1 /*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
7 */
8
9 #include <linux/slab.h>
10 #include <linux/backing-dev.h>
11 #include <linux/mm.h>
12 #include <linux/shm.h>
13 #include <linux/mman.h>
14 #include <linux/pagemap.h>
15 #include <linux/swap.h>
16 #include <linux/syscalls.h>
17 #include <linux/capability.h>
18 #include <linux/init.h>
19 #include <linux/file.h>
20 #include <linux/fs.h>
21 #include <linux/personality.h>
22 #include <linux/security.h>
23 #include <linux/ima.h>
24 #include <linux/hugetlb.h>
25 #include <linux/profile.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/mempolicy.h>
29 #include <linux/rmap.h>
30 #include <linux/mmu_notifier.h>
31 #include <linux/perf_event.h>
32
33 #include <asm/uaccess.h>
34 #include <asm/cacheflush.h>
35 #include <asm/tlb.h>
36 #include <asm/mmu_context.h>
37
38 #include "internal.h"
39
40 #ifndef arch_mmap_check
41 #define arch_mmap_check(addr, len, flags) (0)
42 #endif
43
44 #ifndef arch_rebalance_pgtables
45 #define arch_rebalance_pgtables(addr, len) (addr)
46 #endif
47
48 static void unmap_region(struct mm_struct *mm,
49 struct vm_area_struct *vma, struct vm_area_struct *prev,
50 unsigned long start, unsigned long end);
51
52 /*
53 * WARNING: the debugging will use recursive algorithms so never enable this
54 * unless you know what you are doing.
55 */
56 #undef DEBUG_MM_RB
57
58 /* description of effects of mapping type and prot in current implementation.
59 * this is due to the limited x86 page protection hardware. The expected
60 * behavior is in parens:
61 *
62 * map_type prot
63 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
64 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
65 * w: (no) no w: (no) no w: (yes) yes w: (no) no
66 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
67 *
68 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
69 * w: (no) no w: (no) no w: (copy) copy w: (no) no
70 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
71 *
72 */
73 pgprot_t protection_map[16] = {
74 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
75 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
76 };
77
78 pgprot_t vm_get_page_prot(unsigned long vm_flags)
79 {
80 return __pgprot(pgprot_val(protection_map[vm_flags &
81 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
82 pgprot_val(arch_vm_get_page_prot(vm_flags)));
83 }
84 EXPORT_SYMBOL(vm_get_page_prot);
85
86 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
87 int sysctl_overcommit_ratio = 50; /* default is 50% */
88 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
89 struct percpu_counter vm_committed_as;
90
91 /*
92 * Check that a process has enough memory to allocate a new virtual
93 * mapping. 0 means there is enough memory for the allocation to
94 * succeed and -ENOMEM implies there is not.
95 *
96 * We currently support three overcommit policies, which are set via the
97 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
98 *
99 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
100 * Additional code 2002 Jul 20 by Robert Love.
101 *
102 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
103 *
104 * Note this is a helper function intended to be used by LSMs which
105 * wish to use this logic.
106 */
107 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
108 {
109 unsigned long free, allowed;
110
111 vm_acct_memory(pages);
112
113 /*
114 * Sometimes we want to use more memory than we have
115 */
116 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
117 return 0;
118
119 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
120 unsigned long n;
121
122 free = global_page_state(NR_FILE_PAGES);
123 free += nr_swap_pages;
124
125 /*
126 * Any slabs which are created with the
127 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
128 * which are reclaimable, under pressure. The dentry
129 * cache and most inode caches should fall into this
130 */
131 free += global_page_state(NR_SLAB_RECLAIMABLE);
132
133 /*
134 * Leave the last 3% for root
135 */
136 if (!cap_sys_admin)
137 free -= free / 32;
138
139 if (free > pages)
140 return 0;
141
142 /*
143 * nr_free_pages() is very expensive on large systems,
144 * only call if we're about to fail.
145 */
146 n = nr_free_pages();
147
148 /*
149 * Leave reserved pages. The pages are not for anonymous pages.
150 */
151 if (n <= totalreserve_pages)
152 goto error;
153 else
154 n -= totalreserve_pages;
155
156 /*
157 * Leave the last 3% for root
158 */
159 if (!cap_sys_admin)
160 n -= n / 32;
161 free += n;
162
163 if (free > pages)
164 return 0;
165
166 goto error;
167 }
168
169 allowed = (totalram_pages - hugetlb_total_pages())
170 * sysctl_overcommit_ratio / 100;
171 /*
172 * Leave the last 3% for root
173 */
174 if (!cap_sys_admin)
175 allowed -= allowed / 32;
176 allowed += total_swap_pages;
177
178 /* Don't let a single process grow too big:
179 leave 3% of the size of this process for other processes */
180 if (mm)
181 allowed -= mm->total_vm / 32;
182
183 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
184 return 0;
185 error:
186 vm_unacct_memory(pages);
187
188 return -ENOMEM;
189 }
190
191 /*
192 * Requires inode->i_mapping->i_mmap_lock
193 */
194 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
195 struct file *file, struct address_space *mapping)
196 {
197 if (vma->vm_flags & VM_DENYWRITE)
198 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
199 if (vma->vm_flags & VM_SHARED)
200 mapping->i_mmap_writable--;
201
202 flush_dcache_mmap_lock(mapping);
203 if (unlikely(vma->vm_flags & VM_NONLINEAR))
204 list_del_init(&vma->shared.vm_set.list);
205 else
206 vma_prio_tree_remove(vma, &mapping->i_mmap);
207 flush_dcache_mmap_unlock(mapping);
208 }
209
210 /*
211 * Unlink a file-based vm structure from its prio_tree, to hide
212 * vma from rmap and vmtruncate before freeing its page tables.
213 */
214 void unlink_file_vma(struct vm_area_struct *vma)
215 {
216 struct file *file = vma->vm_file;
217
218 if (file) {
219 struct address_space *mapping = file->f_mapping;
220 spin_lock(&mapping->i_mmap_lock);
221 __remove_shared_vm_struct(vma, file, mapping);
222 spin_unlock(&mapping->i_mmap_lock);
223 }
224 }
225
226 /*
227 * Close a vm structure and free it, returning the next.
228 */
229 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
230 {
231 struct vm_area_struct *next = vma->vm_next;
232
233 might_sleep();
234 if (vma->vm_ops && vma->vm_ops->close)
235 vma->vm_ops->close(vma);
236 if (vma->vm_file) {
237 fput(vma->vm_file);
238 if (vma->vm_flags & VM_EXECUTABLE)
239 removed_exe_file_vma(vma->vm_mm);
240 }
241 mpol_put(vma_policy(vma));
242 kmem_cache_free(vm_area_cachep, vma);
243 return next;
244 }
245
246 SYSCALL_DEFINE1(brk, unsigned long, brk)
247 {
248 unsigned long rlim, retval;
249 unsigned long newbrk, oldbrk;
250 struct mm_struct *mm = current->mm;
251 unsigned long min_brk;
252
253 down_write(&mm->mmap_sem);
254
255 #ifdef CONFIG_COMPAT_BRK
256 min_brk = mm->end_code;
257 #else
258 min_brk = mm->start_brk;
259 #endif
260 if (brk < min_brk)
261 goto out;
262
263 /*
264 * Check against rlimit here. If this check is done later after the test
265 * of oldbrk with newbrk then it can escape the test and let the data
266 * segment grow beyond its set limit the in case where the limit is
267 * not page aligned -Ram Gupta
268 */
269 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
270 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
271 (mm->end_data - mm->start_data) > rlim)
272 goto out;
273
274 newbrk = PAGE_ALIGN(brk);
275 oldbrk = PAGE_ALIGN(mm->brk);
276 if (oldbrk == newbrk)
277 goto set_brk;
278
279 /* Always allow shrinking brk. */
280 if (brk <= mm->brk) {
281 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
282 goto set_brk;
283 goto out;
284 }
285
286 /* Check against existing mmap mappings. */
287 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
288 goto out;
289
290 /* Ok, looks good - let it rip. */
291 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
292 goto out;
293 set_brk:
294 mm->brk = brk;
295 out:
296 retval = mm->brk;
297 up_write(&mm->mmap_sem);
298 return retval;
299 }
300
301 #ifdef DEBUG_MM_RB
302 static int browse_rb(struct rb_root *root)
303 {
304 int i = 0, j;
305 struct rb_node *nd, *pn = NULL;
306 unsigned long prev = 0, pend = 0;
307
308 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
309 struct vm_area_struct *vma;
310 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
311 if (vma->vm_start < prev)
312 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
313 if (vma->vm_start < pend)
314 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
315 if (vma->vm_start > vma->vm_end)
316 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
317 i++;
318 pn = nd;
319 prev = vma->vm_start;
320 pend = vma->vm_end;
321 }
322 j = 0;
323 for (nd = pn; nd; nd = rb_prev(nd)) {
324 j++;
325 }
326 if (i != j)
327 printk("backwards %d, forwards %d\n", j, i), i = 0;
328 return i;
329 }
330
331 void validate_mm(struct mm_struct *mm)
332 {
333 int bug = 0;
334 int i = 0;
335 struct vm_area_struct *tmp = mm->mmap;
336 while (tmp) {
337 tmp = tmp->vm_next;
338 i++;
339 }
340 if (i != mm->map_count)
341 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
342 i = browse_rb(&mm->mm_rb);
343 if (i != mm->map_count)
344 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
345 BUG_ON(bug);
346 }
347 #else
348 #define validate_mm(mm) do { } while (0)
349 #endif
350
351 static struct vm_area_struct *
352 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
353 struct vm_area_struct **pprev, struct rb_node ***rb_link,
354 struct rb_node ** rb_parent)
355 {
356 struct vm_area_struct * vma;
357 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
358
359 __rb_link = &mm->mm_rb.rb_node;
360 rb_prev = __rb_parent = NULL;
361 vma = NULL;
362
363 while (*__rb_link) {
364 struct vm_area_struct *vma_tmp;
365
366 __rb_parent = *__rb_link;
367 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
368
369 if (vma_tmp->vm_end > addr) {
370 vma = vma_tmp;
371 if (vma_tmp->vm_start <= addr)
372 break;
373 __rb_link = &__rb_parent->rb_left;
374 } else {
375 rb_prev = __rb_parent;
376 __rb_link = &__rb_parent->rb_right;
377 }
378 }
379
380 *pprev = NULL;
381 if (rb_prev)
382 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
383 *rb_link = __rb_link;
384 *rb_parent = __rb_parent;
385 return vma;
386 }
387
388 static inline void
389 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
390 struct vm_area_struct *prev, struct rb_node *rb_parent)
391 {
392 struct vm_area_struct *next;
393
394 vma->vm_prev = prev;
395 if (prev) {
396 next = prev->vm_next;
397 prev->vm_next = vma;
398 } else {
399 mm->mmap = vma;
400 if (rb_parent)
401 next = rb_entry(rb_parent,
402 struct vm_area_struct, vm_rb);
403 else
404 next = NULL;
405 }
406 vma->vm_next = next;
407 if (next)
408 next->vm_prev = vma;
409 }
410
411 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
412 struct rb_node **rb_link, struct rb_node *rb_parent)
413 {
414 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
415 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
416 }
417
418 static void __vma_link_file(struct vm_area_struct *vma)
419 {
420 struct file *file;
421
422 file = vma->vm_file;
423 if (file) {
424 struct address_space *mapping = file->f_mapping;
425
426 if (vma->vm_flags & VM_DENYWRITE)
427 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
428 if (vma->vm_flags & VM_SHARED)
429 mapping->i_mmap_writable++;
430
431 flush_dcache_mmap_lock(mapping);
432 if (unlikely(vma->vm_flags & VM_NONLINEAR))
433 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
434 else
435 vma_prio_tree_insert(vma, &mapping->i_mmap);
436 flush_dcache_mmap_unlock(mapping);
437 }
438 }
439
440 static void
441 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
442 struct vm_area_struct *prev, struct rb_node **rb_link,
443 struct rb_node *rb_parent)
444 {
445 __vma_link_list(mm, vma, prev, rb_parent);
446 __vma_link_rb(mm, vma, rb_link, rb_parent);
447 __anon_vma_link(vma);
448 }
449
450 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
451 struct vm_area_struct *prev, struct rb_node **rb_link,
452 struct rb_node *rb_parent)
453 {
454 struct address_space *mapping = NULL;
455
456 if (vma->vm_file)
457 mapping = vma->vm_file->f_mapping;
458
459 if (mapping) {
460 spin_lock(&mapping->i_mmap_lock);
461 vma->vm_truncate_count = mapping->truncate_count;
462 }
463 anon_vma_lock(vma);
464
465 __vma_link(mm, vma, prev, rb_link, rb_parent);
466 __vma_link_file(vma);
467
468 anon_vma_unlock(vma);
469 if (mapping)
470 spin_unlock(&mapping->i_mmap_lock);
471
472 mm->map_count++;
473 validate_mm(mm);
474 }
475
476 /*
477 * Helper for vma_adjust in the split_vma insert case:
478 * insert vm structure into list and rbtree and anon_vma,
479 * but it has already been inserted into prio_tree earlier.
480 */
481 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
482 {
483 struct vm_area_struct *__vma, *prev;
484 struct rb_node **rb_link, *rb_parent;
485
486 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
487 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
488 __vma_link(mm, vma, prev, rb_link, rb_parent);
489 mm->map_count++;
490 }
491
492 static inline void
493 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
494 struct vm_area_struct *prev)
495 {
496 struct vm_area_struct *next = vma->vm_next;
497
498 prev->vm_next = next;
499 if (next)
500 next->vm_prev = prev;
501 rb_erase(&vma->vm_rb, &mm->mm_rb);
502 if (mm->mmap_cache == vma)
503 mm->mmap_cache = prev;
504 }
505
506 /*
507 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
508 * is already present in an i_mmap tree without adjusting the tree.
509 * The following helper function should be used when such adjustments
510 * are necessary. The "insert" vma (if any) is to be inserted
511 * before we drop the necessary locks.
512 */
513 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
514 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
515 {
516 struct mm_struct *mm = vma->vm_mm;
517 struct vm_area_struct *next = vma->vm_next;
518 struct vm_area_struct *importer = NULL;
519 struct address_space *mapping = NULL;
520 struct prio_tree_root *root = NULL;
521 struct file *file = vma->vm_file;
522 struct anon_vma *anon_vma = NULL;
523 long adjust_next = 0;
524 int remove_next = 0;
525
526 if (next && !insert) {
527 if (end >= next->vm_end) {
528 /*
529 * vma expands, overlapping all the next, and
530 * perhaps the one after too (mprotect case 6).
531 */
532 again: remove_next = 1 + (end > next->vm_end);
533 end = next->vm_end;
534 anon_vma = next->anon_vma;
535 importer = vma;
536 } else if (end > next->vm_start) {
537 /*
538 * vma expands, overlapping part of the next:
539 * mprotect case 5 shifting the boundary up.
540 */
541 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
542 anon_vma = next->anon_vma;
543 importer = vma;
544 } else if (end < vma->vm_end) {
545 /*
546 * vma shrinks, and !insert tells it's not
547 * split_vma inserting another: so it must be
548 * mprotect case 4 shifting the boundary down.
549 */
550 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
551 anon_vma = next->anon_vma;
552 importer = next;
553 }
554 }
555
556 if (file) {
557 mapping = file->f_mapping;
558 if (!(vma->vm_flags & VM_NONLINEAR))
559 root = &mapping->i_mmap;
560 spin_lock(&mapping->i_mmap_lock);
561 if (importer &&
562 vma->vm_truncate_count != next->vm_truncate_count) {
563 /*
564 * unmap_mapping_range might be in progress:
565 * ensure that the expanding vma is rescanned.
566 */
567 importer->vm_truncate_count = 0;
568 }
569 if (insert) {
570 insert->vm_truncate_count = vma->vm_truncate_count;
571 /*
572 * Put into prio_tree now, so instantiated pages
573 * are visible to arm/parisc __flush_dcache_page
574 * throughout; but we cannot insert into address
575 * space until vma start or end is updated.
576 */
577 __vma_link_file(insert);
578 }
579 }
580
581 /*
582 * When changing only vma->vm_end, we don't really need
583 * anon_vma lock.
584 */
585 if (vma->anon_vma && (insert || importer || start != vma->vm_start))
586 anon_vma = vma->anon_vma;
587 if (anon_vma) {
588 spin_lock(&anon_vma->lock);
589 /*
590 * Easily overlooked: when mprotect shifts the boundary,
591 * make sure the expanding vma has anon_vma set if the
592 * shrinking vma had, to cover any anon pages imported.
593 */
594 if (importer && !importer->anon_vma) {
595 importer->anon_vma = anon_vma;
596 __anon_vma_link(importer);
597 }
598 }
599
600 if (root) {
601 flush_dcache_mmap_lock(mapping);
602 vma_prio_tree_remove(vma, root);
603 if (adjust_next)
604 vma_prio_tree_remove(next, root);
605 }
606
607 vma->vm_start = start;
608 vma->vm_end = end;
609 vma->vm_pgoff = pgoff;
610 if (adjust_next) {
611 next->vm_start += adjust_next << PAGE_SHIFT;
612 next->vm_pgoff += adjust_next;
613 }
614
615 if (root) {
616 if (adjust_next)
617 vma_prio_tree_insert(next, root);
618 vma_prio_tree_insert(vma, root);
619 flush_dcache_mmap_unlock(mapping);
620 }
621
622 if (remove_next) {
623 /*
624 * vma_merge has merged next into vma, and needs
625 * us to remove next before dropping the locks.
626 */
627 __vma_unlink(mm, next, vma);
628 if (file)
629 __remove_shared_vm_struct(next, file, mapping);
630 if (next->anon_vma)
631 __anon_vma_merge(vma, next);
632 } else if (insert) {
633 /*
634 * split_vma has split insert from vma, and needs
635 * us to insert it before dropping the locks
636 * (it may either follow vma or precede it).
637 */
638 __insert_vm_struct(mm, insert);
639 }
640
641 if (anon_vma)
642 spin_unlock(&anon_vma->lock);
643 if (mapping)
644 spin_unlock(&mapping->i_mmap_lock);
645
646 if (remove_next) {
647 if (file) {
648 fput(file);
649 if (next->vm_flags & VM_EXECUTABLE)
650 removed_exe_file_vma(mm);
651 }
652 mm->map_count--;
653 mpol_put(vma_policy(next));
654 kmem_cache_free(vm_area_cachep, next);
655 /*
656 * In mprotect's case 6 (see comments on vma_merge),
657 * we must remove another next too. It would clutter
658 * up the code too much to do both in one go.
659 */
660 if (remove_next == 2) {
661 next = vma->vm_next;
662 goto again;
663 }
664 }
665
666 validate_mm(mm);
667 }
668
669 /*
670 * If the vma has a ->close operation then the driver probably needs to release
671 * per-vma resources, so we don't attempt to merge those.
672 */
673 static inline int is_mergeable_vma(struct vm_area_struct *vma,
674 struct file *file, unsigned long vm_flags)
675 {
676 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
677 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
678 return 0;
679 if (vma->vm_file != file)
680 return 0;
681 if (vma->vm_ops && vma->vm_ops->close)
682 return 0;
683 return 1;
684 }
685
686 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
687 struct anon_vma *anon_vma2)
688 {
689 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
690 }
691
692 /*
693 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
694 * in front of (at a lower virtual address and file offset than) the vma.
695 *
696 * We cannot merge two vmas if they have differently assigned (non-NULL)
697 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
698 *
699 * We don't check here for the merged mmap wrapping around the end of pagecache
700 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
701 * wrap, nor mmaps which cover the final page at index -1UL.
702 */
703 static int
704 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
705 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
706 {
707 if (is_mergeable_vma(vma, file, vm_flags) &&
708 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
709 if (vma->vm_pgoff == vm_pgoff)
710 return 1;
711 }
712 return 0;
713 }
714
715 /*
716 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
717 * beyond (at a higher virtual address and file offset than) the vma.
718 *
719 * We cannot merge two vmas if they have differently assigned (non-NULL)
720 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
721 */
722 static int
723 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
724 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
725 {
726 if (is_mergeable_vma(vma, file, vm_flags) &&
727 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
728 pgoff_t vm_pglen;
729 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
730 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
731 return 1;
732 }
733 return 0;
734 }
735
736 /*
737 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
738 * whether that can be merged with its predecessor or its successor.
739 * Or both (it neatly fills a hole).
740 *
741 * In most cases - when called for mmap, brk or mremap - [addr,end) is
742 * certain not to be mapped by the time vma_merge is called; but when
743 * called for mprotect, it is certain to be already mapped (either at
744 * an offset within prev, or at the start of next), and the flags of
745 * this area are about to be changed to vm_flags - and the no-change
746 * case has already been eliminated.
747 *
748 * The following mprotect cases have to be considered, where AAAA is
749 * the area passed down from mprotect_fixup, never extending beyond one
750 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
751 *
752 * AAAA AAAA AAAA AAAA
753 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
754 * cannot merge might become might become might become
755 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
756 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
757 * mremap move: PPPPNNNNNNNN 8
758 * AAAA
759 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
760 * might become case 1 below case 2 below case 3 below
761 *
762 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
763 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
764 */
765 struct vm_area_struct *vma_merge(struct mm_struct *mm,
766 struct vm_area_struct *prev, unsigned long addr,
767 unsigned long end, unsigned long vm_flags,
768 struct anon_vma *anon_vma, struct file *file,
769 pgoff_t pgoff, struct mempolicy *policy)
770 {
771 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
772 struct vm_area_struct *area, *next;
773
774 /*
775 * We later require that vma->vm_flags == vm_flags,
776 * so this tests vma->vm_flags & VM_SPECIAL, too.
777 */
778 if (vm_flags & VM_SPECIAL)
779 return NULL;
780
781 if (prev)
782 next = prev->vm_next;
783 else
784 next = mm->mmap;
785 area = next;
786 if (next && next->vm_end == end) /* cases 6, 7, 8 */
787 next = next->vm_next;
788
789 /*
790 * Can it merge with the predecessor?
791 */
792 if (prev && prev->vm_end == addr &&
793 mpol_equal(vma_policy(prev), policy) &&
794 can_vma_merge_after(prev, vm_flags,
795 anon_vma, file, pgoff)) {
796 /*
797 * OK, it can. Can we now merge in the successor as well?
798 */
799 if (next && end == next->vm_start &&
800 mpol_equal(policy, vma_policy(next)) &&
801 can_vma_merge_before(next, vm_flags,
802 anon_vma, file, pgoff+pglen) &&
803 is_mergeable_anon_vma(prev->anon_vma,
804 next->anon_vma)) {
805 /* cases 1, 6 */
806 vma_adjust(prev, prev->vm_start,
807 next->vm_end, prev->vm_pgoff, NULL);
808 } else /* cases 2, 5, 7 */
809 vma_adjust(prev, prev->vm_start,
810 end, prev->vm_pgoff, NULL);
811 return prev;
812 }
813
814 /*
815 * Can this new request be merged in front of next?
816 */
817 if (next && end == next->vm_start &&
818 mpol_equal(policy, vma_policy(next)) &&
819 can_vma_merge_before(next, vm_flags,
820 anon_vma, file, pgoff+pglen)) {
821 if (prev && addr < prev->vm_end) /* case 4 */
822 vma_adjust(prev, prev->vm_start,
823 addr, prev->vm_pgoff, NULL);
824 else /* cases 3, 8 */
825 vma_adjust(area, addr, next->vm_end,
826 next->vm_pgoff - pglen, NULL);
827 return area;
828 }
829
830 return NULL;
831 }
832
833 /*
834 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
835 * neighbouring vmas for a suitable anon_vma, before it goes off
836 * to allocate a new anon_vma. It checks because a repetitive
837 * sequence of mprotects and faults may otherwise lead to distinct
838 * anon_vmas being allocated, preventing vma merge in subsequent
839 * mprotect.
840 */
841 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
842 {
843 struct vm_area_struct *near;
844 unsigned long vm_flags;
845
846 near = vma->vm_next;
847 if (!near)
848 goto try_prev;
849
850 /*
851 * Since only mprotect tries to remerge vmas, match flags
852 * which might be mprotected into each other later on.
853 * Neither mlock nor madvise tries to remerge at present,
854 * so leave their flags as obstructing a merge.
855 */
856 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
857 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
858
859 if (near->anon_vma && vma->vm_end == near->vm_start &&
860 mpol_equal(vma_policy(vma), vma_policy(near)) &&
861 can_vma_merge_before(near, vm_flags,
862 NULL, vma->vm_file, vma->vm_pgoff +
863 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
864 return near->anon_vma;
865 try_prev:
866 /*
867 * It is potentially slow to have to call find_vma_prev here.
868 * But it's only on the first write fault on the vma, not
869 * every time, and we could devise a way to avoid it later
870 * (e.g. stash info in next's anon_vma_node when assigning
871 * an anon_vma, or when trying vma_merge). Another time.
872 */
873 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
874 if (!near)
875 goto none;
876
877 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
878 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
879
880 if (near->anon_vma && near->vm_end == vma->vm_start &&
881 mpol_equal(vma_policy(near), vma_policy(vma)) &&
882 can_vma_merge_after(near, vm_flags,
883 NULL, vma->vm_file, vma->vm_pgoff))
884 return near->anon_vma;
885 none:
886 /*
887 * There's no absolute need to look only at touching neighbours:
888 * we could search further afield for "compatible" anon_vmas.
889 * But it would probably just be a waste of time searching,
890 * or lead to too many vmas hanging off the same anon_vma.
891 * We're trying to allow mprotect remerging later on,
892 * not trying to minimize memory used for anon_vmas.
893 */
894 return NULL;
895 }
896
897 #ifdef CONFIG_PROC_FS
898 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
899 struct file *file, long pages)
900 {
901 const unsigned long stack_flags
902 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
903
904 if (file) {
905 mm->shared_vm += pages;
906 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
907 mm->exec_vm += pages;
908 } else if (flags & stack_flags)
909 mm->stack_vm += pages;
910 if (flags & (VM_RESERVED|VM_IO))
911 mm->reserved_vm += pages;
912 }
913 #endif /* CONFIG_PROC_FS */
914
915 /*
916 * The caller must hold down_write(¤t->mm->mmap_sem).
917 */
918
919 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
920 unsigned long len, unsigned long prot,
921 unsigned long flags, unsigned long pgoff)
922 {
923 struct mm_struct * mm = current->mm;
924 struct inode *inode;
925 unsigned int vm_flags;
926 int error;
927 unsigned long reqprot = prot;
928
929 /*
930 * Does the application expect PROT_READ to imply PROT_EXEC?
931 *
932 * (the exception is when the underlying filesystem is noexec
933 * mounted, in which case we dont add PROT_EXEC.)
934 */
935 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
936 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
937 prot |= PROT_EXEC;
938
939 if (!len)
940 return -EINVAL;
941
942 if (!(flags & MAP_FIXED))
943 addr = round_hint_to_min(addr);
944
945 /* Careful about overflows.. */
946 len = PAGE_ALIGN(len);
947 if (!len)
948 return -ENOMEM;
949
950 /* offset overflow? */
951 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
952 return -EOVERFLOW;
953
954 /* Too many mappings? */
955 if (mm->map_count > sysctl_max_map_count)
956 return -ENOMEM;
957
958 /* Obtain the address to map to. we verify (or select) it and ensure
959 * that it represents a valid section of the address space.
960 */
961 addr = get_unmapped_area(file, addr, len, pgoff, flags);
962 if (addr & ~PAGE_MASK)
963 return addr;
964
965 /* Do simple checking here so the lower-level routines won't have
966 * to. we assume access permissions have been handled by the open
967 * of the memory object, so we don't do any here.
968 */
969 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
970 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
971
972 if (flags & MAP_LOCKED)
973 if (!can_do_mlock())
974 return -EPERM;
975
976 /* mlock MCL_FUTURE? */
977 if (vm_flags & VM_LOCKED) {
978 unsigned long locked, lock_limit;
979 locked = len >> PAGE_SHIFT;
980 locked += mm->locked_vm;
981 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
982 lock_limit >>= PAGE_SHIFT;
983 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
984 return -EAGAIN;
985 }
986
987 inode = file ? file->f_path.dentry->d_inode : NULL;
988
989 if (file) {
990 switch (flags & MAP_TYPE) {
991 case MAP_SHARED:
992 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
993 return -EACCES;
994
995 /*
996 * Make sure we don't allow writing to an append-only
997 * file..
998 */
999 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1000 return -EACCES;
1001
1002 /*
1003 * Make sure there are no mandatory locks on the file.
1004 */
1005 if (locks_verify_locked(inode))
1006 return -EAGAIN;
1007
1008 vm_flags |= VM_SHARED | VM_MAYSHARE;
1009 if (!(file->f_mode & FMODE_WRITE))
1010 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1011
1012 /* fall through */
1013 case MAP_PRIVATE:
1014 if (!(file->f_mode & FMODE_READ))
1015 return -EACCES;
1016 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1017 if (vm_flags & VM_EXEC)
1018 return -EPERM;
1019 vm_flags &= ~VM_MAYEXEC;
1020 }
1021
1022 if (!file->f_op || !file->f_op->mmap)
1023 return -ENODEV;
1024 break;
1025
1026 default:
1027 return -EINVAL;
1028 }
1029 } else {
1030 switch (flags & MAP_TYPE) {
1031 case MAP_SHARED:
1032 /*
1033 * Ignore pgoff.
1034 */
1035 pgoff = 0;
1036 vm_flags |= VM_SHARED | VM_MAYSHARE;
1037 break;
1038 case MAP_PRIVATE:
1039 /*
1040 * Set pgoff according to addr for anon_vma.
1041 */
1042 pgoff = addr >> PAGE_SHIFT;
1043 break;
1044 default:
1045 return -EINVAL;
1046 }
1047 }
1048
1049 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1050 if (error)
1051 return error;
1052 error = ima_file_mmap(file, prot);
1053 if (error)
1054 return error;
1055
1056 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1057 }
1058 EXPORT_SYMBOL(do_mmap_pgoff);
1059
1060 /*
1061 * Some shared mappigns will want the pages marked read-only
1062 * to track write events. If so, we'll downgrade vm_page_prot
1063 * to the private version (using protection_map[] without the
1064 * VM_SHARED bit).
1065 */
1066 int vma_wants_writenotify(struct vm_area_struct *vma)
1067 {
1068 unsigned int vm_flags = vma->vm_flags;
1069
1070 /* If it was private or non-writable, the write bit is already clear */
1071 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1072 return 0;
1073
1074 /* The backer wishes to know when pages are first written to? */
1075 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1076 return 1;
1077
1078 /* The open routine did something to the protections already? */
1079 if (pgprot_val(vma->vm_page_prot) !=
1080 pgprot_val(vm_get_page_prot(vm_flags)))
1081 return 0;
1082
1083 /* Specialty mapping? */
1084 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1085 return 0;
1086
1087 /* Can the mapping track the dirty pages? */
1088 return vma->vm_file && vma->vm_file->f_mapping &&
1089 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1090 }
1091
1092 /*
1093 * We account for memory if it's a private writeable mapping,
1094 * not hugepages and VM_NORESERVE wasn't set.
1095 */
1096 static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
1097 {
1098 /*
1099 * hugetlb has its own accounting separate from the core VM
1100 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1101 */
1102 if (file && is_file_hugepages(file))
1103 return 0;
1104
1105 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1106 }
1107
1108 unsigned long mmap_region(struct file *file, unsigned long addr,
1109 unsigned long len, unsigned long flags,
1110 unsigned int vm_flags, unsigned long pgoff)
1111 {
1112 struct mm_struct *mm = current->mm;
1113 struct vm_area_struct *vma, *prev;
1114 int correct_wcount = 0;
1115 int error;
1116 struct rb_node **rb_link, *rb_parent;
1117 unsigned long charged = 0;
1118 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1119
1120 /* Clear old maps */
1121 error = -ENOMEM;
1122 munmap_back:
1123 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1124 if (vma && vma->vm_start < addr + len) {
1125 if (do_munmap(mm, addr, len))
1126 return -ENOMEM;
1127 goto munmap_back;
1128 }
1129
1130 /* Check against address space limit. */
1131 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1132 return -ENOMEM;
1133
1134 /*
1135 * Set 'VM_NORESERVE' if we should not account for the
1136 * memory use of this mapping.
1137 */
1138 if ((flags & MAP_NORESERVE)) {
1139 /* We honor MAP_NORESERVE if allowed to overcommit */
1140 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1141 vm_flags |= VM_NORESERVE;
1142
1143 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1144 if (file && is_file_hugepages(file))
1145 vm_flags |= VM_NORESERVE;
1146 }
1147
1148 /*
1149 * Private writable mapping: check memory availability
1150 */
1151 if (accountable_mapping(file, vm_flags)) {
1152 charged = len >> PAGE_SHIFT;
1153 if (security_vm_enough_memory(charged))
1154 return -ENOMEM;
1155 vm_flags |= VM_ACCOUNT;
1156 }
1157
1158 /*
1159 * Can we just expand an old mapping?
1160 */
1161 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1162 if (vma)
1163 goto out;
1164
1165 /*
1166 * Determine the object being mapped and call the appropriate
1167 * specific mapper. the address has already been validated, but
1168 * not unmapped, but the maps are removed from the list.
1169 */
1170 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1171 if (!vma) {
1172 error = -ENOMEM;
1173 goto unacct_error;
1174 }
1175
1176 vma->vm_mm = mm;
1177 vma->vm_start = addr;
1178 vma->vm_end = addr + len;
1179 vma->vm_flags = vm_flags;
1180 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1181 vma->vm_pgoff = pgoff;
1182
1183 if (file) {
1184 error = -EINVAL;
1185 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1186 goto free_vma;
1187 if (vm_flags & VM_DENYWRITE) {
1188 error = deny_write_access(file);
1189 if (error)
1190 goto free_vma;
1191 correct_wcount = 1;
1192 }
1193 vma->vm_file = file;
1194 get_file(file);
1195 error = file->f_op->mmap(file, vma);
1196 if (error)
1197 goto unmap_and_free_vma;
1198 if (vm_flags & VM_EXECUTABLE)
1199 added_exe_file_vma(mm);
1200
1201 /* Can addr have changed??
1202 *
1203 * Answer: Yes, several device drivers can do it in their
1204 * f_op->mmap method. -DaveM
1205 */
1206 addr = vma->vm_start;
1207 pgoff = vma->vm_pgoff;
1208 vm_flags = vma->vm_flags;
1209 } else if (vm_flags & VM_SHARED) {
1210 error = shmem_zero_setup(vma);
1211 if (error)
1212 goto free_vma;
1213 }
1214
1215 if (vma_wants_writenotify(vma))
1216 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1217
1218 vma_link(mm, vma, prev, rb_link, rb_parent);
1219 file = vma->vm_file;
1220
1221 /* Once vma denies write, undo our temporary denial count */
1222 if (correct_wcount)
1223 atomic_inc(&inode->i_writecount);
1224 out:
1225 perf_event_mmap(vma);
1226
1227 mm->total_vm += len >> PAGE_SHIFT;
1228 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1229 if (vm_flags & VM_LOCKED) {
1230 /*
1231 * makes pages present; downgrades, drops, reacquires mmap_sem
1232 */
1233 long nr_pages = mlock_vma_pages_range(vma, addr, addr + len);
1234 if (nr_pages < 0)
1235 return nr_pages; /* vma gone! */
1236 mm->locked_vm += (len >> PAGE_SHIFT) - nr_pages;
1237 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1238 make_pages_present(addr, addr + len);
1239 return addr;
1240
1241 unmap_and_free_vma:
1242 if (correct_wcount)
1243 atomic_inc(&inode->i_writecount);
1244 vma->vm_file = NULL;
1245 fput(file);
1246
1247 /* Undo any partial mapping done by a device driver. */
1248 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1249 charged = 0;
1250 free_vma:
1251 kmem_cache_free(vm_area_cachep, vma);
1252 unacct_error:
1253 if (charged)
1254 vm_unacct_memory(charged);
1255 return error;
1256 }
1257
1258 /* Get an address range which is currently unmapped.
1259 * For shmat() with addr=0.
1260 *
1261 * Ugly calling convention alert:
1262 * Return value with the low bits set means error value,
1263 * ie
1264 * if (ret & ~PAGE_MASK)
1265 * error = ret;
1266 *
1267 * This function "knows" that -ENOMEM has the bits set.
1268 */
1269 #ifndef HAVE_ARCH_UNMAPPED_AREA
1270 unsigned long
1271 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1272 unsigned long len, unsigned long pgoff, unsigned long flags)
1273 {
1274 struct mm_struct *mm = current->mm;
1275 struct vm_area_struct *vma;
1276 unsigned long start_addr;
1277
1278 if (len > TASK_SIZE)
1279 return -ENOMEM;
1280
1281 if (flags & MAP_FIXED)
1282 return addr;
1283
1284 if (addr) {
1285 addr = PAGE_ALIGN(addr);
1286 vma = find_vma(mm, addr);
1287 if (TASK_SIZE - len >= addr &&
1288 (!vma || addr + len <= vma->vm_start))
1289 return addr;
1290 }
1291 if (len > mm->cached_hole_size) {
1292 start_addr = addr = mm->free_area_cache;
1293 } else {
1294 start_addr = addr = TASK_UNMAPPED_BASE;
1295 mm->cached_hole_size = 0;
1296 }
1297
1298 full_search:
1299 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1300 /* At this point: (!vma || addr < vma->vm_end). */
1301 if (TASK_SIZE - len < addr) {
1302 /*
1303 * Start a new search - just in case we missed
1304 * some holes.
1305 */
1306 if (start_addr != TASK_UNMAPPED_BASE) {
1307 addr = TASK_UNMAPPED_BASE;
1308 start_addr = addr;
1309 mm->cached_hole_size = 0;
1310 goto full_search;
1311 }
1312 return -ENOMEM;
1313 }
1314 if (!vma || addr + len <= vma->vm_start) {
1315 /*
1316 * Remember the place where we stopped the search:
1317 */
1318 mm->free_area_cache = addr + len;
1319 return addr;
1320 }
1321 if (addr + mm->cached_hole_size < vma->vm_start)
1322 mm->cached_hole_size = vma->vm_start - addr;
1323 addr = vma->vm_end;
1324 }
1325 }
1326 #endif
1327
1328 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1329 {
1330 /*
1331 * Is this a new hole at the lowest possible address?
1332 */
1333 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1334 mm->free_area_cache = addr;
1335 mm->cached_hole_size = ~0UL;
1336 }
1337 }
1338
1339 /*
1340 * This mmap-allocator allocates new areas top-down from below the
1341 * stack's low limit (the base):
1342 */
1343 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1344 unsigned long
1345 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1346 const unsigned long len, const unsigned long pgoff,
1347 const unsigned long flags)
1348 {
1349 struct vm_area_struct *vma;
1350 struct mm_struct *mm = current->mm;
1351 unsigned long addr = addr0;
1352
1353 /* requested length too big for entire address space */
1354 if (len > TASK_SIZE)
1355 return -ENOMEM;
1356
1357 if (flags & MAP_FIXED)
1358 return addr;
1359
1360 /* requesting a specific address */
1361 if (addr) {
1362 addr = PAGE_ALIGN(addr);
1363 vma = find_vma(mm, addr);
1364 if (TASK_SIZE - len >= addr &&
1365 (!vma || addr + len <= vma->vm_start))
1366 return addr;
1367 }
1368
1369 /* check if free_area_cache is useful for us */
1370 if (len <= mm->cached_hole_size) {
1371 mm->cached_hole_size = 0;
1372 mm->free_area_cache = mm->mmap_base;
1373 }
1374
1375 /* either no address requested or can't fit in requested address hole */
1376 addr = mm->free_area_cache;
1377
1378 /* make sure it can fit in the remaining address space */
1379 if (addr > len) {
1380 vma = find_vma(mm, addr-len);
1381 if (!vma || addr <= vma->vm_start)
1382 /* remember the address as a hint for next time */
1383 return (mm->free_area_cache = addr-len);
1384 }
1385
1386 if (mm->mmap_base < len)
1387 goto bottomup;
1388
1389 addr = mm->mmap_base-len;
1390
1391 do {
1392 /*
1393 * Lookup failure means no vma is above this address,
1394 * else if new region fits below vma->vm_start,
1395 * return with success:
1396 */
1397 vma = find_vma(mm, addr);
1398 if (!vma || addr+len <= vma->vm_start)
1399 /* remember the address as a hint for next time */
1400 return (mm->free_area_cache = addr);
1401
1402 /* remember the largest hole we saw so far */
1403 if (addr + mm->cached_hole_size < vma->vm_start)
1404 mm->cached_hole_size = vma->vm_start - addr;
1405
1406 /* try just below the current vma->vm_start */
1407 addr = vma->vm_start-len;
1408 } while (len < vma->vm_start);
1409
1410 bottomup:
1411 /*
1412 * A failed mmap() very likely causes application failure,
1413 * so fall back to the bottom-up function here. This scenario
1414 * can happen with large stack limits and large mmap()
1415 * allocations.
1416 */
1417 mm->cached_hole_size = ~0UL;
1418 mm->free_area_cache = TASK_UNMAPPED_BASE;
1419 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1420 /*
1421 * Restore the topdown base:
1422 */
1423 mm->free_area_cache = mm->mmap_base;
1424 mm->cached_hole_size = ~0UL;
1425
1426 return addr;
1427 }
1428 #endif
1429
1430 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1431 {
1432 /*
1433 * Is this a new hole at the highest possible address?
1434 */
1435 if (addr > mm->free_area_cache)
1436 mm->free_area_cache = addr;
1437
1438 /* dont allow allocations above current base */
1439 if (mm->free_area_cache > mm->mmap_base)
1440 mm->free_area_cache = mm->mmap_base;
1441 }
1442
1443 unsigned long
1444 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1445 unsigned long pgoff, unsigned long flags)
1446 {
1447 unsigned long (*get_area)(struct file *, unsigned long,
1448 unsigned long, unsigned long, unsigned long);
1449
1450 unsigned long error = arch_mmap_check(addr, len, flags);
1451 if (error)
1452 return error;
1453
1454 /* Careful about overflows.. */
1455 if (len > TASK_SIZE)
1456 return -ENOMEM;
1457
1458 get_area = current->mm->get_unmapped_area;
1459 if (file && file->f_op && file->f_op->get_unmapped_area)
1460 get_area = file->f_op->get_unmapped_area;
1461 addr = get_area(file, addr, len, pgoff, flags);
1462 if (IS_ERR_VALUE(addr))
1463 return addr;
1464
1465 if (addr > TASK_SIZE - len)
1466 return -ENOMEM;
1467 if (addr & ~PAGE_MASK)
1468 return -EINVAL;
1469
1470 return arch_rebalance_pgtables(addr, len);
1471 }
1472
1473 EXPORT_SYMBOL(get_unmapped_area);
1474
1475 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1476 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1477 {
1478 struct vm_area_struct *vma = NULL;
1479
1480 if (mm) {
1481 /* Check the cache first. */
1482 /* (Cache hit rate is typically around 35%.) */
1483 vma = mm->mmap_cache;
1484 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1485 struct rb_node * rb_node;
1486
1487 rb_node = mm->mm_rb.rb_node;
1488 vma = NULL;
1489
1490 while (rb_node) {
1491 struct vm_area_struct * vma_tmp;
1492
1493 vma_tmp = rb_entry(rb_node,
1494 struct vm_area_struct, vm_rb);
1495
1496 if (vma_tmp->vm_end > addr) {
1497 vma = vma_tmp;
1498 if (vma_tmp->vm_start <= addr)
1499 break;
1500 rb_node = rb_node->rb_left;
1501 } else
1502 rb_node = rb_node->rb_right;
1503 }
1504 if (vma)
1505 mm->mmap_cache = vma;
1506 }
1507 }
1508 return vma;
1509 }
1510
1511 EXPORT_SYMBOL(find_vma);
1512
1513 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1514 struct vm_area_struct *
1515 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1516 struct vm_area_struct **pprev)
1517 {
1518 struct vm_area_struct *vma = NULL, *prev = NULL;
1519 struct rb_node *rb_node;
1520 if (!mm)
1521 goto out;
1522
1523 /* Guard against addr being lower than the first VMA */
1524 vma = mm->mmap;
1525
1526 /* Go through the RB tree quickly. */
1527 rb_node = mm->mm_rb.rb_node;
1528
1529 while (rb_node) {
1530 struct vm_area_struct *vma_tmp;
1531 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1532
1533 if (addr < vma_tmp->vm_end) {
1534 rb_node = rb_node->rb_left;
1535 } else {
1536 prev = vma_tmp;
1537 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1538 break;
1539 rb_node = rb_node->rb_right;
1540 }
1541 }
1542
1543 out:
1544 *pprev = prev;
1545 return prev ? prev->vm_next : vma;
1546 }
1547
1548 /*
1549 * Verify that the stack growth is acceptable and
1550 * update accounting. This is shared with both the
1551 * grow-up and grow-down cases.
1552 */
1553 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1554 {
1555 struct mm_struct *mm = vma->vm_mm;
1556 struct rlimit *rlim = current->signal->rlim;
1557 unsigned long new_start;
1558
1559 /* address space limit tests */
1560 if (!may_expand_vm(mm, grow))
1561 return -ENOMEM;
1562
1563 /* Stack limit test */
1564 if (size > rlim[RLIMIT_STACK].rlim_cur)
1565 return -ENOMEM;
1566
1567 /* mlock limit tests */
1568 if (vma->vm_flags & VM_LOCKED) {
1569 unsigned long locked;
1570 unsigned long limit;
1571 locked = mm->locked_vm + grow;
1572 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1573 if (locked > limit && !capable(CAP_IPC_LOCK))
1574 return -ENOMEM;
1575 }
1576
1577 /* Check to ensure the stack will not grow into a hugetlb-only region */
1578 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1579 vma->vm_end - size;
1580 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1581 return -EFAULT;
1582
1583 /*
1584 * Overcommit.. This must be the final test, as it will
1585 * update security statistics.
1586 */
1587 if (security_vm_enough_memory_mm(mm, grow))
1588 return -ENOMEM;
1589
1590 /* Ok, everything looks good - let it rip */
1591 mm->total_vm += grow;
1592 if (vma->vm_flags & VM_LOCKED)
1593 mm->locked_vm += grow;
1594 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1595 return 0;
1596 }
1597
1598 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1599 /*
1600 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1601 * vma is the last one with address > vma->vm_end. Have to extend vma.
1602 */
1603 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1604 {
1605 int error;
1606
1607 if (!(vma->vm_flags & VM_GROWSUP))
1608 return -EFAULT;
1609
1610 /*
1611 * We must make sure the anon_vma is allocated
1612 * so that the anon_vma locking is not a noop.
1613 */
1614 if (unlikely(anon_vma_prepare(vma)))
1615 return -ENOMEM;
1616 anon_vma_lock(vma);
1617
1618 /*
1619 * vma->vm_start/vm_end cannot change under us because the caller
1620 * is required to hold the mmap_sem in read mode. We need the
1621 * anon_vma lock to serialize against concurrent expand_stacks.
1622 * Also guard against wrapping around to address 0.
1623 */
1624 if (address < PAGE_ALIGN(address+4))
1625 address = PAGE_ALIGN(address+4);
1626 else {
1627 anon_vma_unlock(vma);
1628 return -ENOMEM;
1629 }
1630 error = 0;
1631
1632 /* Somebody else might have raced and expanded it already */
1633 if (address > vma->vm_end) {
1634 unsigned long size, grow;
1635
1636 size = address - vma->vm_start;
1637 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1638
1639 error = -ENOMEM;
1640 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1641 error = acct_stack_growth(vma, size, grow);
1642 if (!error)
1643 vma->vm_end = address;
1644 }
1645 }
1646 anon_vma_unlock(vma);
1647 return error;
1648 }
1649 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1650
1651 /*
1652 * vma is the first one with address < vma->vm_start. Have to extend vma.
1653 */
1654 static int expand_downwards(struct vm_area_struct *vma,
1655 unsigned long address)
1656 {
1657 int error;
1658
1659 /*
1660 * We must make sure the anon_vma is allocated
1661 * so that the anon_vma locking is not a noop.
1662 */
1663 if (unlikely(anon_vma_prepare(vma)))
1664 return -ENOMEM;
1665
1666 address &= PAGE_MASK;
1667 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1668 if (error)
1669 return error;
1670
1671 anon_vma_lock(vma);
1672
1673 /*
1674 * vma->vm_start/vm_end cannot change under us because the caller
1675 * is required to hold the mmap_sem in read mode. We need the
1676 * anon_vma lock to serialize against concurrent expand_stacks.
1677 */
1678
1679 /* Somebody else might have raced and expanded it already */
1680 if (address < vma->vm_start) {
1681 unsigned long size, grow;
1682
1683 size = vma->vm_end - address;
1684 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1685
1686 error = -ENOMEM;
1687 if (grow <= vma->vm_pgoff) {
1688 error = acct_stack_growth(vma, size, grow);
1689 if (!error) {
1690 vma->vm_start = address;
1691 vma->vm_pgoff -= grow;
1692 }
1693 }
1694 }
1695 anon_vma_unlock(vma);
1696 return error;
1697 }
1698
1699 int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1700 {
1701 return expand_downwards(vma, address);
1702 }
1703
1704 #ifdef CONFIG_STACK_GROWSUP
1705 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1706 {
1707 return expand_upwards(vma, address);
1708 }
1709
1710 struct vm_area_struct *
1711 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1712 {
1713 struct vm_area_struct *vma, *prev;
1714
1715 addr &= PAGE_MASK;
1716 vma = find_vma_prev(mm, addr, &prev);
1717 if (vma && (vma->vm_start <= addr))
1718 return vma;
1719 if (!prev || expand_stack(prev, addr))
1720 return NULL;
1721 if (prev->vm_flags & VM_LOCKED) {
1722 if (mlock_vma_pages_range(prev, addr, prev->vm_end) < 0)
1723 return NULL; /* vma gone! */
1724 }
1725 return prev;
1726 }
1727 #else
1728 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1729 {
1730 return expand_downwards(vma, address);
1731 }
1732
1733 struct vm_area_struct *
1734 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1735 {
1736 struct vm_area_struct * vma;
1737 unsigned long start;
1738
1739 addr &= PAGE_MASK;
1740 vma = find_vma(mm,addr);
1741 if (!vma)
1742 return NULL;
1743 if (vma->vm_start <= addr)
1744 return vma;
1745 if (!(vma->vm_flags & VM_GROWSDOWN))
1746 return NULL;
1747 start = vma->vm_start;
1748 if (expand_stack(vma, addr))
1749 return NULL;
1750 if (vma->vm_flags & VM_LOCKED) {
1751 if (mlock_vma_pages_range(vma, addr, start) < 0)
1752 return NULL; /* vma gone! */
1753 }
1754 return vma;
1755 }
1756 #endif
1757
1758 /*
1759 * Ok - we have the memory areas we should free on the vma list,
1760 * so release them, and do the vma updates.
1761 *
1762 * Called with the mm semaphore held.
1763 */
1764 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1765 {
1766 /* Update high watermark before we lower total_vm */
1767 update_hiwater_vm(mm);
1768 do {
1769 long nrpages = vma_pages(vma);
1770
1771 mm->total_vm -= nrpages;
1772 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1773 vma = remove_vma(vma);
1774 } while (vma);
1775 validate_mm(mm);
1776 }
1777
1778 /*
1779 * Get rid of page table information in the indicated region.
1780 *
1781 * Called with the mm semaphore held.
1782 */
1783 static void unmap_region(struct mm_struct *mm,
1784 struct vm_area_struct *vma, struct vm_area_struct *prev,
1785 unsigned long start, unsigned long end)
1786 {
1787 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1788 struct mmu_gather *tlb;
1789 unsigned long nr_accounted = 0;
1790
1791 lru_add_drain();
1792 tlb = tlb_gather_mmu(mm, 0);
1793 update_hiwater_rss(mm);
1794 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1795 vm_unacct_memory(nr_accounted);
1796 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1797 next? next->vm_start: 0);
1798 tlb_finish_mmu(tlb, start, end);
1799 }
1800
1801 /*
1802 * Create a list of vma's touched by the unmap, removing them from the mm's
1803 * vma list as we go..
1804 */
1805 static void
1806 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1807 struct vm_area_struct *prev, unsigned long end)
1808 {
1809 struct vm_area_struct **insertion_point;
1810 struct vm_area_struct *tail_vma = NULL;
1811 unsigned long addr;
1812
1813 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1814 vma->vm_prev = NULL;
1815 do {
1816 rb_erase(&vma->vm_rb, &mm->mm_rb);
1817 mm->map_count--;
1818 tail_vma = vma;
1819 vma = vma->vm_next;
1820 } while (vma && vma->vm_start < end);
1821 *insertion_point = vma;
1822 if (vma)
1823 vma->vm_prev = prev;
1824 tail_vma->vm_next = NULL;
1825 if (mm->unmap_area == arch_unmap_area)
1826 addr = prev ? prev->vm_end : mm->mmap_base;
1827 else
1828 addr = vma ? vma->vm_start : mm->mmap_base;
1829 mm->unmap_area(mm, addr);
1830 mm->mmap_cache = NULL; /* Kill the cache. */
1831 }
1832
1833 /*
1834 * Split a vma into two pieces at address 'addr', a new vma is allocated
1835 * either for the first part or the tail.
1836 */
1837 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1838 unsigned long addr, int new_below)
1839 {
1840 struct mempolicy *pol;
1841 struct vm_area_struct *new;
1842
1843 if (is_vm_hugetlb_page(vma) && (addr &
1844 ~(huge_page_mask(hstate_vma(vma)))))
1845 return -EINVAL;
1846
1847 if (mm->map_count >= sysctl_max_map_count)
1848 return -ENOMEM;
1849
1850 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1851 if (!new)
1852 return -ENOMEM;
1853
1854 /* most fields are the same, copy all, and then fixup */
1855 *new = *vma;
1856
1857 if (new_below)
1858 new->vm_end = addr;
1859 else {
1860 new->vm_start = addr;
1861 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1862 }
1863
1864 pol = mpol_dup(vma_policy(vma));
1865 if (IS_ERR(pol)) {
1866 kmem_cache_free(vm_area_cachep, new);
1867 return PTR_ERR(pol);
1868 }
1869 vma_set_policy(new, pol);
1870
1871 if (new->vm_file) {
1872 get_file(new->vm_file);
1873 if (vma->vm_flags & VM_EXECUTABLE)
1874 added_exe_file_vma(mm);
1875 }
1876
1877 if (new->vm_ops && new->vm_ops->open)
1878 new->vm_ops->open(new);
1879
1880 if (new_below)
1881 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1882 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1883 else
1884 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1885
1886 return 0;
1887 }
1888
1889 /* Munmap is split into 2 main parts -- this part which finds
1890 * what needs doing, and the areas themselves, which do the
1891 * work. This now handles partial unmappings.
1892 * Jeremy Fitzhardinge <jeremy@goop.org>
1893 */
1894 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1895 {
1896 unsigned long end;
1897 struct vm_area_struct *vma, *prev, *last;
1898
1899 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1900 return -EINVAL;
1901
1902 if ((len = PAGE_ALIGN(len)) == 0)
1903 return -EINVAL;
1904
1905 /* Find the first overlapping VMA */
1906 vma = find_vma_prev(mm, start, &prev);
1907 if (!vma)
1908 return 0;
1909 /* we have start < vma->vm_end */
1910
1911 /* if it doesn't overlap, we have nothing.. */
1912 end = start + len;
1913 if (vma->vm_start >= end)
1914 return 0;
1915
1916 /*
1917 * If we need to split any vma, do it now to save pain later.
1918 *
1919 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1920 * unmapped vm_area_struct will remain in use: so lower split_vma
1921 * places tmp vma above, and higher split_vma places tmp vma below.
1922 */
1923 if (start > vma->vm_start) {
1924 int error = split_vma(mm, vma, start, 0);
1925 if (error)
1926 return error;
1927 prev = vma;
1928 }
1929
1930 /* Does it split the last one? */
1931 last = find_vma(mm, end);
1932 if (last && end > last->vm_start) {
1933 int error = split_vma(mm, last, end, 1);
1934 if (error)
1935 return error;
1936 }
1937 vma = prev? prev->vm_next: mm->mmap;
1938
1939 /*
1940 * unlock any mlock()ed ranges before detaching vmas
1941 */
1942 if (mm->locked_vm) {
1943 struct vm_area_struct *tmp = vma;
1944 while (tmp && tmp->vm_start < end) {
1945 if (tmp->vm_flags & VM_LOCKED) {
1946 mm->locked_vm -= vma_pages(tmp);
1947 munlock_vma_pages_all(tmp);
1948 }
1949 tmp = tmp->vm_next;
1950 }
1951 }
1952
1953 /*
1954 * Remove the vma's, and unmap the actual pages
1955 */
1956 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1957 unmap_region(mm, vma, prev, start, end);
1958
1959 /* Fix up all other VM information */
1960 remove_vma_list(mm, vma);
1961
1962 return 0;
1963 }
1964
1965 EXPORT_SYMBOL(do_munmap);
1966
1967 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1968 {
1969 int ret;
1970 struct mm_struct *mm = current->mm;
1971
1972 profile_munmap(addr);
1973
1974 down_write(&mm->mmap_sem);
1975 ret = do_munmap(mm, addr, len);
1976 up_write(&mm->mmap_sem);
1977 return ret;
1978 }
1979
1980 static inline void verify_mm_writelocked(struct mm_struct *mm)
1981 {
1982 #ifdef CONFIG_DEBUG_VM
1983 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1984 WARN_ON(1);
1985 up_read(&mm->mmap_sem);
1986 }
1987 #endif
1988 }
1989
1990 /*
1991 * this is really a simplified "do_mmap". it only handles
1992 * anonymous maps. eventually we may be able to do some
1993 * brk-specific accounting here.
1994 */
1995 unsigned long do_brk(unsigned long addr, unsigned long len)
1996 {
1997 struct mm_struct * mm = current->mm;
1998 struct vm_area_struct * vma, * prev;
1999 unsigned long flags;
2000 struct rb_node ** rb_link, * rb_parent;
2001 pgoff_t pgoff = addr >> PAGE_SHIFT;
2002 int error;
2003
2004 len = PAGE_ALIGN(len);
2005 if (!len)
2006 return addr;
2007
2008 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2009 if (error)
2010 return error;
2011
2012 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2013
2014 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2015 if (error & ~PAGE_MASK)
2016 return error;
2017
2018 /*
2019 * mlock MCL_FUTURE?
2020 */
2021 if (mm->def_flags & VM_LOCKED) {
2022 unsigned long locked, lock_limit;
2023 locked = len >> PAGE_SHIFT;
2024 locked += mm->locked_vm;
2025 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
2026 lock_limit >>= PAGE_SHIFT;
2027 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2028 return -EAGAIN;
2029 }
2030
2031 /*
2032 * mm->mmap_sem is required to protect against another thread
2033 * changing the mappings in case we sleep.
2034 */
2035 verify_mm_writelocked(mm);
2036
2037 /*
2038 * Clear old maps. this also does some error checking for us
2039 */
2040 munmap_back:
2041 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2042 if (vma && vma->vm_start < addr + len) {
2043 if (do_munmap(mm, addr, len))
2044 return -ENOMEM;
2045 goto munmap_back;
2046 }
2047
2048 /* Check against address space limits *after* clearing old maps... */
2049 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2050 return -ENOMEM;
2051
2052 if (mm->map_count > sysctl_max_map_count)
2053 return -ENOMEM;
2054
2055 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2056 return -ENOMEM;
2057
2058 /* Can we just expand an old private anonymous mapping? */
2059 vma = vma_merge(mm, prev, addr, addr + len, flags,
2060 NULL, NULL, pgoff, NULL);
2061 if (vma)
2062 goto out;
2063
2064 /*
2065 * create a vma struct for an anonymous mapping
2066 */
2067 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2068 if (!vma) {
2069 vm_unacct_memory(len >> PAGE_SHIFT);
2070 return -ENOMEM;
2071 }
2072
2073 vma->vm_mm = mm;
2074 vma->vm_start = addr;
2075 vma->vm_end = addr + len;
2076 vma->vm_pgoff = pgoff;
2077 vma->vm_flags = flags;
2078 vma->vm_page_prot = vm_get_page_prot(flags);
2079 vma_link(mm, vma, prev, rb_link, rb_parent);
2080 out:
2081 mm->total_vm += len >> PAGE_SHIFT;
2082 if (flags & VM_LOCKED) {
2083 if (!mlock_vma_pages_range(vma, addr, addr + len))
2084 mm->locked_vm += (len >> PAGE_SHIFT);
2085 }
2086 return addr;
2087 }
2088
2089 EXPORT_SYMBOL(do_brk);
2090
2091 /* Release all mmaps. */
2092 void exit_mmap(struct mm_struct *mm)
2093 {
2094 struct mmu_gather *tlb;
2095 struct vm_area_struct *vma;
2096 unsigned long nr_accounted = 0;
2097 unsigned long end;
2098
2099 /* mm's last user has gone, and its about to be pulled down */
2100 mmu_notifier_release(mm);
2101
2102 if (mm->locked_vm) {
2103 vma = mm->mmap;
2104 while (vma) {
2105 if (vma->vm_flags & VM_LOCKED)
2106 munlock_vma_pages_all(vma);
2107 vma = vma->vm_next;
2108 }
2109 }
2110
2111 arch_exit_mmap(mm);
2112
2113 vma = mm->mmap;
2114 if (!vma) /* Can happen if dup_mmap() received an OOM */
2115 return;
2116
2117 lru_add_drain();
2118 flush_cache_mm(mm);
2119 tlb = tlb_gather_mmu(mm, 1);
2120 /* update_hiwater_rss(mm) here? but nobody should be looking */
2121 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2122 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2123 vm_unacct_memory(nr_accounted);
2124
2125 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
2126 tlb_finish_mmu(tlb, 0, end);
2127
2128 /*
2129 * Walk the list again, actually closing and freeing it,
2130 * with preemption enabled, without holding any MM locks.
2131 */
2132 while (vma)
2133 vma = remove_vma(vma);
2134
2135 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2136 }
2137
2138 /* Insert vm structure into process list sorted by address
2139 * and into the inode's i_mmap tree. If vm_file is non-NULL
2140 * then i_mmap_lock is taken here.
2141 */
2142 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2143 {
2144 struct vm_area_struct * __vma, * prev;
2145 struct rb_node ** rb_link, * rb_parent;
2146
2147 /*
2148 * The vm_pgoff of a purely anonymous vma should be irrelevant
2149 * until its first write fault, when page's anon_vma and index
2150 * are set. But now set the vm_pgoff it will almost certainly
2151 * end up with (unless mremap moves it elsewhere before that
2152 * first wfault), so /proc/pid/maps tells a consistent story.
2153 *
2154 * By setting it to reflect the virtual start address of the
2155 * vma, merges and splits can happen in a seamless way, just
2156 * using the existing file pgoff checks and manipulations.
2157 * Similarly in do_mmap_pgoff and in do_brk.
2158 */
2159 if (!vma->vm_file) {
2160 BUG_ON(vma->anon_vma);
2161 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2162 }
2163 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2164 if (__vma && __vma->vm_start < vma->vm_end)
2165 return -ENOMEM;
2166 if ((vma->vm_flags & VM_ACCOUNT) &&
2167 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2168 return -ENOMEM;
2169 vma_link(mm, vma, prev, rb_link, rb_parent);
2170 return 0;
2171 }
2172
2173 /*
2174 * Copy the vma structure to a new location in the same mm,
2175 * prior to moving page table entries, to effect an mremap move.
2176 */
2177 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2178 unsigned long addr, unsigned long len, pgoff_t pgoff)
2179 {
2180 struct vm_area_struct *vma = *vmap;
2181 unsigned long vma_start = vma->vm_start;
2182 struct mm_struct *mm = vma->vm_mm;
2183 struct vm_area_struct *new_vma, *prev;
2184 struct rb_node **rb_link, *rb_parent;
2185 struct mempolicy *pol;
2186
2187 /*
2188 * If anonymous vma has not yet been faulted, update new pgoff
2189 * to match new location, to increase its chance of merging.
2190 */
2191 if (!vma->vm_file && !vma->anon_vma)
2192 pgoff = addr >> PAGE_SHIFT;
2193
2194 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2195 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2196 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2197 if (new_vma) {
2198 /*
2199 * Source vma may have been merged into new_vma
2200 */
2201 if (vma_start >= new_vma->vm_start &&
2202 vma_start < new_vma->vm_end)
2203 *vmap = new_vma;
2204 } else {
2205 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2206 if (new_vma) {
2207 *new_vma = *vma;
2208 pol = mpol_dup(vma_policy(vma));
2209 if (IS_ERR(pol)) {
2210 kmem_cache_free(vm_area_cachep, new_vma);
2211 return NULL;
2212 }
2213 vma_set_policy(new_vma, pol);
2214 new_vma->vm_start = addr;
2215 new_vma->vm_end = addr + len;
2216 new_vma->vm_pgoff = pgoff;
2217 if (new_vma->vm_file) {
2218 get_file(new_vma->vm_file);
2219 if (vma->vm_flags & VM_EXECUTABLE)
2220 added_exe_file_vma(mm);
2221 }
2222 if (new_vma->vm_ops && new_vma->vm_ops->open)
2223 new_vma->vm_ops->open(new_vma);
2224 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2225 }
2226 }
2227 return new_vma;
2228 }
2229
2230 /*
2231 * Return true if the calling process may expand its vm space by the passed
2232 * number of pages
2233 */
2234 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2235 {
2236 unsigned long cur = mm->total_vm; /* pages */
2237 unsigned long lim;
2238
2239 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2240
2241 if (cur + npages > lim)
2242 return 0;
2243 return 1;
2244 }
2245
2246
2247 static int special_mapping_fault(struct vm_area_struct *vma,
2248 struct vm_fault *vmf)
2249 {
2250 pgoff_t pgoff;
2251 struct page **pages;
2252
2253 /*
2254 * special mappings have no vm_file, and in that case, the mm
2255 * uses vm_pgoff internally. So we have to subtract it from here.
2256 * We are allowed to do this because we are the mm; do not copy
2257 * this code into drivers!
2258 */
2259 pgoff = vmf->pgoff - vma->vm_pgoff;
2260
2261 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2262 pgoff--;
2263
2264 if (*pages) {
2265 struct page *page = *pages;
2266 get_page(page);
2267 vmf->page = page;
2268 return 0;
2269 }
2270
2271 return VM_FAULT_SIGBUS;
2272 }
2273
2274 /*
2275 * Having a close hook prevents vma merging regardless of flags.
2276 */
2277 static void special_mapping_close(struct vm_area_struct *vma)
2278 {
2279 }
2280
2281 static const struct vm_operations_struct special_mapping_vmops = {
2282 .close = special_mapping_close,
2283 .fault = special_mapping_fault,
2284 };
2285
2286 /*
2287 * Called with mm->mmap_sem held for writing.
2288 * Insert a new vma covering the given region, with the given flags.
2289 * Its pages are supplied by the given array of struct page *.
2290 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2291 * The region past the last page supplied will always produce SIGBUS.
2292 * The array pointer and the pages it points to are assumed to stay alive
2293 * for as long as this mapping might exist.
2294 */
2295 int install_special_mapping(struct mm_struct *mm,
2296 unsigned long addr, unsigned long len,
2297 unsigned long vm_flags, struct page **pages)
2298 {
2299 int ret;
2300 struct vm_area_struct *vma;
2301
2302 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2303 if (unlikely(vma == NULL))
2304 return -ENOMEM;
2305
2306 vma->vm_mm = mm;
2307 vma->vm_start = addr;
2308 vma->vm_end = addr + len;
2309
2310 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2311 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2312
2313 vma->vm_ops = &special_mapping_vmops;
2314 vma->vm_private_data = pages;
2315
2316 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2317 if (ret)
2318 goto out;
2319
2320 ret = insert_vm_struct(mm, vma);
2321 if (ret)
2322 goto out;
2323
2324 mm->total_vm += len >> PAGE_SHIFT;
2325
2326 perf_event_mmap(vma);
2327
2328 return 0;
2329
2330 out:
2331 kmem_cache_free(vm_area_cachep, vma);
2332 return ret;
2333 }
2334
2335 static DEFINE_MUTEX(mm_all_locks_mutex);
2336
2337 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2338 {
2339 if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2340 /*
2341 * The LSB of head.next can't change from under us
2342 * because we hold the mm_all_locks_mutex.
2343 */
2344 spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
2345 /*
2346 * We can safely modify head.next after taking the
2347 * anon_vma->lock. If some other vma in this mm shares
2348 * the same anon_vma we won't take it again.
2349 *
2350 * No need of atomic instructions here, head.next
2351 * can't change from under us thanks to the
2352 * anon_vma->lock.
2353 */
2354 if (__test_and_set_bit(0, (unsigned long *)
2355 &anon_vma->head.next))
2356 BUG();
2357 }
2358 }
2359
2360 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2361 {
2362 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2363 /*
2364 * AS_MM_ALL_LOCKS can't change from under us because
2365 * we hold the mm_all_locks_mutex.
2366 *
2367 * Operations on ->flags have to be atomic because
2368 * even if AS_MM_ALL_LOCKS is stable thanks to the
2369 * mm_all_locks_mutex, there may be other cpus
2370 * changing other bitflags in parallel to us.
2371 */
2372 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2373 BUG();
2374 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
2375 }
2376 }
2377
2378 /*
2379 * This operation locks against the VM for all pte/vma/mm related
2380 * operations that could ever happen on a certain mm. This includes
2381 * vmtruncate, try_to_unmap, and all page faults.
2382 *
2383 * The caller must take the mmap_sem in write mode before calling
2384 * mm_take_all_locks(). The caller isn't allowed to release the
2385 * mmap_sem until mm_drop_all_locks() returns.
2386 *
2387 * mmap_sem in write mode is required in order to block all operations
2388 * that could modify pagetables and free pages without need of
2389 * altering the vma layout (for example populate_range() with
2390 * nonlinear vmas). It's also needed in write mode to avoid new
2391 * anon_vmas to be associated with existing vmas.
2392 *
2393 * A single task can't take more than one mm_take_all_locks() in a row
2394 * or it would deadlock.
2395 *
2396 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2397 * mapping->flags avoid to take the same lock twice, if more than one
2398 * vma in this mm is backed by the same anon_vma or address_space.
2399 *
2400 * We can take all the locks in random order because the VM code
2401 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2402 * takes more than one of them in a row. Secondly we're protected
2403 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2404 *
2405 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2406 * that may have to take thousand of locks.
2407 *
2408 * mm_take_all_locks() can fail if it's interrupted by signals.
2409 */
2410 int mm_take_all_locks(struct mm_struct *mm)
2411 {
2412 struct vm_area_struct *vma;
2413 int ret = -EINTR;
2414
2415 BUG_ON(down_read_trylock(&mm->mmap_sem));
2416
2417 mutex_lock(&mm_all_locks_mutex);
2418
2419 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2420 if (signal_pending(current))
2421 goto out_unlock;
2422 if (vma->vm_file && vma->vm_file->f_mapping)
2423 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2424 }
2425
2426 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2427 if (signal_pending(current))
2428 goto out_unlock;
2429 if (vma->anon_vma)
2430 vm_lock_anon_vma(mm, vma->anon_vma);
2431 }
2432
2433 ret = 0;
2434
2435 out_unlock:
2436 if (ret)
2437 mm_drop_all_locks(mm);
2438
2439 return ret;
2440 }
2441
2442 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2443 {
2444 if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2445 /*
2446 * The LSB of head.next can't change to 0 from under
2447 * us because we hold the mm_all_locks_mutex.
2448 *
2449 * We must however clear the bitflag before unlocking
2450 * the vma so the users using the anon_vma->head will
2451 * never see our bitflag.
2452 *
2453 * No need of atomic instructions here, head.next
2454 * can't change from under us until we release the
2455 * anon_vma->lock.
2456 */
2457 if (!__test_and_clear_bit(0, (unsigned long *)
2458 &anon_vma->head.next))
2459 BUG();
2460 spin_unlock(&anon_vma->lock);
2461 }
2462 }
2463
2464 static void vm_unlock_mapping(struct address_space *mapping)
2465 {
2466 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2467 /*
2468 * AS_MM_ALL_LOCKS can't change to 0 from under us
2469 * because we hold the mm_all_locks_mutex.
2470 */
2471 spin_unlock(&mapping->i_mmap_lock);
2472 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2473 &mapping->flags))
2474 BUG();
2475 }
2476 }
2477
2478 /*
2479 * The mmap_sem cannot be released by the caller until
2480 * mm_drop_all_locks() returns.
2481 */
2482 void mm_drop_all_locks(struct mm_struct *mm)
2483 {
2484 struct vm_area_struct *vma;
2485
2486 BUG_ON(down_read_trylock(&mm->mmap_sem));
2487 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2488
2489 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2490 if (vma->anon_vma)
2491 vm_unlock_anon_vma(vma->anon_vma);
2492 if (vma->vm_file && vma->vm_file->f_mapping)
2493 vm_unlock_mapping(vma->vm_file->f_mapping);
2494 }
2495
2496 mutex_unlock(&mm_all_locks_mutex);
2497 }
2498
2499 /*
2500 * initialise the VMA slab
2501 */
2502 void __init mmap_init(void)
2503 {
2504 int ret;
2505
2506 ret = percpu_counter_init(&vm_committed_as, 0);
2507 VM_BUG_ON(ret);
2508 }
2509
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.