 /*
1000 * Set pgoff according to addr for anon_vma.
1001 */
1002 pgoff = addr >> PAGE_SHIFT;
1003 break;
1004 default:
1005 return -EINVAL;
1006 }
1007 }
1008
1009 error = security_file_mmap(file, reqprot, prot, flags);
1010 if (error)
1011 return error;
1012
1013 /* Clear old maps */
1014 error = -ENOMEM;
1015munmap_back:
1016 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1017 if (vma && vma->vm_start < addr + len) {
1018 if (do_munmap(mm, addr, len))
1019 return -ENOMEM;
1020 goto munmap_back;
1021 }
1022
1023 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001024 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 return -ENOMEM;
1026
1027 if (accountable && (!(flags & MAP_NORESERVE) ||
1028 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1029 if (vm_flags & VM_SHARED) {
1030 /* Check memory availability in shmem_file_setup? */
1031 vm_flags |= VM_ACCOUNT;
1032 } else if (vm_flags & VM_WRITE) {
1033 /*
1034 * Private writable mapping: check memory availability
1035 */
1036 charged = len >> PAGE_SHIFT;
1037 if (security_vm_enough_memory(charged))
1038 return -ENOMEM;
1039 vm_flags |= VM_ACCOUNT;
1040 }
1041 }
1042
1043 /*
1044 * Can we just expand an old private anonymous mapping?
1045 * The VM_SHARED test is necessary because shmem_zero_setup
1046 * will create the file object for a shared anonymous map below.
1047 */
1048 if (!file && !(vm_flags & VM_SHARED) &&
1049 vma_merge(mm, prev, addr, addr + len, vm_flags,
1050 NULL, NULL, pgoff, NULL))
1051 goto out;
1052
1053 /*
1054 * Determine the object being mapped and call the appropriate
1055 * specific mapper. the address has already been validated, but
1056 * not unmapped, but the maps are removed from the list.
1057 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001058 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (!vma) {
1060 error = -ENOMEM;
1061 goto unacct_error;
1062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
1064 vma->vm_mm = mm;
1065 vma->vm_start = addr;
1066 vma->vm_end = addr + len;
1067 vma->vm_flags = vm_flags;
1068 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1069 vma->vm_pgoff = pgoff;
1070
1071 if (file) {
1072 error = -EINVAL;
1073 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1074 goto free_vma;
1075 if (vm_flags & VM_DENYWRITE) {
1076 error = deny_write_access(file);
1077 if (error)
1078 goto free_vma;
1079 correct_wcount = 1;
1080 }
1081 vma->vm_file = file;
1082 get_file(file);
1083 error = file->f_op->mmap(file, vma);
1084 if (error)
1085 goto unmap_and_free_vma;
1086 } else if (vm_flags & VM_SHARED) {
1087 error = shmem_zero_setup(vma);
1088 if (error)
1089 goto free_vma;
1090 }
1091
1092 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1093 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1094 * that memory reservation must be checked; but that reservation
1095 * belongs to shared memory object, not to vma: so now clear it.
1096 */
1097 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1098 vma->vm_flags &= ~VM_ACCOUNT;
1099
1100 /* Can addr have changed??
1101 *
1102 * Answer: Yes, several device drivers can do it in their
1103 * f_op->mmap method. -DaveM
1104 */
1105 addr = vma->vm_start;
1106 pgoff = vma->vm_pgoff;
1107 vm_flags = vma->vm_flags;
1108
1109 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1110 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1111 file = vma->vm_file;
1112 vma_link(mm, vma, prev, rb_link, rb_parent);
1113 if (correct_wcount)
1114 atomic_inc(&inode->i_writecount);
1115 } else {
1116 if (file) {
1117 if (correct_wcount)
1118 atomic_inc(&inode->i_writecount);
1119 fput(file);
1120 }
1121 mpol_free(vma_policy(vma));
1122 kmem_cache_free(vm_area_cachep, vma);
1123 }
1124out:
1125 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001126 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (vm_flags & VM_LOCKED) {
1128 mm->locked_vm += len >> PAGE_SHIFT;
1129 make_pages_present(addr, addr + len);
1130 }
1131 if (flags & MAP_POPULATE) {
1132 up_write(&mm->mmap_sem);
1133 sys_remap_file_pages(addr, len, 0,
1134 pgoff, flags & MAP_NONBLOCK);
1135 down_write(&mm->mmap_sem);
1136 }
1137 return addr;
1138
1139unmap_and_free_vma:
1140 if (correct_wcount)
1141 atomic_inc(&inode->i_writecount);
1142 vma->vm_file = NULL;
1143 fput(file);
1144
1145 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001146 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1147 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148free_vma:
1149 kmem_cache_free(vm_area_cachep, vma);
1150unacct_error:
1151 if (charged)
1152 vm_unacct_memory(charged);
1153 return error;
1154}
1155
1156EXPORT_SYMBOL(do_mmap_pgoff);
1157
1158/* Get an address range which is currently unmapped.
1159 * For shmat() with addr=0.
1160 *
1161 * Ugly calling convention alert:
1162 * Return value with the low bits set means error value,
1163 * ie
1164 * if (ret & ~PAGE_MASK)
1165 * error = ret;
1166 *
1167 * This function "knows" that -ENOMEM has the bits set.
1168 */
1169#ifndef HAVE_ARCH_UNMAPPED_AREA
1170unsigned long
1171arch_get_unmapped_area(struct file *filp, unsigned long addr,
1172 unsigned long len, unsigned long pgoff, unsigned long flags)
1173{
1174 struct mm_struct *mm = current->mm;
1175 struct vm_area_struct *vma;
1176 unsigned long start_addr;
1177
1178 if (len > TASK_SIZE)
1179 return -ENOMEM;
1180
1181 if (addr) {
1182 addr = PAGE_ALIGN(addr);
1183 vma = find_vma(mm, addr);
1184 if (TASK_SIZE - len >= addr &&
1185 (!vma || addr + len <= vma->vm_start))
1186 return addr;
1187 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001188 if (len > mm->cached_hole_size) {
1189 start_addr = addr = mm->free_area_cache;
1190 } else {
1191 start_addr = addr = TASK_UNMAPPED_BASE;
1192 mm->cached_hole_size = 0;
1193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195full_search:
1196 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1197 /* At this point: (!vma || addr < vma->vm_end). */
1198 if (TASK_SIZE - len < addr) {
1199 /*
1200 * Start a new search - just in case we missed
1201 * some holes.
1202 */
1203 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001204 addr = TASK_UNMAPPED_BASE;
1205 start_addr = addr;
1206 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 goto full_search;
1208 }
1209 return -ENOMEM;
1210 }
1211 if (!vma || addr + len <= vma->vm_start) {
1212 /*
1213 * Remember the place where we stopped the search:
1214 */
1215 mm->free_area_cache = addr + len;
1216 return addr;
1217 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001218 if (addr + mm->cached_hole_size < vma->vm_start)
1219 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 addr = vma->vm_end;
1221 }
1222}
1223#endif
1224
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001225void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226{
1227 /*
1228 * Is this a new hole at the lowest possible address?
1229 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001230 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1231 mm->free_area_cache = addr;
1232 mm->cached_hole_size = ~0UL;
1233 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234}
1235
1236/*
1237 * This mmap-allocator allocates new areas top-down from below the
1238 * stack's low limit (the base):
1239 */
1240#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1241unsigned long
1242arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1243 const unsigned long len, const unsigned long pgoff,
1244 const unsigned long flags)
1245{
1246 struct vm_area_struct *vma;
1247 struct mm_struct *mm = current->mm;
1248 unsigned long addr = addr0;
1249
1250 /* requested length too big for entire address space */
1251 if (len > TASK_SIZE)
1252 return -ENOMEM;
1253
1254 /* requesting a specific address */
1255 if (addr) {
1256 addr = PAGE_ALIGN(addr);
1257 vma = find_vma(mm, addr);
1258 if (TASK_SIZE - len >= addr &&
1259 (!vma || addr + len <= vma->vm_start))
1260 return addr;
1261 }
1262
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001263 /* check if free_area_cache is useful for us */
1264 if (len <= mm->cached_hole_size) {
1265 mm->cached_hole_size = 0;
1266 mm->free_area_cache = mm->mmap_base;
1267 }
1268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 /* either no address requested or can't fit in requested address hole */
1270 addr = mm->free_area_cache;
1271
1272 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001273 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 vma = find_vma(mm, addr-len);
1275 if (!vma || addr <= vma->vm_start)
1276 /* remember the address as a hint for next time */
1277 return (mm->free_area_cache = addr-len);
1278 }
1279
Chris Wright73219d12005-06-21 17:14:52 -07001280 if (mm->mmap_base < len)
1281 goto bottomup;
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 addr = mm->mmap_base-len;
1284
1285 do {
1286 /*
1287 * Lookup failure means no vma is above this address,
1288 * else if new region fits below vma->vm_start,
1289 * return with success:
1290 */
1291 vma = find_vma(mm, addr);
1292 if (!vma || addr+len <= vma->vm_start)
1293 /* remember the address as a hint for next time */
1294 return (mm->free_area_cache = addr);
1295
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001296 /* remember the largest hole we saw so far */
1297 if (addr + mm->cached_hole_size < vma->vm_start)
1298 mm->cached_hole_size = vma->vm_start - addr;
1299
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 /* try just below the current vma->vm_start */
1301 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001302 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Chris Wright73219d12005-06-21 17:14:52 -07001304bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 /*
1306 * A failed mmap() very likely causes application failure,
1307 * so fall back to the bottom-up function here. This scenario
1308 * can happen with large stack limits and large mmap()
1309 * allocations.
1310 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001311 mm->cached_hole_size = ~0UL;
1312 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1314 /*
1315 * Restore the topdown base:
1316 */
1317 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001318 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
1320 return addr;
1321}
1322#endif
1323
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001324void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
1326 /*
1327 * Is this a new hole at the highest possible address?
1328 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001329 if (addr > mm->free_area_cache)
1330 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
1332 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001333 if (mm->free_area_cache > mm->mmap_base)
1334 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335}
1336
1337unsigned long
1338get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1339 unsigned long pgoff, unsigned long flags)
1340{
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001341 unsigned long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001343 if (!(flags & MAP_FIXED)) {
1344 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1345
1346 get_area = current->mm->get_unmapped_area;
1347 if (file && file->f_op && file->f_op->get_unmapped_area)
1348 get_area = file->f_op->get_unmapped_area;
1349 addr = get_area(file, addr, len, pgoff, flags);
1350 if (IS_ERR_VALUE(addr))
1351 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 }
1353
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001354 if (addr > TASK_SIZE - len)
1355 return -ENOMEM;
1356 if (addr & ~PAGE_MASK)
1357 return -EINVAL;
1358 if (file && is_file_hugepages(file)) {
1359 /*
1360 * Check if the given range is hugepage aligned, and
1361 * can be made suitable for hugepages.
1362 */
1363 ret = prepare_hugepage_range(addr, len);
1364 } else {
1365 /*
1366 * Ensure that a normal request is not falling in a
1367 * reserved hugepage range. For some archs like IA-64,
1368 * there is a separate region for hugepages.
1369 */
1370 ret = is_hugepage_only_range(current->mm, addr, len);
1371 }
1372 if (ret)
1373 return -EINVAL;
1374 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
1377EXPORT_SYMBOL(get_unmapped_area);
1378
1379/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1380struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1381{
1382 struct vm_area_struct *vma = NULL;
1383
1384 if (mm) {
1385 /* Check the cache first. */
1386 /* (Cache hit rate is typically around 35%.) */
1387 vma = mm->mmap_cache;
1388 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1389 struct rb_node * rb_node;
1390
1391 rb_node = mm->mm_rb.rb_node;
1392 vma = NULL;
1393
1394 while (rb_node) {
1395 struct vm_area_struct * vma_tmp;
1396
1397 vma_tmp = rb_entry(rb_node,
1398 struct vm_area_struct, vm_rb);
1399
1400 if (vma_tmp->vm_end > addr) {
1401 vma = vma_tmp;
1402 if (vma_tmp->vm_start <= addr)
1403 break;
1404 rb_node = rb_node->rb_left;
1405 } else
1406 rb_node = rb_node->rb_right;
1407 }
1408 if (vma)
1409 mm->mmap_cache = vma;
1410 }
1411 }
1412 return vma;
1413}
1414
1415EXPORT_SYMBOL(find_vma);
1416
1417/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1418struct vm_area_struct *
1419find_vma_prev(struct mm_struct *mm, unsigned long addr,
1420 struct vm_area_struct **pprev)
1421{
1422 struct vm_area_struct *vma = NULL, *prev = NULL;
1423 struct rb_node * rb_node;
1424 if (!mm)
1425 goto out;
1426
1427 /* Guard against addr being lower than the first VMA */
1428 vma = mm->mmap;
1429
1430 /* Go through the RB tree quickly. */
1431 rb_node = mm->mm_rb.rb_node;
1432
1433 while (rb_node) {
1434 struct vm_area_struct *vma_tmp;
1435 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1436
1437 if (addr < vma_tmp->vm_end) {
1438 rb_node = rb_node->rb_left;
1439 } else {
1440 prev = vma_tmp;
1441 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1442 break;
1443 rb_node = rb_node->rb_right;
1444 }
1445 }
1446
1447out:
1448 *pprev = prev;
1449 return prev ? prev->vm_next : vma;
1450}
1451
1452/*
1453 * Verify that the stack growth is acceptable and
1454 * update accounting. This is shared with both the
1455 * grow-up and grow-down cases.
1456 */
1457static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1458{
1459 struct mm_struct *mm = vma->vm_mm;
1460 struct rlimit *rlim = current->signal->rlim;
1461
1462 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001463 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return -ENOMEM;
1465
1466 /* Stack limit test */
1467 if (size > rlim[RLIMIT_STACK].rlim_cur)
1468 return -ENOMEM;
1469
1470 /* mlock limit tests */
1471 if (vma->vm_flags & VM_LOCKED) {
1472 unsigned long locked;
1473 unsigned long limit;
1474 locked = mm->locked_vm + grow;
1475 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1476 if (locked > limit && !capable(CAP_IPC_LOCK))
1477 return -ENOMEM;
1478 }
1479
1480 /*
1481 * Overcommit.. This must be the final test, as it will
1482 * update security statistics.
1483 */
1484 if (security_vm_enough_memory(grow))
1485 return -ENOMEM;
1486
1487 /* Ok, everything looks good - let it rip */
1488 mm->total_vm += grow;
1489 if (vma->vm_flags & VM_LOCKED)
1490 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001491 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 return 0;
1493}
1494
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001495#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001497 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1498 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001500#ifndef CONFIG_IA64
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001501static inline
1502#endif
1503int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504{
1505 int error;
1506
1507 if (!(vma->vm_flags & VM_GROWSUP))
1508 return -EFAULT;
1509
1510 /*
1511 * We must make sure the anon_vma is allocated
1512 * so that the anon_vma locking is not a noop.
1513 */
1514 if (unlikely(anon_vma_prepare(vma)))
1515 return -ENOMEM;
1516 anon_vma_lock(vma);
1517
1518 /*
1519 * vma->vm_start/vm_end cannot change under us because the caller
1520 * is required to hold the mmap_sem in read mode. We need the
1521 * anon_vma lock to serialize against concurrent expand_stacks.
1522 */
1523 address += 4 + PAGE_SIZE - 1;
1524 address &= PAGE_MASK;
1525 error = 0;
1526
1527 /* Somebody else might have raced and expanded it already */
1528 if (address > vma->vm_end) {
1529 unsigned long size, grow;
1530
1531 size = address - vma->vm_start;
1532 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1533
1534 error = acct_stack_growth(vma, size, grow);
1535 if (!error)
1536 vma->vm_end = address;
1537 }
1538 anon_vma_unlock(vma);
1539 return error;
1540}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001541#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1542
1543#ifdef CONFIG_STACK_GROWSUP
1544int expand_stack(struct vm_area_struct *vma, unsigned long address)
1545{
1546 return expand_upwards(vma, address);
1547}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549struct vm_area_struct *
1550find_extend_vma(struct mm_struct *mm, unsigned long addr)
1551{
1552 struct vm_area_struct *vma, *prev;
1553
1554 addr &= PAGE_MASK;
1555 vma = find_vma_prev(mm, addr, &prev);
1556 if (vma && (vma->vm_start <= addr))
1557 return vma;
1558 if (!prev || expand_stack(prev, addr))
1559 return NULL;
1560 if (prev->vm_flags & VM_LOCKED) {
1561 make_pages_present(addr, prev->vm_end);
1562 }
1563 return prev;
1564}
1565#else
1566/*
1567 * vma is the first one with address < vma->vm_start. Have to extend vma.
1568 */
1569int expand_stack(struct vm_area_struct *vma, unsigned long address)
1570{
1571 int error;
1572
1573 /*
1574 * We must make sure the anon_vma is allocated
1575 * so that the anon_vma locking is not a noop.
1576 */
1577 if (unlikely(anon_vma_prepare(vma)))
1578 return -ENOMEM;
1579 anon_vma_lock(vma);
1580
1581 /*
1582 * vma->vm_start/vm_end cannot change under us because the caller
1583 * is required to hold the mmap_sem in read mode. We need the
1584 * anon_vma lock to serialize against concurrent expand_stacks.
1585 */
1586 address &= PAGE_MASK;
1587 error = 0;
1588
1589 /* Somebody else might have raced and expanded it already */
1590 if (address < vma->vm_start) {
1591 unsigned long size, grow;
1592
1593 size = vma->vm_end - address;
1594 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1595
1596 error = acct_stack_growth(vma, size, grow);
1597 if (!error) {
1598 vma->vm_start = address;
1599 vma->vm_pgoff -= grow;
1600 }
1601 }
1602 anon_vma_unlock(vma);
1603 return error;
1604}
1605
1606struct vm_area_struct *
1607find_extend_vma(struct mm_struct * mm, unsigned long addr)
1608{
1609 struct vm_area_struct * vma;
1610 unsigned long start;
1611
1612 addr &= PAGE_MASK;
1613 vma = find_vma(mm,addr);
1614 if (!vma)
1615 return NULL;
1616 if (vma->vm_start <= addr)
1617 return vma;
1618 if (!(vma->vm_flags & VM_GROWSDOWN))
1619 return NULL;
1620 start = vma->vm_start;
1621 if (expand_stack(vma, addr))
1622 return NULL;
1623 if (vma->vm_flags & VM_LOCKED) {
1624 make_pages_present(addr, start);
1625 }
1626 return vma;
1627}
1628#endif
1629
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001631 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001633 *
1634 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001636static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001638 /* Update high watermark before we lower total_vm */
1639 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001641 long nrpages = vma_pages(vma);
1642
1643 mm->total_vm -= nrpages;
1644 if (vma->vm_flags & VM_LOCKED)
1645 mm->locked_vm -= nrpages;
1646 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001647 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001648 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 validate_mm(mm);
1650}
1651
1652/*
1653 * Get rid of page table information in the indicated region.
1654 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001655 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 */
1657static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001658 struct vm_area_struct *vma, struct vm_area_struct *prev,
1659 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001661 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 struct mmu_gather *tlb;
1663 unsigned long nr_accounted = 0;
1664
1665 lru_add_drain();
1666 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001667 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001668 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001670 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001671 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 tlb_finish_mmu(tlb, start, end);
1673}
1674
1675/*
1676 * Create a list of vma's touched by the unmap, removing them from the mm's
1677 * vma list as we go..
1678 */
1679static void
1680detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1681 struct vm_area_struct *prev, unsigned long end)
1682{
1683 struct vm_area_struct **insertion_point;
1684 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001685 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
1687 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1688 do {
1689 rb_erase(&vma->vm_rb, &mm->mm_rb);
1690 mm->map_count--;
1691 tail_vma = vma;
1692 vma = vma->vm_next;
1693 } while (vma && vma->vm_start < end);
1694 *insertion_point = vma;
1695 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001696 if (mm->unmap_area == arch_unmap_area)
1697 addr = prev ? prev->vm_end : mm->mmap_base;
1698 else
1699 addr = vma ? vma->vm_start : mm->mmap_base;
1700 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 mm->mmap_cache = NULL; /* Kill the cache. */
1702}
1703
1704/*
1705 * Split a vma into two pieces at address 'addr', a new vma is allocated
1706 * either for the first part or the the tail.
1707 */
1708int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1709 unsigned long addr, int new_below)
1710{
1711 struct mempolicy *pol;
1712 struct vm_area_struct *new;
1713
1714 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1715 return -EINVAL;
1716
1717 if (mm->map_count >= sysctl_max_map_count)
1718 return -ENOMEM;
1719
1720 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1721 if (!new)
1722 return -ENOMEM;
1723
1724 /* most fields are the same, copy all, and then fixup */
1725 *new = *vma;
1726
1727 if (new_below)
1728 new->vm_end = addr;
1729 else {
1730 new->vm_start = addr;
1731 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1732 }
1733
1734 pol = mpol_copy(vma_policy(vma));
1735 if (IS_ERR(pol)) {
1736 kmem_cache_free(vm_area_cachep, new);
1737 return PTR_ERR(pol);
1738 }
1739 vma_set_policy(new, pol);
1740
1741 if (new->vm_file)
1742 get_file(new->vm_file);
1743
1744 if (new->vm_ops && new->vm_ops->open)
1745 new->vm_ops->open(new);
1746
1747 if (new_below)
1748 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1749 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1750 else
1751 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1752
1753 return 0;
1754}
1755
1756/* Munmap is split into 2 main parts -- this part which finds
1757 * what needs doing, and the areas themselves, which do the
1758 * work. This now handles partial unmappings.
1759 * Jeremy Fitzhardinge <jeremy@goop.org>
1760 */
1761int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1762{
1763 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001764 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
1766 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1767 return -EINVAL;
1768
1769 if ((len = PAGE_ALIGN(len)) == 0)
1770 return -EINVAL;
1771
1772 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001773 vma = find_vma_prev(mm, start, &prev);
1774 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001776 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
1778 /* if it doesn't overlap, we have nothing.. */
1779 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001780 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 return 0;
1782
1783 /*
1784 * If we need to split any vma, do it now to save pain later.
1785 *
1786 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1787 * unmapped vm_area_struct will remain in use: so lower split_vma
1788 * places tmp vma above, and higher split_vma places tmp vma below.
1789 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001790 if (start > vma->vm_start) {
1791 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 if (error)
1793 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001794 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 }
1796
1797 /* Does it split the last one? */
1798 last = find_vma(mm, end);
1799 if (last && end > last->vm_start) {
1800 int error = split_vma(mm, last, end, 1);
1801 if (error)
1802 return error;
1803 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001804 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
1806 /*
1807 * Remove the vma's, and unmap the actual pages
1808 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001809 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1810 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
1812 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001813 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
1815 return 0;
1816}
1817
1818EXPORT_SYMBOL(do_munmap);
1819
1820asmlinkage long sys_munmap(unsigned long addr, size_t len)
1821{
1822 int ret;
1823 struct mm_struct *mm = current->mm;
1824
1825 profile_munmap(addr);
1826
1827 down_write(&mm->mmap_sem);
1828 ret = do_munmap(mm, addr, len);
1829 up_write(&mm->mmap_sem);
1830 return ret;
1831}
1832
1833static inline void verify_mm_writelocked(struct mm_struct *mm)
1834{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001835#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1837 WARN_ON(1);
1838 up_read(&mm->mmap_sem);
1839 }
1840#endif
1841}
1842
1843/*
1844 * this is really a simplified "do_mmap". it only handles
1845 * anonymous maps. eventually we may be able to do some
1846 * brk-specific accounting here.
1847 */
1848unsigned long do_brk(unsigned long addr, unsigned long len)
1849{
1850 struct mm_struct * mm = current->mm;
1851 struct vm_area_struct * vma, * prev;
1852 unsigned long flags;
1853 struct rb_node ** rb_link, * rb_parent;
1854 pgoff_t pgoff = addr >> PAGE_SHIFT;
1855
1856 len = PAGE_ALIGN(len);
1857 if (!len)
1858 return addr;
1859
1860 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1861 return -EINVAL;
1862
1863 /*
1864 * mlock MCL_FUTURE?
1865 */
1866 if (mm->def_flags & VM_LOCKED) {
1867 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001868 locked = len >> PAGE_SHIFT;
1869 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07001871 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1873 return -EAGAIN;
1874 }
1875
1876 /*
1877 * mm->mmap_sem is required to protect against another thread
1878 * changing the mappings in case we sleep.
1879 */
1880 verify_mm_writelocked(mm);
1881
1882 /*
1883 * Clear old maps. this also does some error checking for us
1884 */
1885 munmap_back:
1886 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1887 if (vma && vma->vm_start < addr + len) {
1888 if (do_munmap(mm, addr, len))
1889 return -ENOMEM;
1890 goto munmap_back;
1891 }
1892
1893 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001894 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 return -ENOMEM;
1896
1897 if (mm->map_count > sysctl_max_map_count)
1898 return -ENOMEM;
1899
1900 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1901 return -ENOMEM;
1902
1903 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1904
1905 /* Can we just expand an old private anonymous mapping? */
1906 if (vma_merge(mm, prev, addr, addr + len, flags,
1907 NULL, NULL, pgoff, NULL))
1908 goto out;
1909
1910 /*
1911 * create a vma struct for an anonymous mapping
1912 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001913 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 if (!vma) {
1915 vm_unacct_memory(len >> PAGE_SHIFT);
1916 return -ENOMEM;
1917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
1919 vma->vm_mm = mm;
1920 vma->vm_start = addr;
1921 vma->vm_end = addr + len;
1922 vma->vm_pgoff = pgoff;
1923 vma->vm_flags = flags;
1924 vma->vm_page_prot = protection_map[flags & 0x0f];
1925 vma_link(mm, vma, prev, rb_link, rb_parent);
1926out:
1927 mm->total_vm += len >> PAGE_SHIFT;
1928 if (flags & VM_LOCKED) {
1929 mm->locked_vm += len >> PAGE_SHIFT;
1930 make_pages_present(addr, addr + len);
1931 }
1932 return addr;
1933}
1934
1935EXPORT_SYMBOL(do_brk);
1936
1937/* Release all mmaps. */
1938void exit_mmap(struct mm_struct *mm)
1939{
1940 struct mmu_gather *tlb;
Hugh Dickinse0da3822005-04-19 13:29:15 -07001941 struct vm_area_struct *vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001943 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
1945 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001947 tlb = tlb_gather_mmu(mm, 1);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001948 /* Don't update_hiwater_rss(mm) here, do_exit already did */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001949 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07001950 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001952 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001953 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07001956 * Walk the list again, actually closing and freeing it,
1957 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001959 while (vma)
1960 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001961
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001962 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963}
1964
1965/* Insert vm structure into process list sorted by address
1966 * and into the inode's i_mmap tree. If vm_file is non-NULL
1967 * then i_mmap_lock is taken here.
1968 */
1969int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1970{
1971 struct vm_area_struct * __vma, * prev;
1972 struct rb_node ** rb_link, * rb_parent;
1973
1974 /*
1975 * The vm_pgoff of a purely anonymous vma should be irrelevant
1976 * until its first write fault, when page's anon_vma and index
1977 * are set. But now set the vm_pgoff it will almost certainly
1978 * end up with (unless mremap moves it elsewhere before that
1979 * first wfault), so /proc/pid/maps tells a consistent story.
1980 *
1981 * By setting it to reflect the virtual start address of the
1982 * vma, merges and splits can happen in a seamless way, just
1983 * using the existing file pgoff checks and manipulations.
1984 * Similarly in do_mmap_pgoff and in do_brk.
1985 */
1986 if (!vma->vm_file) {
1987 BUG_ON(vma->anon_vma);
1988 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1989 }
1990 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1991 if (__vma && __vma->vm_start < vma->vm_end)
1992 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01001993 if ((vma->vm_flags & VM_ACCOUNT) &&
1994 security_vm_enough_memory(vma_pages(vma)))
1995 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 vma_link(mm, vma, prev, rb_link, rb_parent);
1997 return 0;
1998}
1999
2000/*
2001 * Copy the vma structure to a new location in the same mm,
2002 * prior to moving page table entries, to effect an mremap move.
2003 */
2004struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2005 unsigned long addr, unsigned long len, pgoff_t pgoff)
2006{
2007 struct vm_area_struct *vma = *vmap;
2008 unsigned long vma_start = vma->vm_start;
2009 struct mm_struct *mm = vma->vm_mm;
2010 struct vm_area_struct *new_vma, *prev;
2011 struct rb_node **rb_link, *rb_parent;
2012 struct mempolicy *pol;
2013
2014 /*
2015 * If anonymous vma has not yet been faulted, update new pgoff
2016 * to match new location, to increase its chance of merging.
2017 */
2018 if (!vma->vm_file && !vma->anon_vma)
2019 pgoff = addr >> PAGE_SHIFT;
2020
2021 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2022 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2023 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2024 if (new_vma) {
2025 /*
2026 * Source vma may have been merged into new_vma
2027 */
2028 if (vma_start >= new_vma->vm_start &&
2029 vma_start < new_vma->vm_end)
2030 *vmap = new_vma;
2031 } else {
2032 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2033 if (new_vma) {
2034 *new_vma = *vma;
2035 pol = mpol_copy(vma_policy(vma));
2036 if (IS_ERR(pol)) {
2037 kmem_cache_free(vm_area_cachep, new_vma);
2038 return NULL;
2039 }
2040 vma_set_policy(new_vma, pol);
2041 new_vma->vm_start = addr;
2042 new_vma->vm_end = addr + len;
2043 new_vma->vm_pgoff = pgoff;
2044 if (new_vma->vm_file)
2045 get_file(new_vma->vm_file);
2046 if (new_vma->vm_ops && new_vma->vm_ops->open)
2047 new_vma->vm_ops->open(new_vma);
2048 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2049 }
2050 }
2051 return new_vma;
2052}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002053
2054/*
2055 * Return true if the calling process may expand its vm space by the passed
2056 * number of pages
2057 */
2058int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2059{
2060 unsigned long cur = mm->total_vm; /* pages */
2061 unsigned long lim;
2062
2063 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2064
2065 if (cur + npages > lim)
2066 return 0;
2067 return 1;
2068}