blob: da3fa90a0aae2b9af04ec553209deb6a51b80b48 [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
146 if (atomic_read(&vm_committed_space) < allowed)
147 return 0;
148
149 vm_unacct_memory(pages);
150
151 return -ENOMEM;
152}
153
154EXPORT_SYMBOL(sysctl_overcommit_memory);
155EXPORT_SYMBOL(sysctl_overcommit_ratio);
156EXPORT_SYMBOL(sysctl_max_map_count);
157EXPORT_SYMBOL(vm_committed_space);
158EXPORT_SYMBOL(__vm_enough_memory);
159
160/*
161 * Requires inode->i_mapping->i_mmap_lock
162 */
163static void __remove_shared_vm_struct(struct vm_area_struct *vma,
164 struct file *file, struct address_space *mapping)
165{
166 if (vma->vm_flags & VM_DENYWRITE)
167 atomic_inc(&file->f_dentry->d_inode->i_writecount);
168 if (vma->vm_flags & VM_SHARED)
169 mapping->i_mmap_writable--;
170
171 flush_dcache_mmap_lock(mapping);
172 if (unlikely(vma->vm_flags & VM_NONLINEAR))
173 list_del_init(&vma->shared.vm_set.list);
174 else
175 vma_prio_tree_remove(vma, &mapping->i_mmap);
176 flush_dcache_mmap_unlock(mapping);
177}
178
179/*
180 * Remove one vm structure and free it.
181 */
182static void remove_vm_struct(struct vm_area_struct *vma)
183{
184 struct file *file = vma->vm_file;
185
186 might_sleep();
187 if (file) {
188 struct address_space *mapping = file->f_mapping;
189 spin_lock(&mapping->i_mmap_lock);
190 __remove_shared_vm_struct(vma, file, mapping);
191 spin_unlock(&mapping->i_mmap_lock);
192 }
193 if (vma->vm_ops && vma->vm_ops->close)
194 vma->vm_ops->close(vma);
195 if (file)
196 fput(file);
197 anon_vma_unlink(vma);
198 mpol_free(vma_policy(vma));
199 kmem_cache_free(vm_area_cachep, vma);
200}
201
202/*
203 * sys_brk() for the most part doesn't need the global kernel
204 * lock, except when an application is doing something nasty
205 * like trying to un-brk an area that has already been mapped
206 * to a regular file. in this case, the unmapping will need
207 * to invoke file system routines that need the global lock.
208 */
209asmlinkage unsigned long sys_brk(unsigned long brk)
210{
211 unsigned long rlim, retval;
212 unsigned long newbrk, oldbrk;
213 struct mm_struct *mm = current->mm;
214
215 down_write(&mm->mmap_sem);
216
217 if (brk < mm->end_code)
218 goto out;
219 newbrk = PAGE_ALIGN(brk);
220 oldbrk = PAGE_ALIGN(mm->brk);
221 if (oldbrk == newbrk)
222 goto set_brk;
223
224 /* Always allow shrinking brk. */
225 if (brk <= mm->brk) {
226 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
227 goto set_brk;
228 goto out;
229 }
230
231 /* Check against rlimit.. */
232 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
233 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
234 goto out;
235
236 /* Check against existing mmap mappings. */
237 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
238 goto out;
239
240 /* Ok, looks good - let it rip. */
241 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
242 goto out;
243set_brk:
244 mm->brk = brk;
245out:
246 retval = mm->brk;
247 up_write(&mm->mmap_sem);
248 return retval;
249}
250
251#ifdef DEBUG_MM_RB
252static int browse_rb(struct rb_root *root)
253{
254 int i = 0, j;
255 struct rb_node *nd, *pn = NULL;
256 unsigned long prev = 0, pend = 0;
257
258 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
259 struct vm_area_struct *vma;
260 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
261 if (vma->vm_start < prev)
262 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
263 if (vma->vm_start < pend)
264 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
265 if (vma->vm_start > vma->vm_end)
266 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
267 i++;
268 pn = nd;
269 }
270 j = 0;
271 for (nd = pn; nd; nd = rb_prev(nd)) {
272 j++;
273 }
274 if (i != j)
275 printk("backwards %d, forwards %d\n", j, i), i = 0;
276 return i;
277}
278
279void validate_mm(struct mm_struct *mm)
280{
281 int bug = 0;
282 int i = 0;
283 struct vm_area_struct *tmp = mm->mmap;
284 while (tmp) {
285 tmp = tmp->vm_next;
286 i++;
287 }
288 if (i != mm->map_count)
289 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
290 i = browse_rb(&mm->mm_rb);
291 if (i != mm->map_count)
292 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
293 if (bug)
294 BUG();
295}
296#else
297#define validate_mm(mm) do { } while (0)
298#endif
299
300static struct vm_area_struct *
301find_vma_prepare(struct mm_struct *mm, unsigned long addr,
302 struct vm_area_struct **pprev, struct rb_node ***rb_link,
303 struct rb_node ** rb_parent)
304{
305 struct vm_area_struct * vma;
306 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
307
308 __rb_link = &mm->mm_rb.rb_node;
309 rb_prev = __rb_parent = NULL;
310 vma = NULL;
311
312 while (*__rb_link) {
313 struct vm_area_struct *vma_tmp;
314
315 __rb_parent = *__rb_link;
316 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
317
318 if (vma_tmp->vm_end > addr) {
319 vma = vma_tmp;
320 if (vma_tmp->vm_start <= addr)
321 return vma;
322 __rb_link = &__rb_parent->rb_left;
323 } else {
324 rb_prev = __rb_parent;
325 __rb_link = &__rb_parent->rb_right;
326 }
327 }
328
329 *pprev = NULL;
330 if (rb_prev)
331 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
332 *rb_link = __rb_link;
333 *rb_parent = __rb_parent;
334 return vma;
335}
336
337static inline void
338__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
339 struct vm_area_struct *prev, struct rb_node *rb_parent)
340{
341 if (prev) {
342 vma->vm_next = prev->vm_next;
343 prev->vm_next = vma;
344 } else {
345 mm->mmap = vma;
346 if (rb_parent)
347 vma->vm_next = rb_entry(rb_parent,
348 struct vm_area_struct, vm_rb);
349 else
350 vma->vm_next = NULL;
351 }
352}
353
354void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
355 struct rb_node **rb_link, struct rb_node *rb_parent)
356{
357 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
358 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
359}
360
361static inline void __vma_link_file(struct vm_area_struct *vma)
362{
363 struct file * file;
364
365 file = vma->vm_file;
366 if (file) {
367 struct address_space *mapping = file->f_mapping;
368
369 if (vma->vm_flags & VM_DENYWRITE)
370 atomic_dec(&file->f_dentry->d_inode->i_writecount);
371 if (vma->vm_flags & VM_SHARED)
372 mapping->i_mmap_writable++;
373
374 flush_dcache_mmap_lock(mapping);
375 if (unlikely(vma->vm_flags & VM_NONLINEAR))
376 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
377 else
378 vma_prio_tree_insert(vma, &mapping->i_mmap);
379 flush_dcache_mmap_unlock(mapping);
380 }
381}
382
383static void
384__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
385 struct vm_area_struct *prev, struct rb_node **rb_link,
386 struct rb_node *rb_parent)
387{
388 __vma_link_list(mm, vma, prev, rb_parent);
389 __vma_link_rb(mm, vma, rb_link, rb_parent);
390 __anon_vma_link(vma);
391}
392
393static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
394 struct vm_area_struct *prev, struct rb_node **rb_link,
395 struct rb_node *rb_parent)
396{
397 struct address_space *mapping = NULL;
398
399 if (vma->vm_file)
400 mapping = vma->vm_file->f_mapping;
401
402 if (mapping) {
403 spin_lock(&mapping->i_mmap_lock);
404 vma->vm_truncate_count = mapping->truncate_count;
405 }
406 anon_vma_lock(vma);
407
408 __vma_link(mm, vma, prev, rb_link, rb_parent);
409 __vma_link_file(vma);
410
411 anon_vma_unlock(vma);
412 if (mapping)
413 spin_unlock(&mapping->i_mmap_lock);
414
415 mm->map_count++;
416 validate_mm(mm);
417}
418
419/*
420 * Helper for vma_adjust in the split_vma insert case:
421 * insert vm structure into list and rbtree and anon_vma,
422 * but it has already been inserted into prio_tree earlier.
423 */
424static void
425__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
426{
427 struct vm_area_struct * __vma, * prev;
428 struct rb_node ** rb_link, * rb_parent;
429
430 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
431 if (__vma && __vma->vm_start < vma->vm_end)
432 BUG();
433 __vma_link(mm, vma, prev, rb_link, rb_parent);
434 mm->map_count++;
435}
436
437static inline void
438__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
439 struct vm_area_struct *prev)
440{
441 prev->vm_next = vma->vm_next;
442 rb_erase(&vma->vm_rb, &mm->mm_rb);
443 if (mm->mmap_cache == vma)
444 mm->mmap_cache = prev;
445}
446
447/*
448 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
449 * is already present in an i_mmap tree without adjusting the tree.
450 * The following helper function should be used when such adjustments
451 * are necessary. The "insert" vma (if any) is to be inserted
452 * before we drop the necessary locks.
453 */
454void vma_adjust(struct vm_area_struct *vma, unsigned long start,
455 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
456{
457 struct mm_struct *mm = vma->vm_mm;
458 struct vm_area_struct *next = vma->vm_next;
459 struct vm_area_struct *importer = NULL;
460 struct address_space *mapping = NULL;
461 struct prio_tree_root *root = NULL;
462 struct file *file = vma->vm_file;
463 struct anon_vma *anon_vma = NULL;
464 long adjust_next = 0;
465 int remove_next = 0;
466
467 if (next && !insert) {
468 if (end >= next->vm_end) {
469 /*
470 * vma expands, overlapping all the next, and
471 * perhaps the one after too (mprotect case 6).
472 */
473again: remove_next = 1 + (end > next->vm_end);
474 end = next->vm_end;
475 anon_vma = next->anon_vma;
476 importer = vma;
477 } else if (end > next->vm_start) {
478 /*
479 * vma expands, overlapping part of the next:
480 * mprotect case 5 shifting the boundary up.
481 */
482 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
483 anon_vma = next->anon_vma;
484 importer = vma;
485 } else if (end < vma->vm_end) {
486 /*
487 * vma shrinks, and !insert tells it's not
488 * split_vma inserting another: so it must be
489 * mprotect case 4 shifting the boundary down.
490 */
491 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
492 anon_vma = next->anon_vma;
493 importer = next;
494 }
495 }
496
497 if (file) {
498 mapping = file->f_mapping;
499 if (!(vma->vm_flags & VM_NONLINEAR))
500 root = &mapping->i_mmap;
501 spin_lock(&mapping->i_mmap_lock);
502 if (importer &&
503 vma->vm_truncate_count != next->vm_truncate_count) {
504 /*
505 * unmap_mapping_range might be in progress:
506 * ensure that the expanding vma is rescanned.
507 */
508 importer->vm_truncate_count = 0;
509 }
510 if (insert) {
511 insert->vm_truncate_count = vma->vm_truncate_count;
512 /*
513 * Put into prio_tree now, so instantiated pages
514 * are visible to arm/parisc __flush_dcache_page
515 * throughout; but we cannot insert into address
516 * space until vma start or end is updated.
517 */
518 __vma_link_file(insert);
519 }
520 }
521
522 /*
523 * When changing only vma->vm_end, we don't really need
524 * anon_vma lock: but is that case worth optimizing out?
525 */
526 if (vma->anon_vma)
527 anon_vma = vma->anon_vma;
528 if (anon_vma) {
529 spin_lock(&anon_vma->lock);
530 /*
531 * Easily overlooked: when mprotect shifts the boundary,
532 * make sure the expanding vma has anon_vma set if the
533 * shrinking vma had, to cover any anon pages imported.
534 */
535 if (importer && !importer->anon_vma) {
536 importer->anon_vma = anon_vma;
537 __anon_vma_link(importer);
538 }
539 }
540
541 if (root) {
542 flush_dcache_mmap_lock(mapping);
543 vma_prio_tree_remove(vma, root);
544 if (adjust_next)
545 vma_prio_tree_remove(next, root);
546 }
547
548 vma->vm_start = start;
549 vma->vm_end = end;
550 vma->vm_pgoff = pgoff;
551 if (adjust_next) {
552 next->vm_start += adjust_next << PAGE_SHIFT;
553 next->vm_pgoff += adjust_next;
554 }
555
556 if (root) {
557 if (adjust_next)
558 vma_prio_tree_insert(next, root);
559 vma_prio_tree_insert(vma, root);
560 flush_dcache_mmap_unlock(mapping);
561 }
562
563 if (remove_next) {
564 /*
565 * vma_merge has merged next into vma, and needs
566 * us to remove next before dropping the locks.
567 */
568 __vma_unlink(mm, next, vma);
569 if (file)
570 __remove_shared_vm_struct(next, file, mapping);
571 if (next->anon_vma)
572 __anon_vma_merge(vma, next);
573 } else if (insert) {
574 /*
575 * split_vma has split insert from vma, and needs
576 * us to insert it before dropping the locks
577 * (it may either follow vma or precede it).
578 */
579 __insert_vm_struct(mm, insert);
580 }
581
582 if (anon_vma)
583 spin_unlock(&anon_vma->lock);
584 if (mapping)
585 spin_unlock(&mapping->i_mmap_lock);
586
587 if (remove_next) {
588 if (file)
589 fput(file);
590 mm->map_count--;
591 mpol_free(vma_policy(next));
592 kmem_cache_free(vm_area_cachep, next);
593 /*
594 * In mprotect's case 6 (see comments on vma_merge),
595 * we must remove another next too. It would clutter
596 * up the code too much to do both in one go.
597 */
598 if (remove_next == 2) {
599 next = vma->vm_next;
600 goto again;
601 }
602 }
603
604 validate_mm(mm);
605}
606
607/*
608 * If the vma has a ->close operation then the driver probably needs to release
609 * per-vma resources, so we don't attempt to merge those.
610 */
611#define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
612
613static inline int is_mergeable_vma(struct vm_area_struct *vma,
614 struct file *file, unsigned long vm_flags)
615{
616 if (vma->vm_flags != vm_flags)
617 return 0;
618 if (vma->vm_file != file)
619 return 0;
620 if (vma->vm_ops && vma->vm_ops->close)
621 return 0;
622 return 1;
623}
624
625static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
626 struct anon_vma *anon_vma2)
627{
628 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
629}
630
631/*
632 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
633 * in front of (at a lower virtual address and file offset than) the vma.
634 *
635 * We cannot merge two vmas if they have differently assigned (non-NULL)
636 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
637 *
638 * We don't check here for the merged mmap wrapping around the end of pagecache
639 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
640 * wrap, nor mmaps which cover the final page at index -1UL.
641 */
642static int
643can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
644 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
645{
646 if (is_mergeable_vma(vma, file, vm_flags) &&
647 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
648 if (vma->vm_pgoff == vm_pgoff)
649 return 1;
650 }
651 return 0;
652}
653
654/*
655 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
656 * beyond (at a higher virtual address and file offset than) the vma.
657 *
658 * We cannot merge two vmas if they have differently assigned (non-NULL)
659 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
660 */
661static int
662can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
663 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
664{
665 if (is_mergeable_vma(vma, file, vm_flags) &&
666 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
667 pgoff_t vm_pglen;
668 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
669 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
670 return 1;
671 }
672 return 0;
673}
674
675/*
676 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
677 * whether that can be merged with its predecessor or its successor.
678 * Or both (it neatly fills a hole).
679 *
680 * In most cases - when called for mmap, brk or mremap - [addr,end) is
681 * certain not to be mapped by the time vma_merge is called; but when
682 * called for mprotect, it is certain to be already mapped (either at
683 * an offset within prev, or at the start of next), and the flags of
684 * this area are about to be changed to vm_flags - and the no-change
685 * case has already been eliminated.
686 *
687 * The following mprotect cases have to be considered, where AAAA is
688 * the area passed down from mprotect_fixup, never extending beyond one
689 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
690 *
691 * AAAA AAAA AAAA AAAA
692 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
693 * cannot merge might become might become might become
694 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
695 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
696 * mremap move: PPPPNNNNNNNN 8
697 * AAAA
698 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
699 * might become case 1 below case 2 below case 3 below
700 *
701 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
702 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
703 */
704struct vm_area_struct *vma_merge(struct mm_struct *mm,
705 struct vm_area_struct *prev, unsigned long addr,
706 unsigned long end, unsigned long vm_flags,
707 struct anon_vma *anon_vma, struct file *file,
708 pgoff_t pgoff, struct mempolicy *policy)
709{
710 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
711 struct vm_area_struct *area, *next;
712
713 /*
714 * We later require that vma->vm_flags == vm_flags,
715 * so this tests vma->vm_flags & VM_SPECIAL, too.
716 */
717 if (vm_flags & VM_SPECIAL)
718 return NULL;
719
720 if (prev)
721 next = prev->vm_next;
722 else
723 next = mm->mmap;
724 area = next;
725 if (next && next->vm_end == end) /* cases 6, 7, 8 */
726 next = next->vm_next;
727
728 /*
729 * Can it merge with the predecessor?
730 */
731 if (prev && prev->vm_end == addr &&
732 mpol_equal(vma_policy(prev), policy) &&
733 can_vma_merge_after(prev, vm_flags,
734 anon_vma, file, pgoff)) {
735 /*
736 * OK, it can. Can we now merge in the successor as well?
737 */
738 if (next && end == next->vm_start &&
739 mpol_equal(policy, vma_policy(next)) &&
740 can_vma_merge_before(next, vm_flags,
741 anon_vma, file, pgoff+pglen) &&
742 is_mergeable_anon_vma(prev->anon_vma,
743 next->anon_vma)) {
744 /* cases 1, 6 */
745 vma_adjust(prev, prev->vm_start,
746 next->vm_end, prev->vm_pgoff, NULL);
747 } else /* cases 2, 5, 7 */
748 vma_adjust(prev, prev->vm_start,
749 end, prev->vm_pgoff, NULL);
750 return prev;
751 }
752
753 /*
754 * Can this new request be merged in front of next?
755 */
756 if (next && end == next->vm_start &&
757 mpol_equal(policy, vma_policy(next)) &&
758 can_vma_merge_before(next, vm_flags,
759 anon_vma, file, pgoff+pglen)) {
760 if (prev && addr < prev->vm_end) /* case 4 */
761 vma_adjust(prev, prev->vm_start,
762 addr, prev->vm_pgoff, NULL);
763 else /* cases 3, 8 */
764 vma_adjust(area, addr, next->vm_end,
765 next->vm_pgoff - pglen, NULL);
766 return area;
767 }
768
769 return NULL;
770}
771
772/*
773 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
774 * neighbouring vmas for a suitable anon_vma, before it goes off
775 * to allocate a new anon_vma. It checks because a repetitive
776 * sequence of mprotects and faults may otherwise lead to distinct
777 * anon_vmas being allocated, preventing vma merge in subsequent
778 * mprotect.
779 */
780struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
781{
782 struct vm_area_struct *near;
783 unsigned long vm_flags;
784
785 near = vma->vm_next;
786 if (!near)
787 goto try_prev;
788
789 /*
790 * Since only mprotect tries to remerge vmas, match flags
791 * which might be mprotected into each other later on.
792 * Neither mlock nor madvise tries to remerge at present,
793 * so leave their flags as obstructing a merge.
794 */
795 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
796 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
797
798 if (near->anon_vma && vma->vm_end == near->vm_start &&
799 mpol_equal(vma_policy(vma), vma_policy(near)) &&
800 can_vma_merge_before(near, vm_flags,
801 NULL, vma->vm_file, vma->vm_pgoff +
802 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
803 return near->anon_vma;
804try_prev:
805 /*
806 * It is potentially slow to have to call find_vma_prev here.
807 * But it's only on the first write fault on the vma, not
808 * every time, and we could devise a way to avoid it later
809 * (e.g. stash info in next's anon_vma_node when assigning
810 * an anon_vma, or when trying vma_merge). Another time.
811 */
812 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
813 BUG();
814 if (!near)
815 goto none;
816
817 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
818 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
819
820 if (near->anon_vma && near->vm_end == vma->vm_start &&
821 mpol_equal(vma_policy(near), vma_policy(vma)) &&
822 can_vma_merge_after(near, vm_flags,
823 NULL, vma->vm_file, vma->vm_pgoff))
824 return near->anon_vma;
825none:
826 /*
827 * There's no absolute need to look only at touching neighbours:
828 * we could search further afield for "compatible" anon_vmas.
829 * But it would probably just be a waste of time searching,
830 * or lead to too many vmas hanging off the same anon_vma.
831 * We're trying to allow mprotect remerging later on,
832 * not trying to minimize memory used for anon_vmas.
833 */
834 return NULL;
835}
836
837#ifdef CONFIG_PROC_FS
838void __vm_stat_account(struct mm_struct *mm, unsigned long flags,
839 struct file *file, long pages)
840{
841 const unsigned long stack_flags
842 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
843
844#ifdef CONFIG_HUGETLB
845 if (flags & VM_HUGETLB) {
846 if (!(flags & VM_DONTCOPY))
847 mm->shared_vm += pages;
848 return;
849 }
850#endif /* CONFIG_HUGETLB */
851
852 if (file) {
853 mm->shared_vm += pages;
854 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
855 mm->exec_vm += pages;
856 } else if (flags & stack_flags)
857 mm->stack_vm += pages;
858 if (flags & (VM_RESERVED|VM_IO))
859 mm->reserved_vm += pages;
860}
861#endif /* CONFIG_PROC_FS */
862
863/*
864 * The caller must hold down_write(current->mm->mmap_sem).
865 */
866
867unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
868 unsigned long len, unsigned long prot,
869 unsigned long flags, unsigned long pgoff)
870{
871 struct mm_struct * mm = current->mm;
872 struct vm_area_struct * vma, * prev;
873 struct inode *inode;
874 unsigned int vm_flags;
875 int correct_wcount = 0;
876 int error;
877 struct rb_node ** rb_link, * rb_parent;
878 int accountable = 1;
879 unsigned long charged = 0, reqprot = prot;
880
881 if (file) {
882 if (is_file_hugepages(file))
883 accountable = 0;
884
885 if (!file->f_op || !file->f_op->mmap)
886 return -ENODEV;
887
888 if ((prot & PROT_EXEC) &&
889 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
890 return -EPERM;
891 }
892 /*
893 * Does the application expect PROT_READ to imply PROT_EXEC?
894 *
895 * (the exception is when the underlying filesystem is noexec
896 * mounted, in which case we dont add PROT_EXEC.)
897 */
898 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
899 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
900 prot |= PROT_EXEC;
901
902 if (!len)
903 return -EINVAL;
904
905 /* Careful about overflows.. */
906 len = PAGE_ALIGN(len);
907 if (!len || len > TASK_SIZE)
908 return -ENOMEM;
909
910 /* offset overflow? */
911 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
912 return -EOVERFLOW;
913
914 /* Too many mappings? */
915 if (mm->map_count > sysctl_max_map_count)
916 return -ENOMEM;
917
918 /* Obtain the address to map to. we verify (or select) it and ensure
919 * that it represents a valid section of the address space.
920 */
921 addr = get_unmapped_area(file, addr, len, pgoff, flags);
922 if (addr & ~PAGE_MASK)
923 return addr;
924
925 /* Do simple checking here so the lower-level routines won't have
926 * to. we assume access permissions have been handled by the open
927 * of the memory object, so we don't do any here.
928 */
929 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
930 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
931
932 if (flags & MAP_LOCKED) {
933 if (!can_do_mlock())
934 return -EPERM;
935 vm_flags |= VM_LOCKED;
936 }
937 /* mlock MCL_FUTURE? */
938 if (vm_flags & VM_LOCKED) {
939 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700940 locked = len >> PAGE_SHIFT;
941 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700943 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
945 return -EAGAIN;
946 }
947
948 inode = file ? file->f_dentry->d_inode : NULL;
949
950 if (file) {
951 switch (flags & MAP_TYPE) {
952 case MAP_SHARED:
953 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
954 return -EACCES;
955
956 /*
957 * Make sure we don't allow writing to an append-only
958 * file..
959 */
960 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
961 return -EACCES;
962
963 /*
964 * Make sure there are no mandatory locks on the file.
965 */
966 if (locks_verify_locked(inode))
967 return -EAGAIN;
968
969 vm_flags |= VM_SHARED | VM_MAYSHARE;
970 if (!(file->f_mode & FMODE_WRITE))
971 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
972
973 /* fall through */
974 case MAP_PRIVATE:
975 if (!(file->f_mode & FMODE_READ))
976 return -EACCES;
977 break;
978
979 default:
980 return -EINVAL;
981 }
982 } else {
983 switch (flags & MAP_TYPE) {
984 case MAP_SHARED:
985 vm_flags |= VM_SHARED | VM_MAYSHARE;
986 break;
987 case MAP_PRIVATE:
988 /*
989 * Set pgoff according to addr for anon_vma.
990 */
991 pgoff = addr >> PAGE_SHIFT;
992 break;
993 default:
994 return -EINVAL;
995 }
996 }
997
998 error = security_file_mmap(file, reqprot, prot, flags);
999 if (error)
1000 return error;
1001
1002 /* Clear old maps */
1003 error = -ENOMEM;
1004munmap_back:
1005 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1006 if (vma && vma->vm_start < addr + len) {
1007 if (do_munmap(mm, addr, len))
1008 return -ENOMEM;
1009 goto munmap_back;
1010 }
1011
1012 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001013 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 return -ENOMEM;
1015
1016 if (accountable && (!(flags & MAP_NORESERVE) ||
1017 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1018 if (vm_flags & VM_SHARED) {
1019 /* Check memory availability in shmem_file_setup? */
1020 vm_flags |= VM_ACCOUNT;
1021 } else if (vm_flags & VM_WRITE) {
1022 /*
1023 * Private writable mapping: check memory availability
1024 */
1025 charged = len >> PAGE_SHIFT;
1026 if (security_vm_enough_memory(charged))
1027 return -ENOMEM;
1028 vm_flags |= VM_ACCOUNT;
1029 }
1030 }
1031
1032 /*
1033 * Can we just expand an old private anonymous mapping?
1034 * The VM_SHARED test is necessary because shmem_zero_setup
1035 * will create the file object for a shared anonymous map below.
1036 */
1037 if (!file && !(vm_flags & VM_SHARED) &&
1038 vma_merge(mm, prev, addr, addr + len, vm_flags,
1039 NULL, NULL, pgoff, NULL))
1040 goto out;
1041
1042 /*
1043 * Determine the object being mapped and call the appropriate
1044 * specific mapper. the address has already been validated, but
1045 * not unmapped, but the maps are removed from the list.
1046 */
1047 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1048 if (!vma) {
1049 error = -ENOMEM;
1050 goto unacct_error;
1051 }
1052 memset(vma, 0, sizeof(*vma));
1053
1054 vma->vm_mm = mm;
1055 vma->vm_start = addr;
1056 vma->vm_end = addr + len;
1057 vma->vm_flags = vm_flags;
1058 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1059 vma->vm_pgoff = pgoff;
1060
1061 if (file) {
1062 error = -EINVAL;
1063 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1064 goto free_vma;
1065 if (vm_flags & VM_DENYWRITE) {
1066 error = deny_write_access(file);
1067 if (error)
1068 goto free_vma;
1069 correct_wcount = 1;
1070 }
1071 vma->vm_file = file;
1072 get_file(file);
1073 error = file->f_op->mmap(file, vma);
1074 if (error)
1075 goto unmap_and_free_vma;
1076 } else if (vm_flags & VM_SHARED) {
1077 error = shmem_zero_setup(vma);
1078 if (error)
1079 goto free_vma;
1080 }
1081
1082 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1083 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1084 * that memory reservation must be checked; but that reservation
1085 * belongs to shared memory object, not to vma: so now clear it.
1086 */
1087 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1088 vma->vm_flags &= ~VM_ACCOUNT;
1089
1090 /* Can addr have changed??
1091 *
1092 * Answer: Yes, several device drivers can do it in their
1093 * f_op->mmap method. -DaveM
1094 */
1095 addr = vma->vm_start;
1096 pgoff = vma->vm_pgoff;
1097 vm_flags = vma->vm_flags;
1098
1099 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1100 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1101 file = vma->vm_file;
1102 vma_link(mm, vma, prev, rb_link, rb_parent);
1103 if (correct_wcount)
1104 atomic_inc(&inode->i_writecount);
1105 } else {
1106 if (file) {
1107 if (correct_wcount)
1108 atomic_inc(&inode->i_writecount);
1109 fput(file);
1110 }
1111 mpol_free(vma_policy(vma));
1112 kmem_cache_free(vm_area_cachep, vma);
1113 }
1114out:
1115 mm->total_vm += len >> PAGE_SHIFT;
1116 __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1117 if (vm_flags & VM_LOCKED) {
1118 mm->locked_vm += len >> PAGE_SHIFT;
1119 make_pages_present(addr, addr + len);
1120 }
1121 if (flags & MAP_POPULATE) {
1122 up_write(&mm->mmap_sem);
1123 sys_remap_file_pages(addr, len, 0,
1124 pgoff, flags & MAP_NONBLOCK);
1125 down_write(&mm->mmap_sem);
1126 }
1127 return addr;
1128
1129unmap_and_free_vma:
1130 if (correct_wcount)
1131 atomic_inc(&inode->i_writecount);
1132 vma->vm_file = NULL;
1133 fput(file);
1134
1135 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001136 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1137 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138free_vma:
1139 kmem_cache_free(vm_area_cachep, vma);
1140unacct_error:
1141 if (charged)
1142 vm_unacct_memory(charged);
1143 return error;
1144}
1145
1146EXPORT_SYMBOL(do_mmap_pgoff);
1147
1148/* Get an address range which is currently unmapped.
1149 * For shmat() with addr=0.
1150 *
1151 * Ugly calling convention alert:
1152 * Return value with the low bits set means error value,
1153 * ie
1154 * if (ret & ~PAGE_MASK)
1155 * error = ret;
1156 *
1157 * This function "knows" that -ENOMEM has the bits set.
1158 */
1159#ifndef HAVE_ARCH_UNMAPPED_AREA
1160unsigned long
1161arch_get_unmapped_area(struct file *filp, unsigned long addr,
1162 unsigned long len, unsigned long pgoff, unsigned long flags)
1163{
1164 struct mm_struct *mm = current->mm;
1165 struct vm_area_struct *vma;
1166 unsigned long start_addr;
1167
1168 if (len > TASK_SIZE)
1169 return -ENOMEM;
1170
1171 if (addr) {
1172 addr = PAGE_ALIGN(addr);
1173 vma = find_vma(mm, addr);
1174 if (TASK_SIZE - len >= addr &&
1175 (!vma || addr + len <= vma->vm_start))
1176 return addr;
1177 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001178 if (len > mm->cached_hole_size) {
1179 start_addr = addr = mm->free_area_cache;
1180 } else {
1181 start_addr = addr = TASK_UNMAPPED_BASE;
1182 mm->cached_hole_size = 0;
1183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
1185full_search:
1186 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1187 /* At this point: (!vma || addr < vma->vm_end). */
1188 if (TASK_SIZE - len < addr) {
1189 /*
1190 * Start a new search - just in case we missed
1191 * some holes.
1192 */
1193 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001194 addr = TASK_UNMAPPED_BASE;
1195 start_addr = addr;
1196 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 goto full_search;
1198 }
1199 return -ENOMEM;
1200 }
1201 if (!vma || addr + len <= vma->vm_start) {
1202 /*
1203 * Remember the place where we stopped the search:
1204 */
1205 mm->free_area_cache = addr + len;
1206 return addr;
1207 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001208 if (addr + mm->cached_hole_size < vma->vm_start)
1209 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 addr = vma->vm_end;
1211 }
1212}
1213#endif
1214
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001215void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
1217 /*
1218 * Is this a new hole at the lowest possible address?
1219 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001220 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1221 mm->free_area_cache = addr;
1222 mm->cached_hole_size = ~0UL;
1223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224}
1225
1226/*
1227 * This mmap-allocator allocates new areas top-down from below the
1228 * stack's low limit (the base):
1229 */
1230#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1231unsigned long
1232arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1233 const unsigned long len, const unsigned long pgoff,
1234 const unsigned long flags)
1235{
1236 struct vm_area_struct *vma;
1237 struct mm_struct *mm = current->mm;
1238 unsigned long addr = addr0;
1239
1240 /* requested length too big for entire address space */
1241 if (len > TASK_SIZE)
1242 return -ENOMEM;
1243
1244 /* requesting a specific address */
1245 if (addr) {
1246 addr = PAGE_ALIGN(addr);
1247 vma = find_vma(mm, addr);
1248 if (TASK_SIZE - len >= addr &&
1249 (!vma || addr + len <= vma->vm_start))
1250 return addr;
1251 }
1252
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001253 /* check if free_area_cache is useful for us */
1254 if (len <= mm->cached_hole_size) {
1255 mm->cached_hole_size = 0;
1256 mm->free_area_cache = mm->mmap_base;
1257 }
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 /* either no address requested or can't fit in requested address hole */
1260 addr = mm->free_area_cache;
1261
1262 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001263 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 vma = find_vma(mm, addr-len);
1265 if (!vma || addr <= vma->vm_start)
1266 /* remember the address as a hint for next time */
1267 return (mm->free_area_cache = addr-len);
1268 }
1269
Chris Wright73219d12005-06-21 17:14:52 -07001270 if (mm->mmap_base < len)
1271 goto bottomup;
1272
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 addr = mm->mmap_base-len;
1274
1275 do {
1276 /*
1277 * Lookup failure means no vma is above this address,
1278 * else if new region fits below vma->vm_start,
1279 * return with success:
1280 */
1281 vma = find_vma(mm, addr);
1282 if (!vma || addr+len <= vma->vm_start)
1283 /* remember the address as a hint for next time */
1284 return (mm->free_area_cache = addr);
1285
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001286 /* remember the largest hole we saw so far */
1287 if (addr + mm->cached_hole_size < vma->vm_start)
1288 mm->cached_hole_size = vma->vm_start - addr;
1289
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 /* try just below the current vma->vm_start */
1291 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001292 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
Chris Wright73219d12005-06-21 17:14:52 -07001294bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 /*
1296 * A failed mmap() very likely causes application failure,
1297 * so fall back to the bottom-up function here. This scenario
1298 * can happen with large stack limits and large mmap()
1299 * allocations.
1300 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001301 mm->cached_hole_size = ~0UL;
1302 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1304 /*
1305 * Restore the topdown base:
1306 */
1307 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001308 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310 return addr;
1311}
1312#endif
1313
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001314void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
1316 /*
1317 * Is this a new hole at the highest possible address?
1318 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001319 if (addr > mm->free_area_cache)
1320 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
1322 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001323 if (mm->free_area_cache > mm->mmap_base)
1324 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325}
1326
1327unsigned long
1328get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1329 unsigned long pgoff, unsigned long flags)
1330{
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001331 unsigned long ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001333 if (!(flags & MAP_FIXED)) {
1334 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1335
1336 get_area = current->mm->get_unmapped_area;
1337 if (file && file->f_op && file->f_op->get_unmapped_area)
1338 get_area = file->f_op->get_unmapped_area;
1339 addr = get_area(file, addr, len, pgoff, flags);
1340 if (IS_ERR_VALUE(addr))
1341 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 }
1343
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001344 if (addr > TASK_SIZE - len)
1345 return -ENOMEM;
1346 if (addr & ~PAGE_MASK)
1347 return -EINVAL;
1348 if (file && is_file_hugepages(file)) {
1349 /*
1350 * Check if the given range is hugepage aligned, and
1351 * can be made suitable for hugepages.
1352 */
1353 ret = prepare_hugepage_range(addr, len);
1354 } else {
1355 /*
1356 * Ensure that a normal request is not falling in a
1357 * reserved hugepage range. For some archs like IA-64,
1358 * there is a separate region for hugepages.
1359 */
1360 ret = is_hugepage_only_range(current->mm, addr, len);
1361 }
1362 if (ret)
1363 return -EINVAL;
1364 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365}
1366
1367EXPORT_SYMBOL(get_unmapped_area);
1368
1369/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1370struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1371{
1372 struct vm_area_struct *vma = NULL;
1373
1374 if (mm) {
1375 /* Check the cache first. */
1376 /* (Cache hit rate is typically around 35%.) */
1377 vma = mm->mmap_cache;
1378 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1379 struct rb_node * rb_node;
1380
1381 rb_node = mm->mm_rb.rb_node;
1382 vma = NULL;
1383
1384 while (rb_node) {
1385 struct vm_area_struct * vma_tmp;
1386
1387 vma_tmp = rb_entry(rb_node,
1388 struct vm_area_struct, vm_rb);
1389
1390 if (vma_tmp->vm_end > addr) {
1391 vma = vma_tmp;
1392 if (vma_tmp->vm_start <= addr)
1393 break;
1394 rb_node = rb_node->rb_left;
1395 } else
1396 rb_node = rb_node->rb_right;
1397 }
1398 if (vma)
1399 mm->mmap_cache = vma;
1400 }
1401 }
1402 return vma;
1403}
1404
1405EXPORT_SYMBOL(find_vma);
1406
1407/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1408struct vm_area_struct *
1409find_vma_prev(struct mm_struct *mm, unsigned long addr,
1410 struct vm_area_struct **pprev)
1411{
1412 struct vm_area_struct *vma = NULL, *prev = NULL;
1413 struct rb_node * rb_node;
1414 if (!mm)
1415 goto out;
1416
1417 /* Guard against addr being lower than the first VMA */
1418 vma = mm->mmap;
1419
1420 /* Go through the RB tree quickly. */
1421 rb_node = mm->mm_rb.rb_node;
1422
1423 while (rb_node) {
1424 struct vm_area_struct *vma_tmp;
1425 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1426
1427 if (addr < vma_tmp->vm_end) {
1428 rb_node = rb_node->rb_left;
1429 } else {
1430 prev = vma_tmp;
1431 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1432 break;
1433 rb_node = rb_node->rb_right;
1434 }
1435 }
1436
1437out:
1438 *pprev = prev;
1439 return prev ? prev->vm_next : vma;
1440}
1441
1442/*
1443 * Verify that the stack growth is acceptable and
1444 * update accounting. This is shared with both the
1445 * grow-up and grow-down cases.
1446 */
1447static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1448{
1449 struct mm_struct *mm = vma->vm_mm;
1450 struct rlimit *rlim = current->signal->rlim;
1451
1452 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001453 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 return -ENOMEM;
1455
1456 /* Stack limit test */
1457 if (size > rlim[RLIMIT_STACK].rlim_cur)
1458 return -ENOMEM;
1459
1460 /* mlock limit tests */
1461 if (vma->vm_flags & VM_LOCKED) {
1462 unsigned long locked;
1463 unsigned long limit;
1464 locked = mm->locked_vm + grow;
1465 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1466 if (locked > limit && !capable(CAP_IPC_LOCK))
1467 return -ENOMEM;
1468 }
1469
1470 /*
1471 * Overcommit.. This must be the final test, as it will
1472 * update security statistics.
1473 */
1474 if (security_vm_enough_memory(grow))
1475 return -ENOMEM;
1476
1477 /* Ok, everything looks good - let it rip */
1478 mm->total_vm += grow;
1479 if (vma->vm_flags & VM_LOCKED)
1480 mm->locked_vm += grow;
1481 __vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1482 return 0;
1483}
1484
1485#ifdef CONFIG_STACK_GROWSUP
1486/*
1487 * vma is the first one with address > vma->vm_end. Have to extend vma.
1488 */
1489int expand_stack(struct vm_area_struct * vma, unsigned long address)
1490{
1491 int error;
1492
1493 if (!(vma->vm_flags & VM_GROWSUP))
1494 return -EFAULT;
1495
1496 /*
1497 * We must make sure the anon_vma is allocated
1498 * so that the anon_vma locking is not a noop.
1499 */
1500 if (unlikely(anon_vma_prepare(vma)))
1501 return -ENOMEM;
1502 anon_vma_lock(vma);
1503
1504 /*
1505 * vma->vm_start/vm_end cannot change under us because the caller
1506 * is required to hold the mmap_sem in read mode. We need the
1507 * anon_vma lock to serialize against concurrent expand_stacks.
1508 */
1509 address += 4 + PAGE_SIZE - 1;
1510 address &= PAGE_MASK;
1511 error = 0;
1512
1513 /* Somebody else might have raced and expanded it already */
1514 if (address > vma->vm_end) {
1515 unsigned long size, grow;
1516
1517 size = address - vma->vm_start;
1518 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1519
1520 error = acct_stack_growth(vma, size, grow);
1521 if (!error)
1522 vma->vm_end = address;
1523 }
1524 anon_vma_unlock(vma);
1525 return error;
1526}
1527
1528struct vm_area_struct *
1529find_extend_vma(struct mm_struct *mm, unsigned long addr)
1530{
1531 struct vm_area_struct *vma, *prev;
1532
1533 addr &= PAGE_MASK;
1534 vma = find_vma_prev(mm, addr, &prev);
1535 if (vma && (vma->vm_start <= addr))
1536 return vma;
1537 if (!prev || expand_stack(prev, addr))
1538 return NULL;
1539 if (prev->vm_flags & VM_LOCKED) {
1540 make_pages_present(addr, prev->vm_end);
1541 }
1542 return prev;
1543}
1544#else
1545/*
1546 * vma is the first one with address < vma->vm_start. Have to extend vma.
1547 */
1548int expand_stack(struct vm_area_struct *vma, unsigned long address)
1549{
1550 int error;
1551
1552 /*
1553 * We must make sure the anon_vma is allocated
1554 * so that the anon_vma locking is not a noop.
1555 */
1556 if (unlikely(anon_vma_prepare(vma)))
1557 return -ENOMEM;
1558 anon_vma_lock(vma);
1559
1560 /*
1561 * vma->vm_start/vm_end cannot change under us because the caller
1562 * is required to hold the mmap_sem in read mode. We need the
1563 * anon_vma lock to serialize against concurrent expand_stacks.
1564 */
1565 address &= PAGE_MASK;
1566 error = 0;
1567
1568 /* Somebody else might have raced and expanded it already */
1569 if (address < vma->vm_start) {
1570 unsigned long size, grow;
1571
1572 size = vma->vm_end - address;
1573 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1574
1575 error = acct_stack_growth(vma, size, grow);
1576 if (!error) {
1577 vma->vm_start = address;
1578 vma->vm_pgoff -= grow;
1579 }
1580 }
1581 anon_vma_unlock(vma);
1582 return error;
1583}
1584
1585struct vm_area_struct *
1586find_extend_vma(struct mm_struct * mm, unsigned long addr)
1587{
1588 struct vm_area_struct * vma;
1589 unsigned long start;
1590
1591 addr &= PAGE_MASK;
1592 vma = find_vma(mm,addr);
1593 if (!vma)
1594 return NULL;
1595 if (vma->vm_start <= addr)
1596 return vma;
1597 if (!(vma->vm_flags & VM_GROWSDOWN))
1598 return NULL;
1599 start = vma->vm_start;
1600 if (expand_stack(vma, addr))
1601 return NULL;
1602 if (vma->vm_flags & VM_LOCKED) {
1603 make_pages_present(addr, start);
1604 }
1605 return vma;
1606}
1607#endif
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609/* Normal function to fix up a mapping
1610 * This function is the default for when an area has no specific
1611 * function. This may be used as part of a more specific routine.
1612 *
1613 * By the time this function is called, the area struct has been
1614 * removed from the process mapping list.
1615 */
1616static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1617{
1618 size_t len = area->vm_end - area->vm_start;
1619
1620 area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1621 if (area->vm_flags & VM_LOCKED)
1622 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1623 vm_stat_unaccount(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 remove_vm_struct(area);
1625}
1626
1627/*
1628 * Update the VMA and inode share lists.
1629 *
1630 * Ok - we have the memory areas we should free on the 'free' list,
1631 * so release them, and do the vma updates.
1632 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001633static void unmap_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634{
1635 do {
Hugh Dickins146425a2005-04-19 13:29:18 -07001636 struct vm_area_struct *next = vma->vm_next;
1637 unmap_vma(mm, vma);
1638 vma = next;
1639 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 validate_mm(mm);
1641}
1642
1643/*
1644 * Get rid of page table information in the indicated region.
1645 *
1646 * Called with the page table lock held.
1647 */
1648static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001649 struct vm_area_struct *vma, struct vm_area_struct *prev,
1650 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001652 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 struct mmu_gather *tlb;
1654 unsigned long nr_accounted = 0;
1655
1656 lru_add_drain();
Hugh Dickinse0da3822005-04-19 13:29:15 -07001657 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 tlb = tlb_gather_mmu(mm, 0);
1659 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1660 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001661 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001662 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 tlb_finish_mmu(tlb, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001664 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665}
1666
1667/*
1668 * Create a list of vma's touched by the unmap, removing them from the mm's
1669 * vma list as we go..
1670 */
1671static void
1672detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1673 struct vm_area_struct *prev, unsigned long end)
1674{
1675 struct vm_area_struct **insertion_point;
1676 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001677 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
1679 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1680 do {
1681 rb_erase(&vma->vm_rb, &mm->mm_rb);
1682 mm->map_count--;
1683 tail_vma = vma;
1684 vma = vma->vm_next;
1685 } while (vma && vma->vm_start < end);
1686 *insertion_point = vma;
1687 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001688 if (mm->unmap_area == arch_unmap_area)
1689 addr = prev ? prev->vm_end : mm->mmap_base;
1690 else
1691 addr = vma ? vma->vm_start : mm->mmap_base;
1692 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 mm->mmap_cache = NULL; /* Kill the cache. */
1694}
1695
1696/*
1697 * Split a vma into two pieces at address 'addr', a new vma is allocated
1698 * either for the first part or the the tail.
1699 */
1700int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1701 unsigned long addr, int new_below)
1702{
1703 struct mempolicy *pol;
1704 struct vm_area_struct *new;
1705
1706 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1707 return -EINVAL;
1708
1709 if (mm->map_count >= sysctl_max_map_count)
1710 return -ENOMEM;
1711
1712 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1713 if (!new)
1714 return -ENOMEM;
1715
1716 /* most fields are the same, copy all, and then fixup */
1717 *new = *vma;
1718
1719 if (new_below)
1720 new->vm_end = addr;
1721 else {
1722 new->vm_start = addr;
1723 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1724 }
1725
1726 pol = mpol_copy(vma_policy(vma));
1727 if (IS_ERR(pol)) {
1728 kmem_cache_free(vm_area_cachep, new);
1729 return PTR_ERR(pol);
1730 }
1731 vma_set_policy(new, pol);
1732
1733 if (new->vm_file)
1734 get_file(new->vm_file);
1735
1736 if (new->vm_ops && new->vm_ops->open)
1737 new->vm_ops->open(new);
1738
1739 if (new_below)
1740 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1741 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1742 else
1743 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1744
1745 return 0;
1746}
1747
1748/* Munmap is split into 2 main parts -- this part which finds
1749 * what needs doing, and the areas themselves, which do the
1750 * work. This now handles partial unmappings.
1751 * Jeremy Fitzhardinge <jeremy@goop.org>
1752 */
1753int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1754{
1755 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001756 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
1758 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1759 return -EINVAL;
1760
1761 if ((len = PAGE_ALIGN(len)) == 0)
1762 return -EINVAL;
1763
1764 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001765 vma = find_vma_prev(mm, start, &prev);
1766 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001768 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 /* if it doesn't overlap, we have nothing.. */
1771 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001772 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 return 0;
1774
1775 /*
1776 * If we need to split any vma, do it now to save pain later.
1777 *
1778 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1779 * unmapped vm_area_struct will remain in use: so lower split_vma
1780 * places tmp vma above, and higher split_vma places tmp vma below.
1781 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001782 if (start > vma->vm_start) {
1783 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 if (error)
1785 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001786 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 }
1788
1789 /* Does it split the last one? */
1790 last = find_vma(mm, end);
1791 if (last && end > last->vm_start) {
1792 int error = split_vma(mm, last, end, 1);
1793 if (error)
1794 return error;
1795 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001796 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
1798 /*
1799 * Remove the vma's, and unmap the actual pages
1800 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001801 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1802 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 /* Fix up all other VM information */
Hugh Dickins146425a2005-04-19 13:29:18 -07001805 unmap_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807 return 0;
1808}
1809
1810EXPORT_SYMBOL(do_munmap);
1811
1812asmlinkage long sys_munmap(unsigned long addr, size_t len)
1813{
1814 int ret;
1815 struct mm_struct *mm = current->mm;
1816
1817 profile_munmap(addr);
1818
1819 down_write(&mm->mmap_sem);
1820 ret = do_munmap(mm, addr, len);
1821 up_write(&mm->mmap_sem);
1822 return ret;
1823}
1824
1825static inline void verify_mm_writelocked(struct mm_struct *mm)
1826{
1827#ifdef CONFIG_DEBUG_KERNEL
1828 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1829 WARN_ON(1);
1830 up_read(&mm->mmap_sem);
1831 }
1832#endif
1833}
1834
1835/*
1836 * this is really a simplified "do_mmap". it only handles
1837 * anonymous maps. eventually we may be able to do some
1838 * brk-specific accounting here.
1839 */
1840unsigned long do_brk(unsigned long addr, unsigned long len)
1841{
1842 struct mm_struct * mm = current->mm;
1843 struct vm_area_struct * vma, * prev;
1844 unsigned long flags;
1845 struct rb_node ** rb_link, * rb_parent;
1846 pgoff_t pgoff = addr >> PAGE_SHIFT;
1847
1848 len = PAGE_ALIGN(len);
1849 if (!len)
1850 return addr;
1851
1852 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1853 return -EINVAL;
1854
1855 /*
1856 * mlock MCL_FUTURE?
1857 */
1858 if (mm->def_flags & VM_LOCKED) {
1859 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001860 locked = len >> PAGE_SHIFT;
1861 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07001863 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1865 return -EAGAIN;
1866 }
1867
1868 /*
1869 * mm->mmap_sem is required to protect against another thread
1870 * changing the mappings in case we sleep.
1871 */
1872 verify_mm_writelocked(mm);
1873
1874 /*
1875 * Clear old maps. this also does some error checking for us
1876 */
1877 munmap_back:
1878 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1879 if (vma && vma->vm_start < addr + len) {
1880 if (do_munmap(mm, addr, len))
1881 return -ENOMEM;
1882 goto munmap_back;
1883 }
1884
1885 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001886 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 return -ENOMEM;
1888
1889 if (mm->map_count > sysctl_max_map_count)
1890 return -ENOMEM;
1891
1892 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1893 return -ENOMEM;
1894
1895 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1896
1897 /* Can we just expand an old private anonymous mapping? */
1898 if (vma_merge(mm, prev, addr, addr + len, flags,
1899 NULL, NULL, pgoff, NULL))
1900 goto out;
1901
1902 /*
1903 * create a vma struct for an anonymous mapping
1904 */
1905 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1906 if (!vma) {
1907 vm_unacct_memory(len >> PAGE_SHIFT);
1908 return -ENOMEM;
1909 }
1910 memset(vma, 0, sizeof(*vma));
1911
1912 vma->vm_mm = mm;
1913 vma->vm_start = addr;
1914 vma->vm_end = addr + len;
1915 vma->vm_pgoff = pgoff;
1916 vma->vm_flags = flags;
1917 vma->vm_page_prot = protection_map[flags & 0x0f];
1918 vma_link(mm, vma, prev, rb_link, rb_parent);
1919out:
1920 mm->total_vm += len >> PAGE_SHIFT;
1921 if (flags & VM_LOCKED) {
1922 mm->locked_vm += len >> PAGE_SHIFT;
1923 make_pages_present(addr, addr + len);
1924 }
1925 return addr;
1926}
1927
1928EXPORT_SYMBOL(do_brk);
1929
1930/* Release all mmaps. */
1931void exit_mmap(struct mm_struct *mm)
1932{
1933 struct mmu_gather *tlb;
Hugh Dickinse0da3822005-04-19 13:29:15 -07001934 struct vm_area_struct *vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001936 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
1938 lru_add_drain();
1939
1940 spin_lock(&mm->page_table_lock);
1941
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07001943 tlb = tlb_gather_mmu(mm, 1);
1944 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001945 end = unmap_vmas(&tlb, mm, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 vm_unacct_memory(nr_accounted);
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001947 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001948 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 mm->mmap = mm->mmap_cache = NULL;
1951 mm->mm_rb = RB_ROOT;
1952 set_mm_counter(mm, rss, 0);
1953 mm->total_vm = 0;
1954 mm->locked_vm = 0;
1955
1956 spin_unlock(&mm->page_table_lock);
1957
1958 /*
1959 * Walk the list again, actually closing and freeing it
1960 * without holding any MM locks.
1961 */
1962 while (vma) {
1963 struct vm_area_struct *next = vma->vm_next;
1964 remove_vm_struct(vma);
1965 vma = next;
1966 }
Hugh Dickinse0da3822005-04-19 13:29:15 -07001967
Hugh Dickinse2cdef82005-04-19 13:29:19 -07001968 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969}
1970
1971/* Insert vm structure into process list sorted by address
1972 * and into the inode's i_mmap tree. If vm_file is non-NULL
1973 * then i_mmap_lock is taken here.
1974 */
1975int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1976{
1977 struct vm_area_struct * __vma, * prev;
1978 struct rb_node ** rb_link, * rb_parent;
1979
1980 /*
1981 * The vm_pgoff of a purely anonymous vma should be irrelevant
1982 * until its first write fault, when page's anon_vma and index
1983 * are set. But now set the vm_pgoff it will almost certainly
1984 * end up with (unless mremap moves it elsewhere before that
1985 * first wfault), so /proc/pid/maps tells a consistent story.
1986 *
1987 * By setting it to reflect the virtual start address of the
1988 * vma, merges and splits can happen in a seamless way, just
1989 * using the existing file pgoff checks and manipulations.
1990 * Similarly in do_mmap_pgoff and in do_brk.
1991 */
1992 if (!vma->vm_file) {
1993 BUG_ON(vma->anon_vma);
1994 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1995 }
1996 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1997 if (__vma && __vma->vm_start < vma->vm_end)
1998 return -ENOMEM;
1999 vma_link(mm, vma, prev, rb_link, rb_parent);
2000 return 0;
2001}
2002
2003/*
2004 * Copy the vma structure to a new location in the same mm,
2005 * prior to moving page table entries, to effect an mremap move.
2006 */
2007struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2008 unsigned long addr, unsigned long len, pgoff_t pgoff)
2009{
2010 struct vm_area_struct *vma = *vmap;
2011 unsigned long vma_start = vma->vm_start;
2012 struct mm_struct *mm = vma->vm_mm;
2013 struct vm_area_struct *new_vma, *prev;
2014 struct rb_node **rb_link, *rb_parent;
2015 struct mempolicy *pol;
2016
2017 /*
2018 * If anonymous vma has not yet been faulted, update new pgoff
2019 * to match new location, to increase its chance of merging.
2020 */
2021 if (!vma->vm_file && !vma->anon_vma)
2022 pgoff = addr >> PAGE_SHIFT;
2023
2024 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2025 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2026 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2027 if (new_vma) {
2028 /*
2029 * Source vma may have been merged into new_vma
2030 */
2031 if (vma_start >= new_vma->vm_start &&
2032 vma_start < new_vma->vm_end)
2033 *vmap = new_vma;
2034 } else {
2035 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2036 if (new_vma) {
2037 *new_vma = *vma;
2038 pol = mpol_copy(vma_policy(vma));
2039 if (IS_ERR(pol)) {
2040 kmem_cache_free(vm_area_cachep, new_vma);
2041 return NULL;
2042 }
2043 vma_set_policy(new_vma, pol);
2044 new_vma->vm_start = addr;
2045 new_vma->vm_end = addr + len;
2046 new_vma->vm_pgoff = pgoff;
2047 if (new_vma->vm_file)
2048 get_file(new_vma->vm_file);
2049 if (new_vma->vm_ops && new_vma->vm_ops->open)
2050 new_vma->vm_ops->open(new_vma);
2051 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2052 }
2053 }
2054 return new_vma;
2055}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002056
2057/*
2058 * Return true if the calling process may expand its vm space by the passed
2059 * number of pages
2060 */
2061int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2062{
2063 unsigned long cur = mm->total_vm; /* pages */
2064 unsigned long lim;
2065
2066 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2067
2068 if (cur + npages > lim)
2069 return 0;
2070 return 1;
2071}