blob: 404319477e71d61679166bc954ed246719affe89 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
6 * Address space accounting code <alan@redhat.com>
7 */
8
9#include <linux/slab.h>
10#include <linux/mm.h>
11#include <linux/shm.h>
12#include <linux/mman.h>
13#include <linux/pagemap.h>
14#include <linux/swap.h>
15#include <linux/syscalls.h>
16#include <linux/init.h>
17#include <linux/file.h>
18#include <linux/fs.h>
19#include <linux/personality.h>
20#include <linux/security.h>
21#include <linux/hugetlb.h>
22#include <linux/profile.h>
23#include <linux/module.h>
24#include <linux/mount.h>
25#include <linux/mempolicy.h>
26#include <linux/rmap.h>
27
28#include <asm/uaccess.h>
29#include <asm/cacheflush.h>
30#include <asm/tlb.h>
31
Hugh Dickinse0da3822005-04-19 13:29:15 -070032static void unmap_region(struct mm_struct *mm,
33 struct vm_area_struct *vma, struct vm_area_struct *prev,
34 unsigned long start, unsigned long end);
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/*
37 * WARNING: the debugging will use recursive algorithms so never enable this
38 * unless you know what you are doing.
39 */
40#undef DEBUG_MM_RB
41
42/* description of effects of mapping type and prot in current implementation.
43 * this is due to the limited x86 page protection hardware. The expected
44 * behavior is in parens:
45 *
46 * map_type prot
47 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
48 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
49 * w: (no) no w: (no) no w: (yes) yes w: (no) no
50 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
51 *
52 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
53 * w: (no) no w: (no) no w: (copy) copy w: (no) no
54 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
55 *
56 */
57pgprot_t protection_map[16] = {
58 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
59 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
60};
61
62int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63int sysctl_overcommit_ratio = 50; /* default is 50% */
64int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
65atomic_t vm_committed_space = ATOMIC_INIT(0);
66
67/*
68 * Check that a process has enough memory to allocate a new virtual
69 * mapping. 0 means there is enough memory for the allocation to
70 * succeed and -ENOMEM implies there is not.
71 *
72 * We currently support three overcommit policies, which are set via the
73 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
74 *
75 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
76 * Additional code 2002 Jul 20 by Robert Love.
77 *
78 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
79 *
80 * Note this is a helper function intended to be used by LSMs which
81 * wish to use this logic.
82 */
83int __vm_enough_memory(long pages, int cap_sys_admin)
84{
85 unsigned long free, allowed;
86
87 vm_acct_memory(pages);
88
89 /*
90 * Sometimes we want to use more memory than we have
91 */
92 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
93 return 0;
94
95 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
96 unsigned long n;
97
98 free = get_page_cache_size();
99 free += nr_swap_pages;
100
101 /*
102 * Any slabs which are created with the
103 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
104 * which are reclaimable, under pressure. The dentry
105 * cache and most inode caches should fall into this
106 */
107 free += atomic_read(&slab_reclaim_pages);
108
109 /*
110 * Leave the last 3% for root
111 */
112 if (!cap_sys_admin)
113 free -= free / 32;
114
115 if (free > pages)
116 return 0;
117
118 /*
119 * nr_free_pages() is very expensive on large systems,
120 * only call if we're about to fail.
121 */
122 n = nr_free_pages();
123 if (!cap_sys_admin)
124 n -= n / 32;
125 free += n;
126
127 if (free > pages)
128 return 0;
129 vm_unacct_memory(pages);
130 return -ENOMEM;
131 }
132
133 allowed = (totalram_pages - hugetlb_total_pages())
134 * sysctl_overcommit_ratio / 100;
135 /*
136 * Leave the last 3% for root
137 */
138 if (!cap_sys_admin)
139 allowed -= allowed / 32;
140 allowed += total_swap_pages;
141
142 /* Don't let a single process grow too big:
143 leave 3% of the size of this process for other processes */
144 allowed -= current->mm->total_vm / 32;
145
Simon Derr2f60f8d2005-08-04 19:52:03 -0700146 /*
147 * cast `allowed' as a signed long because vm_committed_space
148 * sometimes has a negative value
149 */
150 if (atomic_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 return 0;
152
153 vm_unacct_memory(pages);
154
155 return -ENOMEM;
156}
157
158EXPORT_SYMBOL(sysctl_overcommit_memory);
159EXPORT_SYMBOL(sysctl_overcommit_ratio);
160EXPORT_SYMBOL(sysctl_max_map_count);
161EXPORT_SYMBOL(vm_committed_space);
162EXPORT_SYMBOL(__vm_enough_memory);
163
164/*
165 * Requires inode->i_mapping->i_mmap_lock
166 */
167static void __remove_shared_vm_struct(struct vm_area_struct *vma,
168 struct file *file, struct address_space *mapping)
169{
170 if (vma->vm_flags & VM_DENYWRITE)
171 atomic_inc(&file->f_dentry->d_inode->i_writecount);
172 if (vma->vm_flags & VM_SHARED)
173 mapping->i_mmap_writable--;
174
175 flush_dcache_mmap_lock(mapping);
176 if (unlikely(vma->vm_flags & VM_NONLINEAR))
177 list_del_init(&vma->shared.vm_set.list);
178 else
179 vma_prio_tree_remove(vma, &mapping->i_mmap);
180 flush_dcache_mmap_unlock(mapping);
181}
182
183/*
184 * Remove one vm structure and free it.
185 */
186static void remove_vm_struct(struct vm_area_struct *vma)
187{
188 struct file *file = vma->vm_file;
189
190 might_sleep();
191 if (file) {
192 struct address_space *mapping = file->f_mapping;
193 spin_lock(&mapping->i_mmap_lock);
194 __remove_shared_vm_struct(vma, file, mapping);
195 spin_unlock(&mapping->i_mmap_lock);
196 }
197 if (vma->vm_ops && vma->vm_ops->close)
198 vma->vm_ops->close(vma);
199 if (file)
200 fput(file);
201 anon_vma_unlink(vma);
202 mpol_free(vma_policy(vma));
203 kmem_cache_free(vm_area_cachep, vma);
204}
205
206/*
207 * sys_brk() for the most part doesn't need the global kernel
208 * lock, except when an application is doing something nasty
209 * like trying to un-brk an area that has already been mapped
210 * to a regular file. in this case, the unmapping will need
211 * to invoke file system routines that need the global lock.
212 */
213asmlinkage unsigned long sys_brk(unsigned long brk)
214{
215 unsigned long rlim, retval;
216 unsigned long newbrk, oldbrk;
217 struct mm_struct *mm = current->mm;
218
219 down_write(&mm->mmap_sem);
220
221 if (brk < mm->end_code)
222 goto out;
223 newbrk = PAGE_ALIGN(brk);
224 oldbrk = PAGE_ALIGN(mm->brk);
225 if (oldbrk == newbrk)
226 goto set_brk;
227
228 /* Always allow shrinking brk. */
229 if (brk <= mm->brk) {
230 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
231 goto set_brk;
232 goto out;
233 }
234
235 /* Check against rlimit.. */
236 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
237 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
238 goto out;
239
240 /* Check against existing mmap mappings. */
241 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
242 goto out;
243
244 /* Ok, looks good - let it rip. */
245 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
246 goto out;
247set_brk:
248 mm->brk = brk;
249out:
250 retval = mm->brk;
251 up_write(&mm->mmap_sem);
252 return retval;
253}
254
255#ifdef DEBUG_MM_RB
256static int browse_rb(struct rb_root *root)
257{
258 int i = 0, j;
259 struct rb_node *nd, *pn = NULL;
260 unsigned long prev = 0, pend = 0;
261
262 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
263 struct vm_area_struct *vma;
264 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
265 if (vma->vm_start < prev)
266 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
267 if (vma->vm_start < pend)
268 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
269 if (vma->vm_start > vma->vm_end)
270 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
271 i++;
272 pn = nd;
273 }
274 j = 0;
275 for (nd = pn; nd; nd = rb_prev(nd)) {
276 j++;
277 }
278 if (i != j)
279 printk("backwards %d, forwards %d\n", j, i), i = 0;
280 return i;
281}
282
283void validate_mm(struct mm_struct *mm)
284{
285 int bug = 0;
286 int i = 0;
287 struct vm_area_struct *tmp = mm->mmap;
288 while (tmp) {
289 tmp = tmp->vm_next;
290 i++;
291 }
292 if (i != mm->map_count)
293 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
294 i = browse_rb(&mm->mm_rb);
295 if (i != mm->map_count)
296 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
297 if (bug)
298 BUG();
299}
300#else
301#define validate_mm(mm) do { } while (0)
302#endif
303
304static struct vm_area_struct *
305find_vma_prepare(struct mm_struct *mm, unsigned long addr,
306 struct vm_area_struct **pprev, struct rb_node ***rb_link,
307 struct rb_node ** rb_parent)
308{
309 struct vm_area_struct * vma;
310 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
311
312 __rb_link = &mm->mm_rb.rb_node;
313 rb_prev = __rb_parent = NULL;
314 vma = NULL;
315
316 while (*__rb_link) {
317 struct vm_area_struct *vma_tmp;
318
319 __rb_parent = *__rb_link;
320 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
321
322 if (vma_tmp->vm_end > addr) {
323 vma = vma_tmp;
324 if (vma_tmp->vm_start <= addr)
325 return vma;
326 __rb_link = &__rb_parent->rb_left;
327 } else {
328 rb_prev = __rb_parent;
329 __rb_link = &__rb_parent->rb_right;
330 }
331 }
332
333 *pprev = NULL;
334 if (rb_prev)
335 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
336 *rb_link = __rb_link;
337 *rb_parent = __rb_parent;
338 return vma;
339}
340
341static inline void
342__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
343 struct vm_area_struct *prev, struct rb_node *rb_parent)
344{
345 if (prev) {
346 vma->vm_next = prev->vm_next;
347 prev->vm_next = vma;
348 } else {
349 mm->mmap = vma;
350 if (rb_parent)
351 vma->vm_next = rb_entry(rb_parent,
352 struct vm_area_struct, vm_rb);
353 else
354 vma->vm_next = NULL;
355 }
356}
357
358void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
359 struct rb_node **rb_link, struct rb_node *rb_parent)
360{
361 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
362 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
363}
364
365static inline void __vma_link_file(struct vm_area_struct *vma)
366{
367 struct file * file;
368
369 file = vma->vm_file;
370 if (file) {
371 struct address_space *mapping = file->f_mapping;
372
373 if (vma->vm_flags & VM_DENYWRITE)
374 atomic_dec(&file->f_dentry->d_inode->i_writecount);
375 if (vma->vm_flags & VM_SHARED)
376 mapping->i_mmap_writable++;
377
378 flush_dcache_mmap_lock(mapping);
379 if (unlikely(vma->vm_flags & VM_NONLINEAR))
380 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
381 else
382 vma_prio_tree_insert(vma, &mapping->i_mmap);
383 flush_dcache_mmap_unlock(mapping);
384 }
385}
386
387static void
388__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
389 struct vm_area_struct *prev, struct rb_node **rb_link,
390 struct rb_node *rb_parent)
391{
392 __vma_link_list(mm, vma, prev, rb_parent);
393 __vma_link_rb(mm, vma, rb_link, rb_parent);
394 __anon_vma_link(vma);
395}
396
397static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
398 struct vm_area_struct *prev, struct rb_node **rb_link,
399 struct rb_node *rb_parent)
400{
401 struct address_space *mapping = NULL;
402
403 if (vma->vm_file)
404 mapping = vma->vm_file->f_mapping;
405
406 if (mapping) {
407 spin_lock(&mapping->i_mmap_lock);
408 vma->vm_truncate_count = mapping->truncate_count;
409 }
410 anon_vma_lock(vma);
411
412 __vma_link(mm, vma, prev, rb_link, rb_parent);
413 __vma_link_file(vma);
414
415 anon_vma_unlock(vma);
416 if (mapping)
417 spin_unlock(&mapping->i_mmap_lock);
418
419 mm->map_count++;
420 validate_mm(mm);
421}
422
423/*
424 * Helper for vma_adjust in the split_vma insert case:
425 * insert vm structure into list and rbtree and anon_vma,
426 * but it has already been inserted into prio_tree earlier.
427 */
428static void
429__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
430{
431 struct vm_area_struct * __vma, * prev;
432 struct rb_node ** rb_link, * rb_parent;
433
434 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
435 if (__vma && __vma->vm_start < vma->vm_end)
436 BUG();
437 __vma_link(mm, vma, prev, rb_link, rb_parent);
438 mm->map_count++;
439}
440
441static inline void
442__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
443 struct vm_area_struct *prev)
444{
445 prev->vm_next = vma->vm_next;
446 rb_erase(&vma->vm_rb, &mm->mm_rb);
447 if (mm->mmap_cache == vma)
448 mm->mmap_cache = prev;
449}
450
451/*
452 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
453 * is already present in an i_mmap tree without adjusting the tree.
454 * The following helper function should be used when such adjustments
455 * are necessary. The "insert" vma (if any) is to be inserted
456 * before we drop the necessary locks.
457 */
458void vma_adjust(struct vm_area_struct *vma, unsigned long start,
459 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
460{
461 struct mm_struct *mm = vma->vm_mm;
462 struct vm_area_struct *next = vma->vm_next;
463 struct vm_area_struct *importer = NULL;
464 struct address_space *mapping = NULL;
465 struct prio_tree_root *root = NULL;
466 struct file *file = vma->vm_file;
467 struct anon_vma *anon_vma = NULL;
468 long adjust_next = 0;
469 int remove_next = 0;
470
471 if (next && !insert) {
472 if (end >= next->vm_end) {
473 /*
474 * vma expands, overlapping all the next, and
475 * perhaps the one after too (mprotect case 6).
476 */
477again: remove_next = 1 + (end > next->vm_end);
478 end = next->vm_end;
479 anon_vma = next->anon_vma;
480 importer = vma;
481 } else if (end > next->vm_start) {
482 /*
483 * vma expands, overlapping part of the next:
484 * mprotect case 5 shifting the boundary up.
485 */
486 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
487 anon_vma = next->anon_vma;
488 importer = vma;
489 } else if (end < vma->vm_end) {
490 /*
491 * vma shrinks, and !insert tells it's not
492 * split_vma inserting another: so it must be
493 * mprotect case 4 shifting the boundary down.
494 */
495 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
496 anon_vma = next->anon_vma;
497 importer = next;
498 }
499 }
500
501 if (file) {
502 mapping = file->f_mapping;
503 if (!(vma->vm_flags & VM_NONLINEAR))
504 root = &mapping->i_mmap;
505 spin_lock(&mapping->i_mmap_lock);
506 if (importer &&
507 vma->vm_truncate_count != next->vm_truncate_count) {
508 /*
509 * unmap_mapping_range might be in progress:
510 * ensure that the expanding vma is rescanned.
511 */
512 importer->vm_truncate_count = 0;
513 }
514 if (insert) {
515 insert->vm_truncate_count = vma->vm_truncate_count;
516 /*
517 * Put into prio_tree now, so instantiated pages
518 * are visible to arm/parisc __flush_dcache_page
519 * throughout; but we cannot insert into address
520 * space until vma start or end is updated.
521 */
522 __vma_link_file(insert);
523 }
524 }
525
526 /*
527 * When changing only vma->vm_end, we don't really need
528 * anon_vma lock: but is that case worth optimizing out?
529 */
530 if (vma->anon_vma)
531 anon_vma = vma->anon_vma;
532 if (anon_vma) {
533 spin_lock(&anon_vma->lock);
534 /*
535 * Easily overlooked: when mprotect shifts the boundary,
536 * make sure the expanding vma has anon_vma set if the
537 * shrinking vma had, to cover any anon pages imported.
538 */
539 if (importer && !importer->anon_vma) {
540 importer->anon_vma = anon_vma;
541 __anon_vma_link(importer);
542 }
543 }
544
545 if (root) {
546 flush_dcache_mmap_lock(mapping);
547 vma_prio_tree_remove(vma, root);
548 if (adjust_next)
549 vma_prio_tree_remove(next, root);
550 }
551
552 vma->vm_start = start;
553 vma->vm_end = end;
554 vma->vm_pgoff = pgoff;
555 if (adjust_next) {
556 next->vm_start += adjust_next << PAGE_SHIFT;
557 next->vm_pgoff += adjust_next;
558 }
559
560 if (root) {
561 if (adjust_next)
562 vma_prio_tree_insert(next, root);
563 vma_prio_tree_insert(vma, root);
564 flush_dcache_mmap_unlock(mapping);
565 }
566
567 if (remove_next) {
568 /*
569 * vma_merge has merged next into vma, and needs
570 * us to remove next before dropping the locks.
571 */
572 __vma_unlink(mm, next, vma);
573 if (file)
574 __remove_shared_vm_struct(next, file, mapping);
575 if (next->anon_vma)
576 __anon_vma_merge(vma, next);
577 } else if (insert) {
578 /*
579 * split_vma has split insert from vma, and needs
580 * us to insert it before dropping the locks
581 * (it may either follow vma or precede it).
582 */
583 __insert_vm_struct(mm, insert);
584 }
585
586 if (anon_vma)
587 spin_unlock(&anon_vma->lock);
588 if (mapping)
589 spin_unlock(&mapping->i_mmap_lock);
590
591 if (remove_next) {
592 if (file)
593 fput(file);
594 mm->map_count--;
595 mpol_free(vma_policy(next));
596 kmem_cache_free(vm_area_cachep, next);
597 /*
598 * In mprotect's case 6 (see comments on vma_merge),
599 * we must remove another next too. It would clutter
600 * up the code too much to do both in one go.
601 */
602 if (remove_next == 2) {
603 next = vma->vm_next;
604 goto again;
605 }
606 }
607
608 validate_mm(mm);
609}
610
611/*
612 * If the vma has a ->close operation then the driver probably needs to release
613 * per-vma resources, so we don't attempt to merge those.
614 */
615#define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
616
617static inline int is_mergeable_vma(struct vm_area_struct *vma,
618 struct file *file, unsigned long vm_flags)
619{
620 if (vma->vm_flags != vm_flags)
621 return 0;
622 if (vma->vm_file != file)
623 return 0;
624 if (vma->vm_ops && vma->vm_ops->close)
625 return 0;
626 return 1;
627}
628
629static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
630 struct anon_vma *anon_vma2)
631{
632 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
633}
634
635/*
636 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
637 * in front of (at a lower virtual address and file offset than) the vma.
638 *
639 * We cannot merge two vmas if they have differently assigned (non-NULL)
640 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
641 *
642 * We don't check here for the merged mmap wrapping around the end of pagecache
643 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
644 * wrap, nor mmaps which cover the final page at index -1UL.
645 */
646static int
647can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
648 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
649{
650 if (is_mergeable_vma(vma, file, vm_flags) &&
651 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
652 if (vma->vm_pgoff == vm_pgoff)
653 return 1;
654 }
655 return 0;
656}
657
658/*
659 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
660 * beyond (at a higher virtual address and file offset than) the vma.
661 *
662 * We cannot merge two vmas if they have differently assigned (non-NULL)
663 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
664 */
665static int
666can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
667 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
668{
669 if (is_mergeable_vma(vma, file, vm_flags) &&
670 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
671 pgoff_t vm_pglen;
672 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
673 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
674 return 1;
675 }
676 return 0;
677}
678
679/*
680 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
681 * whether that can be merged with its predecessor or its successor.
682 * Or both (it neatly fills a hole).
683 *
684 * In most cases - when called for mmap, brk or mremap - [addr,end) is
685 * certain not to be mapped by the time vma_merge is called; but when
686 * called for mprotect, it is certain to be already mapped (either at
687 * an offset within prev, or at the start of next), and the flags of
688 * this area are about to be changed to vm_flags - and the no-change
689 * case has already been eliminated.
690 *
691 * The following mprotect cases have to be considered, where AAAA is
692 * the area passed down from mprotect_fixup, never extending beyond one
693 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
694 *
695 * AAAA AAAA AAAA AAAA
696 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
697 * cannot merge might become might become might become
698 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
699 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
700 * mremap move: PPPPNNNNNNNN 8
701 * AAAA
702 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
703 * might become case 1 below case 2 below case 3 below
704 *
705 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
706 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
707 */
708struct vm_area_struct *vma_merge(struct mm_struct *mm,
709 struct vm_area_struct *prev, unsigned long addr,
710 unsigned long end, unsigned long vm_flags,
711 struct anon_vma *anon_vma, struct file *file,
712 pgoff_t pgoff, struct mempolicy *policy)
713{
714 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
715 struct vm_area_struct *area, *next;
716
717 /*
718 * We later require that vma->vm_flags == vm_flags,
719 * so this tests vma->vm_flags & VM_SPECIAL, too.
720 */
721 if (vm_flags & VM_SPECIAL)
722 return NULL;
723
724 if (prev)
725 next = prev->vm_next;
726 else
727 next = mm->mmap;
728 area = next;
729 if (next && next->vm_end == end) /* cases 6, 7, 8 */
730 next = next->vm_next;
731
732 /*
733 * Can it merge with the predecessor?
734 */
735 if (prev && prev->vm_end == addr &&
736 mpol_equal(vma_policy(prev), policy) &&
737 can_vma_merge_after(prev, vm_flags,
738 anon_vma, file, pgoff)) {
739 /*
740 * OK, it can. Can we now merge in the successor as well?
741 */
742 if (next && end == next->vm_start &&
743 mpol_equal(policy, vma_policy(next)) &&
744 can_vma_merge_before(next, vm_flags,
745 anon_vma, file, pgoff+pglen) &&
746 is_mergeable_anon_vma(prev->anon_vma,
747 next->anon_vma)) {
748 /* cases 1, 6 */
749 vma_adjust(prev, prev->vm_start,
750 next->vm_end, prev->vm_pgoff, NULL);
751 } else /* cases 2, 5, 7 */
752 vma_adjust(prev, prev->vm_start,
753 end, prev->vm_pgoff, NULL);
754 return prev;
755 }
756
757 /*
758 * Can this new request be merged in front of next?
759 */
760 if (next && end == next->vm_start &&
761 mpol_equal(policy, vma_policy(next)) &&
762 can_vma_merge_before(next, vm_flags,
763 anon_vma, file, pgoff+pglen)) {
764 if (prev && addr < prev->vm_end) /* case 4 */
765 vma_adjust(prev, prev->vm_start,
766 addr, prev->vm_pgoff, NULL);
767 else /* cases 3, 8 */
768 vma_adjust(area, addr, next->vm_end,
769 next->vm_pgoff - pglen, NULL);
770 return area;
771 }
772
773 return NULL;
774}
775
776/*
777 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
778 * neighbouring vmas for a suitable anon_vma, before it goes off
779 * to allocate a new anon_vma. It checks because a repetitive
780 * sequence of mprotects and faults may otherwise lead to distinct
781 * anon_vmas being allocated, preventing vma merge in subsequent
782 * mprotect.
783 */
784struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
785{
786 struct vm_area_struct *near;
787 unsigned long vm_flags;
788
789 near = vma->vm_next;
790 if (!near)
791 goto try_prev;
792
793 /*
794 * Since only mprotect tries to remerge vmas, match flags
795 * which might be mprotected into each other later on.
796 * Neither mlock nor madvise tries to remerge at present,
797 * so leave their flags as obstructing a merge.
798 */
799 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
800 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
801
802 if (near->anon_vma && vma->vm_end == near->vm_start &&
803 mpol_equal(vma_policy(vma), vma_policy(near)) &&
804 can_vma_merge_before(near, vm_flags,
805 NULL, vma->vm_file, vma->vm_pgoff +
806 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
807 return near->anon_vma;
808try_prev:
809 /*
810 * It is potentially slow to have to call find_vma_prev here.
811 * But it's only on the first write fault on the vma, not
812 * every time, and we could devise a way to avoid it later
813 * (e.g. stash info in next's anon_vma_node when assigning
814 * an anon_vma, or when trying vma_merge). Another time.
815 */
816 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
817 BUG();
818 if (!near)
819 goto none;
820
821 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
822 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
823
824 if (near->anon_vma && near->vm_end == vma->vm_start &&
825 mpol_equal(vma_policy(near), vma_policy(vma)) &&
826 can_vma_merge_after(near, vm_flags,
827 NULL, vma->vm_file, vma->vm_pgoff))
828 return near->anon_vma;
829none:
830 /*
831 * There's no absolute need to look only at touching neighbours:
832 * we could search further afield for "compatible" anon_vmas.
833 * But it would probably just be a waste of time searching,
834 * or lead to too many vmas hanging off the same anon_vma.
835 * We're trying to allow mprotect remerging later on,
836 * not trying to minimize memory used for anon_vmas.
837 */
838 return NULL;
839}
840
841#ifdef CONFIG_PROC_FS
842void __vm_stat_account(struct mm_struct *mm, unsigned long flags,
843 struct file *file, long pages)
844{
845 const unsigned long stack_flags
846 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
847
848#ifdef CONFIG_HUGETLB
849 if (flags & VM_HUGETLB) {
850 if (!(flags & VM_DONTCOPY))
851 mm->shared_vm += pages;
852 return;
853 }
854#endif /* CONFIG_HUGETLB */
855
856 if (file) {
857 mm->shared_vm += pages;
858 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
859 mm->exec_vm += pages;
860 } else if (flags & stack_flags)
861 mm->stack_vm += pages;
862 if (flags & (VM_RESERVED|VM_IO))
863 mm->reserved_vm += pages;
864}
865#endif /* CONFIG_PROC_FS */
866
867/*
868 * The caller must hold down_write(current->mm->mmap_sem).
869 */
870
871unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
872 unsigned long len, unsigned long prot,
873 unsigned long flags, unsigned long pgoff)
874{
875 struct mm_struct * mm = current->mm;
876 struct vm_area_struct * vma, * prev;
877 struct inode *inode;
878 unsigned int vm_flags;
879 int correct_wcount = 0;
880 int error;
881 struct rb_node ** rb_link, * rb_parent;
882 int accountable = 1;
883 unsigned long charged = 0, reqprot = prot;
884
885 if (file) {
886 if (is_file_hugepages(file))
887 accountable = 0;
888
889 if (!file->f_op || !file->f_op->mmap)
890 return -ENODEV;
891
892 if ((prot & PROT_EXEC) &&
893 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
894 return -EPERM;
895 }
896 /*
897 * Does the application expect PROT_READ to imply PROT_EXEC?
898 *
899 * (the exception is when the underlying filesystem is noexec
900 * mounted, in which case we dont add PROT_EXEC.)
901 */
902 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
903 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
904 prot |= PROT_EXEC;
905
906 if (!len)
907 return -EINVAL;
908
909 /* Careful about overflows.. */
910 len = PAGE_ALIGN(len);
911 if (!len || len > TASK_SIZE)
912 return -ENOMEM;
913
914 /* offset overflow? */
915 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
916 return -EOVERFLOW;
917
918 /* Too many mappings? */
919 if (mm->map_count > sysctl_max_map_count)
920 return -ENOMEM;
921
922 /* Obtain the address to map to. we verify (or select) it and ensure
923 * that it represents a valid section of the address space.
924 */
925 addr = get_unmapped_area(file, addr, len, pgoff, flags);
926 if (addr & ~PAGE_MASK)
927 return addr;
928
929 /* Do simple checking here so the lower-level routines won't have
930 * to. we assume access permissions have been handled by the open
931 * of the memory object, so we don't do any here.
932 */
933 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
934 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
935
936 if (flags & MAP_LOCKED) {
937 if (!can_do_mlock())
938 return -EPERM;
939 vm_flags |= VM_LOCKED;
940 }
941 /* mlock MCL_FUTURE? */
942 if (vm_flags & VM_LOCKED) {
943 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700944 locked = len >> PAGE_SHIFT;
945 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700947 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
949 return -EAGAIN;
950 }
951
952 inode = file ? file->f_dentry->d_inode : NULL;
953
954 if (file) {
955 switch (flags & MAP_TYPE) {
956 case MAP_SHARED:
957 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
958 return -EACCES;
959
960 /*
961 * Make sure we don't allow writing to an append-only
962 * file..
963 */
964 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
965 return -EACCES;
966
967 /*
968 * Make sure there are no mandatory locks on the file.
969 */
970 if (locks_verify_locked(inode))
971 return -EAGAIN;
972
973 vm_flags |= VM_SHARED | VM_MAYSHARE;
974 if (!(file->f_mode & FMODE_WRITE))
975 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
976
977 /* fall through */
978 case MAP_PRIVATE:
979 if (!(file->f_mode & FMODE_READ))
980 return -EACCES;
981 break;
982
983 default:
984 return -EINVAL;
985 }
986 } else {
987 switch (flags & MAP_TYPE) {
988 case MAP_SHARED:
989 vm_flags |= VM_SHARED | VM_MAYSHARE;
990 break;
991 case MAP_PRIVATE:
992 /*
993 * Set pgoff according to addr for anon_vma.
994 */
995 pgoff = addr >> PAGE_SHIFT;
996 break;
997 default:
998 return -EINVAL;
999 }
1000 }
1001
1002 error = security_file_mmap(file, reqprot, prot, flags);
1003 if (error)
1004 return error;
1005
1006 /* Clear old maps */
1007 error = -ENOMEM;
1008munmap_back:
1009 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1010 if (vma && vma->vm_start < addr + len) {
1011 if (do_munmap(mm, addr, len))
1012 return -ENOMEM;
1013 goto munmap_back;
1014 }
1015
1016 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001017 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 return -ENOMEM;
1019
1020 if (accountable && (!(flags & MAP_NORESERVE) ||
1021 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1022 if (vm_flags & VM_SHARED) {
1023 /* Check memory availability in shmem_file_setup? */
1024 vm_flags |= VM_ACCOUNT;
1025 } else if (vm_flags & VM_WRITE) {
1026 /*
1027 * Private writable mapping: check memory availability
1028 */
1029 charged = len >> PAGE_SHIFT;
1030 if (security_vm_enough_memory(charged))
1031 return -ENOMEM;
1032 vm_flags |= VM_ACCOUNT;
1033 }
1034 }
1035
1036 /*
1037 * Can we just expand an old private anonymous mapping?
1038 * The VM_SHARED test is necessary because shmem_zero_setup
1039 * will create the file object for a shared anonymous map below.
1040 */
1041 if (!file && !(vm_flags & VM_SHARED) &&
1042 vma_merge(mm, prev, addr, addr + len, vm_flags,
1043 NULL, NULL, pgoff, NULL))
1044 goto out;
1045
1046 /*
1047 * Determine the object being mapped and call the appropriate
1048 * specific mapper. the address has already been validated, but
1049 * not unmapped, but the maps are removed from the list.
1050 */
1051 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1052 if (!vma) {
1053 error = -ENOMEM;
1054 goto unacct_error;
1055 }
1056 memset(vma, 0, sizeof(*vma));
1057
1058 vma->vm_mm = mm;
1059 vma->vm_start = addr;
1060 vma->vm_end = addr + len;
1061 vma->vm_flags = vm_flags;
1062 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1063 vma->vm_pgoff = pgoff;
1064
1065 if (file) {
1066 error = -EINVAL;
1067 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1068 goto free_vma;
1069 if (vm_flags & VM_DENYWRITE) {
1070 error = deny_write_access(file);
1071 if (error)
1072 goto free_vma;
1073 correct_wcount = 1;
1074 }
1075 vma->vm_file = file;
1076 get_file(file);
1077 error = file->f_op->mmap(file, vma);
1078 if (error)
1079 goto unmap_and_free_vma;
1080 } else if (vm_flags & VM_SHARED) {
1081 error = shmem_zero_setup(vma);
1082 if (error)
1083 goto free_vma;
1084 }
1085
1086 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1087 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1088 * that memory reservation must be checked; but that reservation
1089 * belongs to shared memory object, not to vma: so now clear it.
1090 */
1091 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1092 vma->vm_flags &= ~VM_ACCOUNT;
1093
1094 /* Can addr have changed??
1095 *
1096 * Answer: Yes, several device drivers can do it in their
1097 * f_op->mmap method. -DaveM
1098 */
1099 addr = vma->vm_start;
1100 pgoff = vma->vm_pgoff;
1101 vm_flags = vma->vm_flags;
1102
1103 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1104 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1105 file = vma->vm_file;
1106 vma_link(mm, vma, prev, rb_link, rb_parent);
1107 if (correct_wcount)
1108 atomic_inc(&inode->i_writecount);
1109 } else {
1110 if (file) {
1111 if (correct_wcount)
1112 atomic_inc(&inode->i_writecount);
1113 fput(file);
1114 }
1115 mpol_free(vma_policy(vma));
1116 kmem_cache_free(vm_area_cachep, vma);
1117 }
1118out:
1119 mm->total_vm += len >> PAGE_SHIFT;
1120 __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1121 if (vm_flags & VM_LOCKED) {
1122 mm->locked_vm += len >> PAGE_SHIFT;
1123 make_pages_present(addr, addr + len);
1124 }
1125 if (flags & MAP_POPULATE) {
1126 up_write(&mm->mmap_sem);
1127 sys_remap_file_pages(addr, len, 0,
1128 pgoff, flags & MAP_NONBLOCK);
1129 down_write(&mm->mmap_sem);
1130 }
1131 return addr;
1132
1133unmap_and_free_vma:
1134 if (correct_wcount)
1135 atomic_inc(&inode->i_writecount);
1136 vma->vm_file = NULL;
1137 fput(file);
1138
1139 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001140 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1141 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142free_vma:
1143 kmem_cache_free(vm_area_cachep, vma);
1144unacct_error:
1145 if (charged)
1146 vm_unacct_memory(charged);
1147 return error;
1148}
1149
1150EXPORT_SYMBOL(do_mmap_pgoff);
1151
1152/* Get an address range which is currently unmapped.
1153 * For shmat() with addr=0.
1154 *
1155 * Ugly calling convention alert:
1156 * Return value with the low bits set means error value,
1157 * ie
1158 * if (ret & ~PAGE_MASK)
1159 * error = ret;
1160 *
1161 * This function "knows" that -ENOMEM has the bits set.
1162 */
1163#ifndef HAVE_ARCH_UNMAPPED_AREA
1164unsigned long
1165arch_get_unmapped_area(struct file *filp, unsigned long addr,
1166 unsigned long len, unsigned long pgoff, unsigned long flags)
1167{
1168 struct mm_struct *mm = current->mm;
1169 struct vm_area_struct *vma;
1170 unsigned long start_addr;
1171
1172 if (len > TASK_SIZE)
1173 return -ENOMEM;
1174
1175 if (addr) {
1176 addr = PAGE_ALIGN(addr);
1177 vma = find_vma(mm, addr);
1178 if (TASK_SIZE - len >= addr &&
1179 (!vma || addr + len <= vma->vm_start))
1180 return addr;
1181 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001182 if (len > mm->cached_hole_size) {
1183 start_addr = addr = mm->free_area_cache;
1184 } else {
1185 start_addr = addr = TASK_UNMAPPED_BASE;
1186 mm->cached_hole_size = 0;
1187 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
1189full_search:
1190 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1191 /* At this point: (!vma || addr < vma->vm_end). */
1192 if (TASK_SIZE - len < addr) {
1193 /*
1194 * Start a new search - just in case we missed
1195 * some holes.
1196 */
1197 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001198 addr = TASK_UNMAPPED_BASE;
1199 start_addr = addr;
1200 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 goto full_search;
1202 }
1203 return -ENOMEM;
1204 }
1205 if (!vma || addr + len <= vma->vm_start) {
1206 /*
1207 * Remember the place where we stopped the search:
1208 */
1209 mm->free_area_cache = addr + len;
1210 return addr;
1211 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001212 if (addr + mm->cached_hole_size < vma->vm_start)
1213 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 addr = vma->vm_end;
1215 }
1216}
1217#endif
1218
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001219void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
1221 /*
1222 * Is this a new hole at the lowest possible address?
1223 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001224 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1225 mm->free_area_cache = addr;
1226 mm->cached_hole_size = ~0UL;
1227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228}
1229
1230/*
1231 * This mmap-allocator allocates new areas top-down from below the
1232 * stack's low limit (the base):
1233 */
1234#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1235unsigned long
1236arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1237 const unsigned long len, const unsigned long pgoff,
1238 const unsigned long flags)
1239{
1240 struct vm_area_struct *vma;
1241 struct mm_struct *mm = current->mm;
1242 unsigned long addr = addr0;
1243
1244 /* requested length too big for entire address space */
1245 if (len > TASK_SIZE)
1246 return -ENOMEM;
1247
1248 /* requesting a specific address */
1249 if (addr) {
1250 addr = PAGE_ALIGN(addr);
1251 vma = find_vma(mm, addr);
1252 if (TASK_SIZE - len >= addr &&
1253 (!vma || addr + len <= vma->vm_start))
1254 return addr;
1255 }
1256
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001257 /* check if free_area_cache is useful for us */
1258 if (len <= mm->cached_hole_size) {
1259 mm->cached_hole_size = 0;
1260 mm->free_area_cache = mm->mmap_base;
1261 }
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 /* either no address requested or can't fit in requested address hole */
1264 addr = mm->free_area_cache;
1265
1266 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001267 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 vma = find_vma(mm, addr-len);
1269 if (!vma || addr <= vma->vm_start)
1270 /* remember the address as a hint for next time */
1271 return (mm->free_area_cache = addr-len);
1272 }
1273
Chris Wright73219d12005-06-21 17:14:52 -07001274 if (mm->mmap_base < len)
1275 goto bottomup;
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 addr = mm->mmap_base-len;
1278
1279 do {
1280 /*
1281 * Lookup failure means no vma is above this address,
1282 * else if new region fits below vma->vm_start,
1283 * return with success:
1284 */
1285 vma = find_vma(mm, addr);
1286 if (!vma || addr+len <= vma->vm_start)
1287 /* remember the address as a hint for next time */
1288 return (mm->free_area_cache = addr);
1289
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001290 /* remember the largest hole we saw so far */
1291 if (addr + mm->cached_hole_size < vma->vm_start)
1292 mm->cached_hole_size = vma->vm_start - addr;
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 /* try just below the current vma->vm_start */
1295 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001296 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Chris Wright73219d12005-06-21 17:14:52 -07001298bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 /*
1300 * A failed mmap() very likely causes application failure,
1301 * so fall back to the bottom-up function here. This scenario
1302 * can happen with large stack limits and large mmap()
1303 * allocations.
1304 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001305 mm->cached_hole_size = ~0UL;
1306 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1308 /*
1309 * Restore the topdown base:
1310 */
1311 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001312 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
1314 return addr;
1315}
1316#endif
1317
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001318void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319{
1320 /*
1321 * Is this a new hole at the highest possible address?
1322 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001323 if (addr > mm->free_area_cache)
1324 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001327 if (mm->free_area_cache > mm->mmap_base)
1328 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330
1331unsigned long
1332get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1333 unsigned long pgoff, unsigned long flags)
1334{
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001335 unsigned long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001337 if (!(flags & MAP_FIXED)) {
1338 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1339
1340 get_area = current->mm->get_unmapped_area;
1341 if (file && file->f_op && file->f_op->get_unmapped_area)
1342 get_area = file->f_op->get_unmapped_area;
1343 addr = get_area(file, addr, len, pgoff, flags);
1344 if (IS_ERR_VALUE(addr))
1345 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 }
1347
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001348 if (addr > TASK_SIZE - len)
1349 return -ENOMEM;
1350 if (addr & ~PAGE_MASK)
1351 return -EINVAL;
1352 if (file && is_file_hugepages(file)) {
1353 /*
1354 * Check if the given range is hugepage aligned, and
1355 * can be made suitable for hugepages.
1356 */
1357 ret = prepare_hugepage_range(addr, len);
1358 } else {
1359 /*
1360 * Ensure that a normal request is not falling in a
1361 * reserved hugepage range. For some archs like IA-64,
1362 * there is a separate region for hugepages.
1363 */
1364 ret = is_hugepage_only_range(current->mm, addr, len);
1365 }
1366 if (ret)
1367 return -EINVAL;
1368 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369}
1370
1371EXPORT_SYMBOL(get_unmapped_area);
1372
1373/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1374struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1375{
1376 struct vm_area_struct *vma = NULL;
1377
1378 if (mm) {
1379 /* Check the cache first. */
1380 /* (Cache hit rate is typically around 35%.) */
1381 vma = mm->mmap_cache;
1382 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1383 struct rb_node * rb_node;
1384
1385 rb_node = mm->mm_rb.rb_node;
1386 vma = NULL;
1387
1388 while (rb_node) {
1389 struct vm_area_struct * vma_tmp;
1390
1391 vma_tmp = rb_entry(rb_node,
1392 struct vm_area_struct, vm_rb);
1393
1394 if (vma_tmp->vm_end > addr) {
1395 vma = vma_tmp;
1396 if (vma_tmp->vm_start <= addr)
1397 break;
1398 rb_node = rb_node->rb_left;
1399 } else
1400 rb_node = rb_node->rb_right;
1401 }
1402 if (vma)
1403 mm->mmap_cache = vma;
1404 }
1405 }
1406 return vma;
1407}
1408
1409EXPORT_SYMBOL(find_vma);
1410
1411/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1412struct vm_area_struct *
1413find_vma_prev(struct mm_struct *mm, unsigned long addr,
1414 struct vm_area_struct **pprev)
1415{
1416 struct vm_area_struct *vma = NULL, *prev = NULL;
1417 struct rb_node * rb_node;
1418 if (!mm)
1419 goto out;
1420
1421 /* Guard against addr being lower than the first VMA */
1422 vma = mm->mmap;
1423
1424 /* Go through the RB tree quickly. */
1425 rb_node = mm->mm_rb.rb_node;
1426
1427 while (rb_node) {
1428 struct vm_area_struct *vma_tmp;
1429 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1430
1431 if (addr < vma_tmp->vm_end) {
1432 rb_node = rb_node->rb_left;
1433 } else {
1434 prev = vma_tmp;
1435 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1436 break;
1437 rb_node = rb_node->rb_right;
1438 }
1439 }
1440
1441out:
1442 *pprev = prev;
1443 return prev ? prev->vm_next : vma;
1444}
1445
1446/*
1447 * Verify that the stack growth is acceptable and
1448 * update accounting. This is shared with both the
1449 * grow-up and grow-down cases.
1450 */
1451static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1452{
1453 struct mm_struct *mm = vma->vm_mm;
1454 struct rlimit *rlim = current->signal->rlim;
1455
1456 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001457 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 return -ENOMEM;
1459
1460 /* Stack limit test */
1461 if (size > rlim[RLIMIT_STACK].rlim_cur)
1462 return -ENOMEM;
1463
1464 /* mlock limit tests */
1465 if (vma->vm_flags & VM_LOCKED) {
1466 unsigned long locked;
1467 unsigned long limit;
1468 locked = mm->locked_vm + grow;
1469 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1470 if (locked > limit && !capable(CAP_IPC_LOCK))
1471 return -ENOMEM;
1472 }
1473
1474 /*
1475 * Overcommit.. This must be the final test, as it will
1476 * update security statistics.
1477 */
1478 if (security_vm_enough_memory(grow))
1479 return -ENOMEM;
1480
1481 /* Ok, everything looks good - let it rip */
1482 mm->total_vm += grow;
1483 if (vma->vm_flags & VM_LOCKED)
1484 mm->locked_vm += grow;
1485 __vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1486 return 0;
1487}
1488
1489#ifdef CONFIG_STACK_GROWSUP
1490/*
1491 * vma is the first one with address > vma->vm_end. Have to extend vma.
1492 */
1493int expand_stack(struct vm_area_struct * vma, unsigned long address)
1494{
1495 int error;
1496
1497 if (!(vma->vm_flags & VM_GROWSUP))
1498 return -EFAULT;
1499
1500 /*
1501 * We must make sure the anon_vma is allocated
1502 * so that the anon_vma locking is not a noop.
1503 */
1504 if (unlikely(anon_vma_prepare(vma)))
1505 return -ENOMEM;
1506 anon_vma_lock(vma);
1507
1508 /*
1509 * vma->vm_start/vm_end cannot change under us because the caller
1510 * is required to hold the mmap_sem in read mode. We need the
1511 * anon_vma lock to serialize against concurrent expand_stacks.
1512 */
1513 address += 4 + PAGE_SIZE - 1;
1514 address &= PAGE_MASK;
1515 error = 0;
1516
1517 /* Somebody else might have raced and expanded it already */
1518 if (address > vma->vm_end) {
1519 unsigned long size, grow;
1520
1521 size = address - vma->vm_start;
1522 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1523
1524 error = acct_stack_growth(vma, size, grow);
1525 if (!error)
1526 vma->vm_end = address;
1527 }
1528 anon_vma_unlock(vma);
1529 return error;
1530}
1531
1532struct vm_area_struct *
1533find_extend_vma(struct mm_struct *mm, unsigned long addr)
1534{
1535 struct vm_area_struct *vma, *prev;
1536
1537 addr &= PAGE_MASK;
1538 vma = find_vma_prev(mm, addr, &prev);
1539 if (vma && (vma->vm_start <= addr))
1540 return vma;
1541 if (!prev || expand_stack(prev, addr))
1542 return NULL;
1543 if (prev->vm_flags & VM_LOCKED) {
1544 make_pages_present(addr, prev->vm_end);
1545 }
1546 return prev;
1547}
1548#else
1549/*
1550 * vma is the first one with address < vma->vm_start. Have to extend vma.
1551 */
1552int expand_stack(struct vm_area_struct *vma, unsigned long address)
1553{
1554 int error;
1555
1556 /*
1557 * We must make sure the anon_vma is allocated
1558 * so that the anon_vma locking is not a noop.
1559 */
1560 if (unlikely(anon_vma_prepare(vma)))
1561 return -ENOMEM;
1562 anon_vma_lock(vma);
1563
1564 /*
1565 * vma->vm_start/vm_end cannot change under us because the caller
1566 * is required to hold the mmap_sem in read mode. We need the
1567 * anon_vma lock to serialize against concurrent expand_stacks.
1568 */
1569 address &= PAGE_MASK;
1570 error = 0;
1571
1572 /* Somebody else might have raced and expanded it already */
1573 if (address < vma->vm_start) {
1574 unsigned long size, grow;
1575
1576 size = vma->vm_end - address;
1577 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1578
1579 error = acct_stack_growth(vma, size, grow);
1580 if (!error) {
1581 vma->vm_start = address;
1582 vma->vm_pgoff -= grow;
1583 }
1584 }
1585 anon_vma_unlock(vma);
1586 return error;
1587}
1588
1589struct vm_area_struct *
1590find_extend_vma(struct mm_struct * mm, unsigned long addr)
1591{
1592 struct vm_area_struct * vma;
1593 unsigned long start;
1594
1595 addr &= PAGE_MASK;
1596 vma = find_vma(mm,addr);
1597 if (!vma)
1598 return NULL;
1599 if (vma->vm_start <= addr)
1600 return vma;
1601 if (!(vma->vm_flags & VM_GROWSDOWN))
1602 return NULL;
1603 start = vma->vm_start;
1604 if (expand_stack(vma, addr))
1605 return NULL;
1606 if (vma->vm_flags & VM_LOCKED) {
1607 make_pages_present(addr, start);
1608 }
1609 return vma;
1610}
1611#endif
1612
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613/* Normal function to fix up a mapping
1614 * This function is the default for when an area has no specific
1615 * function. This may be used as part of a more specific routine.
1616 *
1617 * By the time this function is called, the area struct has been
1618 * removed from the process mapping list.
1619 */
1620static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1621{
1622 size_t len = area->vm_end - area->vm_start;
1623
1624 area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1625 if (area->vm_flags & VM_LOCKED)
1626 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1627 vm_stat_unaccount(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 remove_vm_struct(area);
1629}
1630
1631/*
1632 * Update the VMA and inode share lists.
1633 *
1634 * Ok - we have the memory areas we should free on the 'free' list,
1635 * so release them, and do the vma updates.
1636 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001637static void unmap_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
1639 do {
Hugh Dickins146425a2005-04-19 13:29:18 -07001640 struct vm_area_struct *next = vma->vm_next;
1641 unmap_vma(mm, vma);
1642 vma = next;
1643 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 validate_mm(mm);
1645}
1646
1647/*
1648 * Get rid of page table information in the indicated region.
1649 *
1650 * Called with the page table lock held.
1651 */
1652static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001653 struct vm_area_struct *vma, struct vm_area_struct *prev,
1654 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001656 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 struct mmu_gather *tlb;
1658 unsigned long nr_accounted = 0;
1659
1660 lru_add_drain();
Hugh Dickinse0da3822005-04-19 13:29:15 -07001661 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 tlb = tlb_gather_mmu(mm, 0);
1663 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1664 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001665 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001666 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 tlb_finish_mmu(tlb, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001668 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669}
1670
1671/*
1672 * Create a list of vma's touched by the unmap, removing them from the mm's
1673 * vma list as we go..
1674 */
1675static void
1676detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1677 struct vm_area_struct *prev, unsigned long end)
1678{
1679 struct vm_area_struct **insertion_point;
1680 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001681 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
1683 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1684 do {
1685 rb_erase(&vma->vm_rb, &mm->mm_rb);
1686 mm->map_count--;
1687 tail_vma = vma;
1688 vma = vma->vm_next;
1689 } while (vma && vma->vm_start < end);
1690 *insertion_point = vma;
1691 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001692 if (mm->unmap_area == arch_unmap_area)
1693 addr = prev ? prev->vm_end : mm->mmap_base;
1694 else
1695 addr = vma ? vma->vm_start : mm->mmap_base;
1696 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 mm->mmap_cache = NULL; /* Kill the cache. */
1698}
1699
1700/*
1701 * Split a vma into two pieces at address 'addr', a new vma is allocated
1702 * either for the first part or the the tail.
1703 */
1704int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1705 unsigned long addr, int new_below)
1706{
1707 struct mempolicy *pol;
1708 struct vm_area_struct *new;
1709
1710 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1711 return -EINVAL;
1712
1713 if (mm->map_count >= sysctl_max_map_count)
1714 return -ENOMEM;
1715
1716 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1717 if (!new)
1718 return -ENOMEM;
1719
1720 /* most fields are the same, copy all, and then fixup */
1721 *new = *vma;
1722
1723 if (new_below)
1724 new->vm_end = addr;
1725 else {
1726 new->vm_start = addr;
1727 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1728 }
1729
1730 pol = mpol_copy(vma_policy(vma));
1731 if (IS_ERR(pol)) {
1732 kmem_cache_free(vm_area_cachep, new);
1733 return PTR_ERR(pol);
1734 }
1735 vma_set_policy(new, pol);
1736
1737 if (new->vm_file)
1738 get_file(new->vm_file);
1739
1740 if (new->vm_ops && new->vm_ops->open)
1741 new->vm_ops->open(new);
1742
1743 if (new_below)
1744 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1745 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1746 else
1747 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1748
1749 return 0;
1750}
1751
1752/* Munmap is split into 2 main parts -- this part which finds
1753 * what needs doing, and the areas themselves, which do the
1754 * work. This now handles partial unmappings.
1755 * Jeremy Fitzhardinge <jeremy@goop.org>
1756 */
1757int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1758{
1759 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001760 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761
1762 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1763 return -EINVAL;
1764
1765 if ((len = PAGE_ALIGN(len)) == 0)
1766 return -EINVAL;
1767
1768 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001769 vma = find_vma_prev(mm, start, &prev);
1770 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001772 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 /* if it doesn't overlap, we have nothing.. */
1775 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001776 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 return 0;
1778
1779 /*
1780 * If we need to split any vma, do it now to save pain later.
1781 *
1782 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1783 * unmapped vm_area_struct will remain in use: so lower split_vma
1784 * places tmp vma above, and higher split_vma places tmp vma below.
1785 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001786 if (start > vma->vm_start) {
1787 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 if (error)
1789 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001790 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 }
1792
1793 /* Does it split the last one? */
1794 last = find_vma(mm, end);
1795 if (last && end > last->vm_start) {
1796 int error = split_vma(mm, last, end, 1);
1797 if (error)
1798 return error;
1799 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001800 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
1802 /*
1803 * Remove the vma's, and unmap the actual pages
1804 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001805 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1806 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
1808 /* Fix up all other VM information */
Hugh Dickins146425a2005-04-19 13:29:18 -07001809 unmap_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810
1811 return 0;
1812}
1813
1814EXPORT_SYMBOL(do_munmap);
1815
1816asmlinkage long sys_munmap(unsigned long addr, size_t len)
1817{
1818 int ret;
1819 struct mm_struct *mm = current->mm;
1820
1821 profile_munmap(addr);
1822
1823 down_write(&mm->mmap_sem);
1824 ret = do_munmap(mm, addr, len);
1825 up_write(&mm->mmap_sem);
1826 return ret;
1827}
1828
1829static inline void verify_mm_writelocked(struct mm_struct *mm)
1830{
1831#ifdef CONFIG_DEBUG_KERNEL
1832 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1833 WARN_ON(1);
1834 up_read(&mm->mmap_sem);
1835 }
1836#endif
1837}
1838
1839/*
1840 * this is really a simplified "do_mmap". it only handles
1841 * anonymous maps. eventually we may be able to do some
1842 * brk-specific accounting here.
1843 */
1844unsigned long do_brk(unsigned long addr, unsigned long len)
1845{
1846 struct mm_struct * mm = current->mm;
1847 struct vm_area_struct * vma, * prev;
1848 unsigned long flags;
1849 struct rb_node ** rb_link, * rb_parent;
1850 pgoff_t pgoff = addr >> PAGE_SHIFT;
1851
1852 len = PAGE_ALIGN(len);
1853 if (!len)
1854 return addr;
1855
1856 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1857 return -EINVAL;
1858
1859 /*
1860 * mlock MCL_FUTURE?
1861 */
1862 if (mm->def_flags & VM_LOCKED) {
1863 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001864 locked = len >> PAGE_SHIFT;
1865 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07001867 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1869 return -EAGAIN;
1870 }
1871
1872 /*
1873 * mm->mmap_sem is required to protect against another thread
1874 * changing the mappings in case we sleep.
1875 */
1876 verify_mm_writelocked(mm);
1877
1878 /*
1879 * Clear old maps. this also does some error checking for us
1880 */
1881 munmap_back:
1882 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1883 if (vma && vma->vm_start < addr + len) {
1884 if (do_munmap(mm, addr, len))
1885 return -ENOMEM;
1886 goto munmap_back;
1887 }
1888
1889 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001890 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 return -ENOMEM;
1892
1893 if (mm->map_count > sysctl_max_map_count)
1894 return -ENOMEM;
1895
1896 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1897 return -ENOMEM;
1898
1899 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1900
1901 /* Can we just expand an old private anonymous mapping? */
1902 if (vma_merge(mm, prev, addr, addr + len, flags,
1903 NULL, NULL, pgoff, NULL))
1904 goto out;
1905
1906 /*
1907 * create a vma struct for an anonymous mapping
1908 */
1909 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1910 if (!vma) {
1911 vm_unacct_memory(len >> PAGE_SHIFT);
1912 return -ENOMEM;
1913 }
1914 memset(vma, 0, sizeof(*vma));
1915
1916 vma->vm_mm = mm;
1917 vma->vm_start = addr;
1918 vma->vm_end = addr + len;
1919 vma->vm_pgoff = pgoff;
1920 vma->vm_flags = flags;
1921 vma->vm_page_prot = protection_map[flags & 0x0f];
1922 vma_link(mm, vma, prev, rb_link, rb_parent);
1923out:
1924 mm->total_vm += len >> PAGE_SHIFT;
1925 if (flags & VM_LOCKED) {
1926 mm->locked_vm += len >> PAGE_SHIFT;
1927 make_pages_present(addr, addr + len);
1928 }
1929 return addr;
1930}
1931
1932EXPORT_SYMBOL(do_brk);
1933
1934/* Release all mmaps. */
1935void exit_mmap(struct mm_struct *mm)
1936{
1937 struct mmu_gather *tlb;
Hugh Dickinse0da3822005-04-19 13:29:15 -07001938 struct vm_area_struct *vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001940 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
1942 lru_add_drain();
1943
1944 spin_lock(&mm->page_table_lock);
1945
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);
1948 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001949 end = unmap_vmas(&tlb, mm, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001951 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001952 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 mm->mmap = mm->mmap_cache = NULL;
1955 mm->mm_rb = RB_ROOT;
1956 set_mm_counter(mm, rss, 0);
1957 mm->total_vm = 0;
1958 mm->locked_vm = 0;
1959
1960 spin_unlock(&mm->page_table_lock);
1961
1962 /*
1963 * Walk the list again, actually closing and freeing it
1964 * without holding any MM locks.
1965 */
1966 while (vma) {
1967 struct vm_area_struct *next = vma->vm_next;
1968 remove_vm_struct(vma);
1969 vma = next;
1970 }
Hugh Dickinse0da3822005-04-19 13:29:15 -07001971
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001972 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973}
1974
1975/* Insert vm structure into process list sorted by address
1976 * and into the inode's i_mmap tree. If vm_file is non-NULL
1977 * then i_mmap_lock is taken here.
1978 */
1979int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1980{
1981 struct vm_area_struct * __vma, * prev;
1982 struct rb_node ** rb_link, * rb_parent;
1983
1984 /*
1985 * The vm_pgoff of a purely anonymous vma should be irrelevant
1986 * until its first write fault, when page's anon_vma and index
1987 * are set. But now set the vm_pgoff it will almost certainly
1988 * end up with (unless mremap moves it elsewhere before that
1989 * first wfault), so /proc/pid/maps tells a consistent story.
1990 *
1991 * By setting it to reflect the virtual start address of the
1992 * vma, merges and splits can happen in a seamless way, just
1993 * using the existing file pgoff checks and manipulations.
1994 * Similarly in do_mmap_pgoff and in do_brk.
1995 */
1996 if (!vma->vm_file) {
1997 BUG_ON(vma->anon_vma);
1998 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1999 }
2000 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2001 if (__vma && __vma->vm_start < vma->vm_end)
2002 return -ENOMEM;
2003 vma_link(mm, vma, prev, rb_link, rb_parent);
2004 return 0;
2005}
2006
2007/*
2008 * Copy the vma structure to a new location in the same mm,
2009 * prior to moving page table entries, to effect an mremap move.
2010 */
2011struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2012 unsigned long addr, unsigned long len, pgoff_t pgoff)
2013{
2014 struct vm_area_struct *vma = *vmap;
2015 unsigned long vma_start = vma->vm_start;
2016 struct mm_struct *mm = vma->vm_mm;
2017 struct vm_area_struct *new_vma, *prev;
2018 struct rb_node **rb_link, *rb_parent;
2019 struct mempolicy *pol;
2020
2021 /*
2022 * If anonymous vma has not yet been faulted, update new pgoff
2023 * to match new location, to increase its chance of merging.
2024 */
2025 if (!vma->vm_file && !vma->anon_vma)
2026 pgoff = addr >> PAGE_SHIFT;
2027
2028 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2029 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2030 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2031 if (new_vma) {
2032 /*
2033 * Source vma may have been merged into new_vma
2034 */
2035 if (vma_start >= new_vma->vm_start &&
2036 vma_start < new_vma->vm_end)
2037 *vmap = new_vma;
2038 } else {
2039 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2040 if (new_vma) {
2041 *new_vma = *vma;
2042 pol = mpol_copy(vma_policy(vma));
2043 if (IS_ERR(pol)) {
2044 kmem_cache_free(vm_area_cachep, new_vma);
2045 return NULL;
2046 }
2047 vma_set_policy(new_vma, pol);
2048 new_vma->vm_start = addr;
2049 new_vma->vm_end = addr + len;
2050 new_vma->vm_pgoff = pgoff;
2051 if (new_vma->vm_file)
2052 get_file(new_vma->vm_file);
2053 if (new_vma->vm_ops && new_vma->vm_ops->open)
2054 new_vma->vm_ops->open(new_vma);
2055 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2056 }
2057 }
2058 return new_vma;
2059}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002060
2061/*
2062 * Return true if the calling process may expand its vm space by the passed
2063 * number of pages
2064 */
2065int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2066{
2067 unsigned long cur = mm->total_vm; /* pages */
2068 unsigned long lim;
2069
2070 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2071
2072 if (cur + npages > lim)
2073 return 0;
2074 return 1;
2075}