security: trim security.h
[deliverable/linux.git] / mm / mmap.c
CommitLineData
1da177e4
LT
1/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
046c6884 6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
1da177e4
LT
7 */
8
9#include <linux/slab.h>
4af3c9cc 10#include <linux/backing-dev.h>
1da177e4
LT
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>
c59ede7b 17#include <linux/capability.h>
1da177e4
LT
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/hugetlb.h>
24#include <linux/profile.h>
b95f1b31 25#include <linux/export.h>
1da177e4
LT
26#include <linux/mount.h>
27#include <linux/mempolicy.h>
28#include <linux/rmap.h>
cddb8a5c 29#include <linux/mmu_notifier.h>
cdd6c482 30#include <linux/perf_event.h>
120a795d 31#include <linux/audit.h>
b15d00b6 32#include <linux/khugepaged.h>
1da177e4
LT
33
34#include <asm/uaccess.h>
35#include <asm/cacheflush.h>
36#include <asm/tlb.h>
d6dd61c8 37#include <asm/mmu_context.h>
1da177e4 38
42b77728
JB
39#include "internal.h"
40
3a459756
KK
41#ifndef arch_mmap_check
42#define arch_mmap_check(addr, len, flags) (0)
43#endif
44
08e7d9b5
MS
45#ifndef arch_rebalance_pgtables
46#define arch_rebalance_pgtables(addr, len) (addr)
47#endif
48
e0da382c
HD
49static void unmap_region(struct mm_struct *mm,
50 struct vm_area_struct *vma, struct vm_area_struct *prev,
51 unsigned long start, unsigned long end);
52
1da177e4
LT
53/*
54 * WARNING: the debugging will use recursive algorithms so never enable this
55 * unless you know what you are doing.
56 */
57#undef DEBUG_MM_RB
58
59/* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
62 *
63 * map_type prot
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
68 *
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 *
73 */
74pgprot_t protection_map[16] = {
75 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
76 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
77};
78
804af2cf
HD
79pgprot_t vm_get_page_prot(unsigned long vm_flags)
80{
b845f313
DK
81 return __pgprot(pgprot_val(protection_map[vm_flags &
82 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
83 pgprot_val(arch_vm_get_page_prot(vm_flags)));
804af2cf
HD
84}
85EXPORT_SYMBOL(vm_get_page_prot);
86
34679d7e
SL
87int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
88int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
c3d8c141 89int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
34679d7e
SL
90/*
91 * Make sure vm_committed_as in one cacheline and not cacheline shared with
92 * other variables. It can be updated by several CPUs frequently.
93 */
94struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
1da177e4
LT
95
96/*
97 * Check that a process has enough memory to allocate a new virtual
98 * mapping. 0 means there is enough memory for the allocation to
99 * succeed and -ENOMEM implies there is not.
100 *
101 * We currently support three overcommit policies, which are set via the
102 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
103 *
104 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
105 * Additional code 2002 Jul 20 by Robert Love.
106 *
107 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
108 *
109 * Note this is a helper function intended to be used by LSMs which
110 * wish to use this logic.
111 */
34b4e4aa 112int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1da177e4
LT
113{
114 unsigned long free, allowed;
115
116 vm_acct_memory(pages);
117
118 /*
119 * Sometimes we want to use more memory than we have
120 */
121 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
122 return 0;
123
124 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
c15bef30
DF
125 free = global_page_state(NR_FREE_PAGES);
126 free += global_page_state(NR_FILE_PAGES);
127
128 /*
129 * shmem pages shouldn't be counted as free in this
130 * case, they can't be purged, only swapped out, and
131 * that won't affect the overall amount of available
132 * memory in the system.
133 */
134 free -= global_page_state(NR_SHMEM);
1da177e4 135
1da177e4
LT
136 free += nr_swap_pages;
137
138 /*
139 * Any slabs which are created with the
140 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
141 * which are reclaimable, under pressure. The dentry
142 * cache and most inode caches should fall into this
143 */
972d1a7b 144 free += global_page_state(NR_SLAB_RECLAIMABLE);
1da177e4 145
6d9f7839
HA
146 /*
147 * Leave reserved pages. The pages are not for anonymous pages.
148 */
c15bef30 149 if (free <= totalreserve_pages)
6d9f7839
HA
150 goto error;
151 else
c15bef30 152 free -= totalreserve_pages;
6d9f7839
HA
153
154 /*
155 * Leave the last 3% for root
156 */
1da177e4 157 if (!cap_sys_admin)
c15bef30 158 free -= free / 32;
1da177e4
LT
159
160 if (free > pages)
161 return 0;
6d9f7839
HA
162
163 goto error;
1da177e4
LT
164 }
165
166 allowed = (totalram_pages - hugetlb_total_pages())
167 * sysctl_overcommit_ratio / 100;
168 /*
169 * Leave the last 3% for root
170 */
171 if (!cap_sys_admin)
172 allowed -= allowed / 32;
173 allowed += total_swap_pages;
174
175 /* Don't let a single process grow too big:
176 leave 3% of the size of this process for other processes */
731572d3
AC
177 if (mm)
178 allowed -= mm->total_vm / 32;
1da177e4 179
00a62ce9 180 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1da177e4 181 return 0;
6d9f7839 182error:
1da177e4
LT
183 vm_unacct_memory(pages);
184
185 return -ENOMEM;
186}
187
1da177e4 188/*
3d48ae45 189 * Requires inode->i_mapping->i_mmap_mutex
1da177e4
LT
190 */
191static void __remove_shared_vm_struct(struct vm_area_struct *vma,
192 struct file *file, struct address_space *mapping)
193{
194 if (vma->vm_flags & VM_DENYWRITE)
d3ac7f89 195 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
1da177e4
LT
196 if (vma->vm_flags & VM_SHARED)
197 mapping->i_mmap_writable--;
198
199 flush_dcache_mmap_lock(mapping);
200 if (unlikely(vma->vm_flags & VM_NONLINEAR))
201 list_del_init(&vma->shared.vm_set.list);
202 else
203 vma_prio_tree_remove(vma, &mapping->i_mmap);
204 flush_dcache_mmap_unlock(mapping);
205}
206
207/*
a8fb5618
HD
208 * Unlink a file-based vm structure from its prio_tree, to hide
209 * vma from rmap and vmtruncate before freeing its page tables.
1da177e4 210 */
a8fb5618 211void unlink_file_vma(struct vm_area_struct *vma)
1da177e4
LT
212{
213 struct file *file = vma->vm_file;
214
1da177e4
LT
215 if (file) {
216 struct address_space *mapping = file->f_mapping;
3d48ae45 217 mutex_lock(&mapping->i_mmap_mutex);
1da177e4 218 __remove_shared_vm_struct(vma, file, mapping);
3d48ae45 219 mutex_unlock(&mapping->i_mmap_mutex);
1da177e4 220 }
a8fb5618
HD
221}
222
223/*
224 * Close a vm structure and free it, returning the next.
225 */
226static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
227{
228 struct vm_area_struct *next = vma->vm_next;
229
a8fb5618 230 might_sleep();
1da177e4
LT
231 if (vma->vm_ops && vma->vm_ops->close)
232 vma->vm_ops->close(vma);
925d1c40 233 if (vma->vm_file) {
a8fb5618 234 fput(vma->vm_file);
925d1c40
MH
235 if (vma->vm_flags & VM_EXECUTABLE)
236 removed_exe_file_vma(vma->vm_mm);
237 }
f0be3d32 238 mpol_put(vma_policy(vma));
1da177e4 239 kmem_cache_free(vm_area_cachep, vma);
a8fb5618 240 return next;
1da177e4
LT
241}
242
6a6160a7 243SYSCALL_DEFINE1(brk, unsigned long, brk)
1da177e4
LT
244{
245 unsigned long rlim, retval;
246 unsigned long newbrk, oldbrk;
247 struct mm_struct *mm = current->mm;
a5b4592c 248 unsigned long min_brk;
1da177e4
LT
249
250 down_write(&mm->mmap_sem);
251
a5b4592c 252#ifdef CONFIG_COMPAT_BRK
5520e894
JK
253 /*
254 * CONFIG_COMPAT_BRK can still be overridden by setting
255 * randomize_va_space to 2, which will still cause mm->start_brk
256 * to be arbitrarily shifted
257 */
4471a675 258 if (current->brk_randomized)
5520e894
JK
259 min_brk = mm->start_brk;
260 else
261 min_brk = mm->end_data;
a5b4592c
JK
262#else
263 min_brk = mm->start_brk;
264#endif
265 if (brk < min_brk)
1da177e4 266 goto out;
1e624196
RG
267
268 /*
269 * Check against rlimit here. If this check is done later after the test
270 * of oldbrk with newbrk then it can escape the test and let the data
271 * segment grow beyond its set limit the in case where the limit is
272 * not page aligned -Ram Gupta
273 */
59e99e5b 274 rlim = rlimit(RLIMIT_DATA);
c1d171a0
JK
275 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
276 (mm->end_data - mm->start_data) > rlim)
1e624196
RG
277 goto out;
278
1da177e4
LT
279 newbrk = PAGE_ALIGN(brk);
280 oldbrk = PAGE_ALIGN(mm->brk);
281 if (oldbrk == newbrk)
282 goto set_brk;
283
284 /* Always allow shrinking brk. */
285 if (brk <= mm->brk) {
286 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
287 goto set_brk;
288 goto out;
289 }
290
1da177e4
LT
291 /* Check against existing mmap mappings. */
292 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
293 goto out;
294
295 /* Ok, looks good - let it rip. */
296 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
297 goto out;
298set_brk:
299 mm->brk = brk;
300out:
301 retval = mm->brk;
302 up_write(&mm->mmap_sem);
303 return retval;
304}
305
306#ifdef DEBUG_MM_RB
307static int browse_rb(struct rb_root *root)
308{
309 int i = 0, j;
310 struct rb_node *nd, *pn = NULL;
311 unsigned long prev = 0, pend = 0;
312
313 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
314 struct vm_area_struct *vma;
315 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
316 if (vma->vm_start < prev)
317 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
318 if (vma->vm_start < pend)
319 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
320 if (vma->vm_start > vma->vm_end)
321 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
322 i++;
323 pn = nd;
d1af65d1
DM
324 prev = vma->vm_start;
325 pend = vma->vm_end;
1da177e4
LT
326 }
327 j = 0;
328 for (nd = pn; nd; nd = rb_prev(nd)) {
329 j++;
330 }
331 if (i != j)
332 printk("backwards %d, forwards %d\n", j, i), i = 0;
333 return i;
334}
335
336void validate_mm(struct mm_struct *mm)
337{
338 int bug = 0;
339 int i = 0;
340 struct vm_area_struct *tmp = mm->mmap;
341 while (tmp) {
342 tmp = tmp->vm_next;
343 i++;
344 }
345 if (i != mm->map_count)
346 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
347 i = browse_rb(&mm->mm_rb);
348 if (i != mm->map_count)
349 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
46a350ef 350 BUG_ON(bug);
1da177e4
LT
351}
352#else
353#define validate_mm(mm) do { } while (0)
354#endif
355
356static struct vm_area_struct *
357find_vma_prepare(struct mm_struct *mm, unsigned long addr,
358 struct vm_area_struct **pprev, struct rb_node ***rb_link,
359 struct rb_node ** rb_parent)
360{
361 struct vm_area_struct * vma;
362 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
363
364 __rb_link = &mm->mm_rb.rb_node;
365 rb_prev = __rb_parent = NULL;
366 vma = NULL;
367
368 while (*__rb_link) {
369 struct vm_area_struct *vma_tmp;
370
371 __rb_parent = *__rb_link;
372 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
373
374 if (vma_tmp->vm_end > addr) {
375 vma = vma_tmp;
376 if (vma_tmp->vm_start <= addr)
dfe195fb 377 break;
1da177e4
LT
378 __rb_link = &__rb_parent->rb_left;
379 } else {
380 rb_prev = __rb_parent;
381 __rb_link = &__rb_parent->rb_right;
382 }
383 }
384
385 *pprev = NULL;
386 if (rb_prev)
387 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
388 *rb_link = __rb_link;
389 *rb_parent = __rb_parent;
390 return vma;
391}
392
1da177e4
LT
393void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
394 struct rb_node **rb_link, struct rb_node *rb_parent)
395{
396 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
397 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
398}
399
cb8f488c 400static void __vma_link_file(struct vm_area_struct *vma)
1da177e4 401{
48aae425 402 struct file *file;
1da177e4
LT
403
404 file = vma->vm_file;
405 if (file) {
406 struct address_space *mapping = file->f_mapping;
407
408 if (vma->vm_flags & VM_DENYWRITE)
d3ac7f89 409 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
1da177e4
LT
410 if (vma->vm_flags & VM_SHARED)
411 mapping->i_mmap_writable++;
412
413 flush_dcache_mmap_lock(mapping);
414 if (unlikely(vma->vm_flags & VM_NONLINEAR))
415 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
416 else
417 vma_prio_tree_insert(vma, &mapping->i_mmap);
418 flush_dcache_mmap_unlock(mapping);
419 }
420}
421
422static void
423__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
424 struct vm_area_struct *prev, struct rb_node **rb_link,
425 struct rb_node *rb_parent)
426{
427 __vma_link_list(mm, vma, prev, rb_parent);
428 __vma_link_rb(mm, vma, rb_link, rb_parent);
1da177e4
LT
429}
430
431static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
432 struct vm_area_struct *prev, struct rb_node **rb_link,
433 struct rb_node *rb_parent)
434{
435 struct address_space *mapping = NULL;
436
437 if (vma->vm_file)
438 mapping = vma->vm_file->f_mapping;
439
97a89413 440 if (mapping)
3d48ae45 441 mutex_lock(&mapping->i_mmap_mutex);
1da177e4
LT
442
443 __vma_link(mm, vma, prev, rb_link, rb_parent);
444 __vma_link_file(vma);
445
1da177e4 446 if (mapping)
3d48ae45 447 mutex_unlock(&mapping->i_mmap_mutex);
1da177e4
LT
448
449 mm->map_count++;
450 validate_mm(mm);
451}
452
453/*
454 * Helper for vma_adjust in the split_vma insert case:
455 * insert vm structure into list and rbtree and anon_vma,
456 * but it has already been inserted into prio_tree earlier.
457 */
48aae425 458static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 459{
48aae425
Z
460 struct vm_area_struct *__vma, *prev;
461 struct rb_node **rb_link, *rb_parent;
1da177e4
LT
462
463 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
46a350ef 464 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
1da177e4
LT
465 __vma_link(mm, vma, prev, rb_link, rb_parent);
466 mm->map_count++;
467}
468
469static inline void
470__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
471 struct vm_area_struct *prev)
472{
297c5eee
LT
473 struct vm_area_struct *next = vma->vm_next;
474
475 prev->vm_next = next;
476 if (next)
477 next->vm_prev = prev;
1da177e4
LT
478 rb_erase(&vma->vm_rb, &mm->mm_rb);
479 if (mm->mmap_cache == vma)
480 mm->mmap_cache = prev;
481}
482
483/*
484 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
485 * is already present in an i_mmap tree without adjusting the tree.
486 * The following helper function should be used when such adjustments
487 * are necessary. The "insert" vma (if any) is to be inserted
488 * before we drop the necessary locks.
489 */
5beb4930 490int vma_adjust(struct vm_area_struct *vma, unsigned long start,
1da177e4
LT
491 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
492{
493 struct mm_struct *mm = vma->vm_mm;
494 struct vm_area_struct *next = vma->vm_next;
495 struct vm_area_struct *importer = NULL;
496 struct address_space *mapping = NULL;
497 struct prio_tree_root *root = NULL;
012f1800 498 struct anon_vma *anon_vma = NULL;
1da177e4 499 struct file *file = vma->vm_file;
1da177e4
LT
500 long adjust_next = 0;
501 int remove_next = 0;
502
503 if (next && !insert) {
287d97ac
LT
504 struct vm_area_struct *exporter = NULL;
505
1da177e4
LT
506 if (end >= next->vm_end) {
507 /*
508 * vma expands, overlapping all the next, and
509 * perhaps the one after too (mprotect case 6).
510 */
511again: remove_next = 1 + (end > next->vm_end);
512 end = next->vm_end;
287d97ac 513 exporter = next;
1da177e4
LT
514 importer = vma;
515 } else if (end > next->vm_start) {
516 /*
517 * vma expands, overlapping part of the next:
518 * mprotect case 5 shifting the boundary up.
519 */
520 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
287d97ac 521 exporter = next;
1da177e4
LT
522 importer = vma;
523 } else if (end < vma->vm_end) {
524 /*
525 * vma shrinks, and !insert tells it's not
526 * split_vma inserting another: so it must be
527 * mprotect case 4 shifting the boundary down.
528 */
529 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
287d97ac 530 exporter = vma;
1da177e4
LT
531 importer = next;
532 }
1da177e4 533
5beb4930
RR
534 /*
535 * Easily overlooked: when mprotect shifts the boundary,
536 * make sure the expanding vma has anon_vma set if the
537 * shrinking vma had, to cover any anon pages imported.
538 */
287d97ac
LT
539 if (exporter && exporter->anon_vma && !importer->anon_vma) {
540 if (anon_vma_clone(importer, exporter))
5beb4930 541 return -ENOMEM;
287d97ac 542 importer->anon_vma = exporter->anon_vma;
5beb4930
RR
543 }
544 }
545
1da177e4
LT
546 if (file) {
547 mapping = file->f_mapping;
548 if (!(vma->vm_flags & VM_NONLINEAR))
549 root = &mapping->i_mmap;
3d48ae45 550 mutex_lock(&mapping->i_mmap_mutex);
1da177e4 551 if (insert) {
1da177e4
LT
552 /*
553 * Put into prio_tree now, so instantiated pages
554 * are visible to arm/parisc __flush_dcache_page
555 * throughout; but we cannot insert into address
556 * space until vma start or end is updated.
557 */
558 __vma_link_file(insert);
559 }
560 }
561
94fcc585
AA
562 vma_adjust_trans_huge(vma, start, end, adjust_next);
563
012f1800
RR
564 /*
565 * When changing only vma->vm_end, we don't really need anon_vma
566 * lock. This is a fairly rare case by itself, but the anon_vma
567 * lock may be shared between many sibling processes. Skipping
568 * the lock for brk adjustments makes a difference sometimes.
569 */
5f70b962 570 if (vma->anon_vma && (importer || start != vma->vm_start)) {
012f1800
RR
571 anon_vma = vma->anon_vma;
572 anon_vma_lock(anon_vma);
573 }
574
1da177e4
LT
575 if (root) {
576 flush_dcache_mmap_lock(mapping);
577 vma_prio_tree_remove(vma, root);
578 if (adjust_next)
579 vma_prio_tree_remove(next, root);
580 }
581
582 vma->vm_start = start;
583 vma->vm_end = end;
584 vma->vm_pgoff = pgoff;
585 if (adjust_next) {
586 next->vm_start += adjust_next << PAGE_SHIFT;
587 next->vm_pgoff += adjust_next;
588 }
589
590 if (root) {
591 if (adjust_next)
592 vma_prio_tree_insert(next, root);
593 vma_prio_tree_insert(vma, root);
594 flush_dcache_mmap_unlock(mapping);
595 }
596
597 if (remove_next) {
598 /*
599 * vma_merge has merged next into vma, and needs
600 * us to remove next before dropping the locks.
601 */
602 __vma_unlink(mm, next, vma);
603 if (file)
604 __remove_shared_vm_struct(next, file, mapping);
1da177e4
LT
605 } else if (insert) {
606 /*
607 * split_vma has split insert from vma, and needs
608 * us to insert it before dropping the locks
609 * (it may either follow vma or precede it).
610 */
611 __insert_vm_struct(mm, insert);
612 }
613
012f1800
RR
614 if (anon_vma)
615 anon_vma_unlock(anon_vma);
1da177e4 616 if (mapping)
3d48ae45 617 mutex_unlock(&mapping->i_mmap_mutex);
1da177e4
LT
618
619 if (remove_next) {
925d1c40 620 if (file) {
1da177e4 621 fput(file);
925d1c40
MH
622 if (next->vm_flags & VM_EXECUTABLE)
623 removed_exe_file_vma(mm);
624 }
5beb4930
RR
625 if (next->anon_vma)
626 anon_vma_merge(vma, next);
1da177e4 627 mm->map_count--;
f0be3d32 628 mpol_put(vma_policy(next));
1da177e4
LT
629 kmem_cache_free(vm_area_cachep, next);
630 /*
631 * In mprotect's case 6 (see comments on vma_merge),
632 * we must remove another next too. It would clutter
633 * up the code too much to do both in one go.
634 */
635 if (remove_next == 2) {
636 next = vma->vm_next;
637 goto again;
638 }
639 }
640
641 validate_mm(mm);
5beb4930
RR
642
643 return 0;
1da177e4
LT
644}
645
646/*
647 * If the vma has a ->close operation then the driver probably needs to release
648 * per-vma resources, so we don't attempt to merge those.
649 */
1da177e4
LT
650static inline int is_mergeable_vma(struct vm_area_struct *vma,
651 struct file *file, unsigned long vm_flags)
652{
8314c4f2
HD
653 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
654 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
1da177e4
LT
655 return 0;
656 if (vma->vm_file != file)
657 return 0;
658 if (vma->vm_ops && vma->vm_ops->close)
659 return 0;
660 return 1;
661}
662
663static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
965f55de
SL
664 struct anon_vma *anon_vma2,
665 struct vm_area_struct *vma)
1da177e4 666{
965f55de
SL
667 /*
668 * The list_is_singular() test is to avoid merging VMA cloned from
669 * parents. This can improve scalability caused by anon_vma lock.
670 */
671 if ((!anon_vma1 || !anon_vma2) && (!vma ||
672 list_is_singular(&vma->anon_vma_chain)))
673 return 1;
674 return anon_vma1 == anon_vma2;
1da177e4
LT
675}
676
677/*
678 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
679 * in front of (at a lower virtual address and file offset than) the vma.
680 *
681 * We cannot merge two vmas if they have differently assigned (non-NULL)
682 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
683 *
684 * We don't check here for the merged mmap wrapping around the end of pagecache
685 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
686 * wrap, nor mmaps which cover the final page at index -1UL.
687 */
688static int
689can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
690 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
691{
692 if (is_mergeable_vma(vma, file, vm_flags) &&
965f55de 693 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1da177e4
LT
694 if (vma->vm_pgoff == vm_pgoff)
695 return 1;
696 }
697 return 0;
698}
699
700/*
701 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
702 * beyond (at a higher virtual address and file offset than) the vma.
703 *
704 * We cannot merge two vmas if they have differently assigned (non-NULL)
705 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
706 */
707static int
708can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
709 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
710{
711 if (is_mergeable_vma(vma, file, vm_flags) &&
965f55de 712 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1da177e4
LT
713 pgoff_t vm_pglen;
714 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
715 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
716 return 1;
717 }
718 return 0;
719}
720
721/*
722 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
723 * whether that can be merged with its predecessor or its successor.
724 * Or both (it neatly fills a hole).
725 *
726 * In most cases - when called for mmap, brk or mremap - [addr,end) is
727 * certain not to be mapped by the time vma_merge is called; but when
728 * called for mprotect, it is certain to be already mapped (either at
729 * an offset within prev, or at the start of next), and the flags of
730 * this area are about to be changed to vm_flags - and the no-change
731 * case has already been eliminated.
732 *
733 * The following mprotect cases have to be considered, where AAAA is
734 * the area passed down from mprotect_fixup, never extending beyond one
735 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
736 *
737 * AAAA AAAA AAAA AAAA
738 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
739 * cannot merge might become might become might become
740 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
741 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
742 * mremap move: PPPPNNNNNNNN 8
743 * AAAA
744 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
745 * might become case 1 below case 2 below case 3 below
746 *
747 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
748 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
749 */
750struct vm_area_struct *vma_merge(struct mm_struct *mm,
751 struct vm_area_struct *prev, unsigned long addr,
752 unsigned long end, unsigned long vm_flags,
753 struct anon_vma *anon_vma, struct file *file,
754 pgoff_t pgoff, struct mempolicy *policy)
755{
756 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
757 struct vm_area_struct *area, *next;
5beb4930 758 int err;
1da177e4
LT
759
760 /*
761 * We later require that vma->vm_flags == vm_flags,
762 * so this tests vma->vm_flags & VM_SPECIAL, too.
763 */
764 if (vm_flags & VM_SPECIAL)
765 return NULL;
766
767 if (prev)
768 next = prev->vm_next;
769 else
770 next = mm->mmap;
771 area = next;
772 if (next && next->vm_end == end) /* cases 6, 7, 8 */
773 next = next->vm_next;
774
775 /*
776 * Can it merge with the predecessor?
777 */
778 if (prev && prev->vm_end == addr &&
779 mpol_equal(vma_policy(prev), policy) &&
780 can_vma_merge_after(prev, vm_flags,
781 anon_vma, file, pgoff)) {
782 /*
783 * OK, it can. Can we now merge in the successor as well?
784 */
785 if (next && end == next->vm_start &&
786 mpol_equal(policy, vma_policy(next)) &&
787 can_vma_merge_before(next, vm_flags,
788 anon_vma, file, pgoff+pglen) &&
789 is_mergeable_anon_vma(prev->anon_vma,
965f55de 790 next->anon_vma, NULL)) {
1da177e4 791 /* cases 1, 6 */
5beb4930 792 err = vma_adjust(prev, prev->vm_start,
1da177e4
LT
793 next->vm_end, prev->vm_pgoff, NULL);
794 } else /* cases 2, 5, 7 */
5beb4930 795 err = vma_adjust(prev, prev->vm_start,
1da177e4 796 end, prev->vm_pgoff, NULL);
5beb4930
RR
797 if (err)
798 return NULL;
b15d00b6 799 khugepaged_enter_vma_merge(prev);
1da177e4
LT
800 return prev;
801 }
802
803 /*
804 * Can this new request be merged in front of next?
805 */
806 if (next && end == next->vm_start &&
807 mpol_equal(policy, vma_policy(next)) &&
808 can_vma_merge_before(next, vm_flags,
809 anon_vma, file, pgoff+pglen)) {
810 if (prev && addr < prev->vm_end) /* case 4 */
5beb4930 811 err = vma_adjust(prev, prev->vm_start,
1da177e4
LT
812 addr, prev->vm_pgoff, NULL);
813 else /* cases 3, 8 */
5beb4930 814 err = vma_adjust(area, addr, next->vm_end,
1da177e4 815 next->vm_pgoff - pglen, NULL);
5beb4930
RR
816 if (err)
817 return NULL;
b15d00b6 818 khugepaged_enter_vma_merge(area);
1da177e4
LT
819 return area;
820 }
821
822 return NULL;
823}
824
d0e9fe17
LT
825/*
826 * Rough compatbility check to quickly see if it's even worth looking
827 * at sharing an anon_vma.
828 *
829 * They need to have the same vm_file, and the flags can only differ
830 * in things that mprotect may change.
831 *
832 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
833 * we can merge the two vma's. For example, we refuse to merge a vma if
834 * there is a vm_ops->close() function, because that indicates that the
835 * driver is doing some kind of reference counting. But that doesn't
836 * really matter for the anon_vma sharing case.
837 */
838static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
839{
840 return a->vm_end == b->vm_start &&
841 mpol_equal(vma_policy(a), vma_policy(b)) &&
842 a->vm_file == b->vm_file &&
843 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
844 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
845}
846
847/*
848 * Do some basic sanity checking to see if we can re-use the anon_vma
849 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
850 * the same as 'old', the other will be the new one that is trying
851 * to share the anon_vma.
852 *
853 * NOTE! This runs with mm_sem held for reading, so it is possible that
854 * the anon_vma of 'old' is concurrently in the process of being set up
855 * by another page fault trying to merge _that_. But that's ok: if it
856 * is being set up, that automatically means that it will be a singleton
857 * acceptable for merging, so we can do all of this optimistically. But
858 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
859 *
860 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
861 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
862 * is to return an anon_vma that is "complex" due to having gone through
863 * a fork).
864 *
865 * We also make sure that the two vma's are compatible (adjacent,
866 * and with the same memory policies). That's all stable, even with just
867 * a read lock on the mm_sem.
868 */
869static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
870{
871 if (anon_vma_compatible(a, b)) {
872 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
873
874 if (anon_vma && list_is_singular(&old->anon_vma_chain))
875 return anon_vma;
876 }
877 return NULL;
878}
879
1da177e4
LT
880/*
881 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
882 * neighbouring vmas for a suitable anon_vma, before it goes off
883 * to allocate a new anon_vma. It checks because a repetitive
884 * sequence of mprotects and faults may otherwise lead to distinct
885 * anon_vmas being allocated, preventing vma merge in subsequent
886 * mprotect.
887 */
888struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
889{
d0e9fe17 890 struct anon_vma *anon_vma;
1da177e4 891 struct vm_area_struct *near;
1da177e4
LT
892
893 near = vma->vm_next;
894 if (!near)
895 goto try_prev;
896
d0e9fe17
LT
897 anon_vma = reusable_anon_vma(near, vma, near);
898 if (anon_vma)
899 return anon_vma;
1da177e4 900try_prev:
9be34c9d 901 near = vma->vm_prev;
1da177e4
LT
902 if (!near)
903 goto none;
904
d0e9fe17
LT
905 anon_vma = reusable_anon_vma(near, near, vma);
906 if (anon_vma)
907 return anon_vma;
1da177e4
LT
908none:
909 /*
910 * There's no absolute need to look only at touching neighbours:
911 * we could search further afield for "compatible" anon_vmas.
912 * But it would probably just be a waste of time searching,
913 * or lead to too many vmas hanging off the same anon_vma.
914 * We're trying to allow mprotect remerging later on,
915 * not trying to minimize memory used for anon_vmas.
916 */
917 return NULL;
918}
919
920#ifdef CONFIG_PROC_FS
ab50b8ed 921void vm_stat_account(struct mm_struct *mm, unsigned long flags,
1da177e4
LT
922 struct file *file, long pages)
923{
924 const unsigned long stack_flags
925 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
926
1da177e4
LT
927 if (file) {
928 mm->shared_vm += pages;
929 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
930 mm->exec_vm += pages;
931 } else if (flags & stack_flags)
932 mm->stack_vm += pages;
933 if (flags & (VM_RESERVED|VM_IO))
934 mm->reserved_vm += pages;
935}
936#endif /* CONFIG_PROC_FS */
937
40401530
AV
938/*
939 * If a hint addr is less than mmap_min_addr change hint to be as
940 * low as possible but still greater than mmap_min_addr
941 */
942static inline unsigned long round_hint_to_min(unsigned long hint)
943{
944 hint &= PAGE_MASK;
945 if (((void *)hint != NULL) &&
946 (hint < mmap_min_addr))
947 return PAGE_ALIGN(mmap_min_addr);
948 return hint;
949}
950
1da177e4 951/*
27f5de79 952 * The caller must hold down_write(&current->mm->mmap_sem).
1da177e4
LT
953 */
954
48aae425 955unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
1da177e4
LT
956 unsigned long len, unsigned long prot,
957 unsigned long flags, unsigned long pgoff)
958{
959 struct mm_struct * mm = current->mm;
1da177e4 960 struct inode *inode;
ca16d140 961 vm_flags_t vm_flags;
1da177e4 962 int error;
0165ab44 963 unsigned long reqprot = prot;
1da177e4 964
1da177e4
LT
965 /*
966 * Does the application expect PROT_READ to imply PROT_EXEC?
967 *
968 * (the exception is when the underlying filesystem is noexec
969 * mounted, in which case we dont add PROT_EXEC.)
970 */
971 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
d3ac7f89 972 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
1da177e4
LT
973 prot |= PROT_EXEC;
974
975 if (!len)
976 return -EINVAL;
977
7cd94146
EP
978 if (!(flags & MAP_FIXED))
979 addr = round_hint_to_min(addr);
980
1da177e4
LT
981 /* Careful about overflows.. */
982 len = PAGE_ALIGN(len);
9206de95 983 if (!len)
1da177e4
LT
984 return -ENOMEM;
985
986 /* offset overflow? */
987 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
988 return -EOVERFLOW;
989
990 /* Too many mappings? */
991 if (mm->map_count > sysctl_max_map_count)
992 return -ENOMEM;
993
994 /* Obtain the address to map to. we verify (or select) it and ensure
995 * that it represents a valid section of the address space.
996 */
997 addr = get_unmapped_area(file, addr, len, pgoff, flags);
998 if (addr & ~PAGE_MASK)
999 return addr;
1000
1001 /* Do simple checking here so the lower-level routines won't have
1002 * to. we assume access permissions have been handled by the open
1003 * of the memory object, so we don't do any here.
1004 */
1005 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1006 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1007
cdf7b341 1008 if (flags & MAP_LOCKED)
1da177e4
LT
1009 if (!can_do_mlock())
1010 return -EPERM;
ba470de4 1011
1da177e4
LT
1012 /* mlock MCL_FUTURE? */
1013 if (vm_flags & VM_LOCKED) {
1014 unsigned long locked, lock_limit;
93ea1d0a
CW
1015 locked = len >> PAGE_SHIFT;
1016 locked += mm->locked_vm;
59e99e5b 1017 lock_limit = rlimit(RLIMIT_MEMLOCK);
93ea1d0a 1018 lock_limit >>= PAGE_SHIFT;
1da177e4
LT
1019 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1020 return -EAGAIN;
1021 }
1022
d3ac7f89 1023 inode = file ? file->f_path.dentry->d_inode : NULL;
1da177e4
LT
1024
1025 if (file) {
1026 switch (flags & MAP_TYPE) {
1027 case MAP_SHARED:
1028 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1029 return -EACCES;
1030
1031 /*
1032 * Make sure we don't allow writing to an append-only
1033 * file..
1034 */
1035 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1036 return -EACCES;
1037
1038 /*
1039 * Make sure there are no mandatory locks on the file.
1040 */
1041 if (locks_verify_locked(inode))
1042 return -EAGAIN;
1043
1044 vm_flags |= VM_SHARED | VM_MAYSHARE;
1045 if (!(file->f_mode & FMODE_WRITE))
1046 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1047
1048 /* fall through */
1049 case MAP_PRIVATE:
1050 if (!(file->f_mode & FMODE_READ))
1051 return -EACCES;
d3ac7f89 1052 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
80c5606c
LT
1053 if (vm_flags & VM_EXEC)
1054 return -EPERM;
1055 vm_flags &= ~VM_MAYEXEC;
1056 }
80c5606c
LT
1057
1058 if (!file->f_op || !file->f_op->mmap)
1059 return -ENODEV;
1da177e4
LT
1060 break;
1061
1062 default:
1063 return -EINVAL;
1064 }
1065 } else {
1066 switch (flags & MAP_TYPE) {
1067 case MAP_SHARED:
ce363942
TH
1068 /*
1069 * Ignore pgoff.
1070 */
1071 pgoff = 0;
1da177e4
LT
1072 vm_flags |= VM_SHARED | VM_MAYSHARE;
1073 break;
1074 case MAP_PRIVATE:
1075 /*
1076 * Set pgoff according to addr for anon_vma.
1077 */
1078 pgoff = addr >> PAGE_SHIFT;
1079 break;
1080 default:
1081 return -EINVAL;
1082 }
1083 }
1084
ed032189 1085 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1da177e4
LT
1086 if (error)
1087 return error;
ed032189 1088
5a6fe125 1089 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
0165ab44
MS
1090}
1091EXPORT_SYMBOL(do_mmap_pgoff);
1092
66f0dc48
HD
1093SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1094 unsigned long, prot, unsigned long, flags,
1095 unsigned long, fd, unsigned long, pgoff)
1096{
1097 struct file *file = NULL;
1098 unsigned long retval = -EBADF;
1099
1100 if (!(flags & MAP_ANONYMOUS)) {
120a795d 1101 audit_mmap_fd(fd, flags);
66f0dc48
HD
1102 if (unlikely(flags & MAP_HUGETLB))
1103 return -EINVAL;
1104 file = fget(fd);
1105 if (!file)
1106 goto out;
1107 } else if (flags & MAP_HUGETLB) {
1108 struct user_struct *user = NULL;
1109 /*
1110 * VM_NORESERVE is used because the reservations will be
1111 * taken when vm_ops->mmap() is called
1112 * A dummy user value is used because we are not locking
1113 * memory so no accounting is necessary
1114 */
1115 len = ALIGN(len, huge_page_size(&default_hstate));
1116 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1117 &user, HUGETLB_ANONHUGE_INODE);
1118 if (IS_ERR(file))
1119 return PTR_ERR(file);
1120 }
1121
1122 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1123
1124 down_write(&current->mm->mmap_sem);
1125 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1126 up_write(&current->mm->mmap_sem);
1127
1128 if (file)
1129 fput(file);
1130out:
1131 return retval;
1132}
1133
a4679373
CH
1134#ifdef __ARCH_WANT_SYS_OLD_MMAP
1135struct mmap_arg_struct {
1136 unsigned long addr;
1137 unsigned long len;
1138 unsigned long prot;
1139 unsigned long flags;
1140 unsigned long fd;
1141 unsigned long offset;
1142};
1143
1144SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1145{
1146 struct mmap_arg_struct a;
1147
1148 if (copy_from_user(&a, arg, sizeof(a)))
1149 return -EFAULT;
1150 if (a.offset & ~PAGE_MASK)
1151 return -EINVAL;
1152
1153 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1154 a.offset >> PAGE_SHIFT);
1155}
1156#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1157
4e950f6f
AD
1158/*
1159 * Some shared mappigns will want the pages marked read-only
1160 * to track write events. If so, we'll downgrade vm_page_prot
1161 * to the private version (using protection_map[] without the
1162 * VM_SHARED bit).
1163 */
1164int vma_wants_writenotify(struct vm_area_struct *vma)
1165{
ca16d140 1166 vm_flags_t vm_flags = vma->vm_flags;
4e950f6f
AD
1167
1168 /* If it was private or non-writable, the write bit is already clear */
1169 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1170 return 0;
1171
1172 /* The backer wishes to know when pages are first written to? */
1173 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1174 return 1;
1175
1176 /* The open routine did something to the protections already? */
1177 if (pgprot_val(vma->vm_page_prot) !=
3ed75eb8 1178 pgprot_val(vm_get_page_prot(vm_flags)))
4e950f6f
AD
1179 return 0;
1180
1181 /* Specialty mapping? */
1182 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1183 return 0;
1184
1185 /* Can the mapping track the dirty pages? */
1186 return vma->vm_file && vma->vm_file->f_mapping &&
1187 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1188}
1189
fc8744ad
LT
1190/*
1191 * We account for memory if it's a private writeable mapping,
5a6fe125 1192 * not hugepages and VM_NORESERVE wasn't set.
fc8744ad 1193 */
ca16d140 1194static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
fc8744ad 1195{
5a6fe125
MG
1196 /*
1197 * hugetlb has its own accounting separate from the core VM
1198 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1199 */
1200 if (file && is_file_hugepages(file))
1201 return 0;
1202
fc8744ad
LT
1203 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1204}
1205
0165ab44
MS
1206unsigned long mmap_region(struct file *file, unsigned long addr,
1207 unsigned long len, unsigned long flags,
ca16d140 1208 vm_flags_t vm_flags, unsigned long pgoff)
0165ab44
MS
1209{
1210 struct mm_struct *mm = current->mm;
1211 struct vm_area_struct *vma, *prev;
1212 int correct_wcount = 0;
1213 int error;
1214 struct rb_node **rb_link, *rb_parent;
1215 unsigned long charged = 0;
1216 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1217
1da177e4
LT
1218 /* Clear old maps */
1219 error = -ENOMEM;
1220munmap_back:
1221 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1222 if (vma && vma->vm_start < addr + len) {
1223 if (do_munmap(mm, addr, len))
1224 return -ENOMEM;
1225 goto munmap_back;
1226 }
1227
1228 /* Check against address space limit. */
119f657c 1229 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1da177e4
LT
1230 return -ENOMEM;
1231
fc8744ad
LT
1232 /*
1233 * Set 'VM_NORESERVE' if we should not account for the
5a6fe125 1234 * memory use of this mapping.
fc8744ad 1235 */
5a6fe125
MG
1236 if ((flags & MAP_NORESERVE)) {
1237 /* We honor MAP_NORESERVE if allowed to overcommit */
1238 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1239 vm_flags |= VM_NORESERVE;
1240
1241 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1242 if (file && is_file_hugepages(file))
1243 vm_flags |= VM_NORESERVE;
1244 }
cdfd4325 1245
fc8744ad
LT
1246 /*
1247 * Private writable mapping: check memory availability
1248 */
5a6fe125 1249 if (accountable_mapping(file, vm_flags)) {
fc8744ad 1250 charged = len >> PAGE_SHIFT;
191c5424 1251 if (security_vm_enough_memory_mm(mm, charged))
fc8744ad
LT
1252 return -ENOMEM;
1253 vm_flags |= VM_ACCOUNT;
1da177e4
LT
1254 }
1255
1256 /*
de33c8db 1257 * Can we just expand an old mapping?
1da177e4 1258 */
de33c8db
LT
1259 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1260 if (vma)
1261 goto out;
1da177e4
LT
1262
1263 /*
1264 * Determine the object being mapped and call the appropriate
1265 * specific mapper. the address has already been validated, but
1266 * not unmapped, but the maps are removed from the list.
1267 */
c5e3b83e 1268 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1da177e4
LT
1269 if (!vma) {
1270 error = -ENOMEM;
1271 goto unacct_error;
1272 }
1da177e4
LT
1273
1274 vma->vm_mm = mm;
1275 vma->vm_start = addr;
1276 vma->vm_end = addr + len;
1277 vma->vm_flags = vm_flags;
3ed75eb8 1278 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1da177e4 1279 vma->vm_pgoff = pgoff;
5beb4930 1280 INIT_LIST_HEAD(&vma->anon_vma_chain);
1da177e4
LT
1281
1282 if (file) {
1283 error = -EINVAL;
1284 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1285 goto free_vma;
1286 if (vm_flags & VM_DENYWRITE) {
1287 error = deny_write_access(file);
1288 if (error)
1289 goto free_vma;
1290 correct_wcount = 1;
1291 }
1292 vma->vm_file = file;
1293 get_file(file);
1294 error = file->f_op->mmap(file, vma);
1295 if (error)
1296 goto unmap_and_free_vma;
925d1c40
MH
1297 if (vm_flags & VM_EXECUTABLE)
1298 added_exe_file_vma(mm);
f8dbf0a7
HS
1299
1300 /* Can addr have changed??
1301 *
1302 * Answer: Yes, several device drivers can do it in their
1303 * f_op->mmap method. -DaveM
1304 */
1305 addr = vma->vm_start;
1306 pgoff = vma->vm_pgoff;
1307 vm_flags = vma->vm_flags;
1da177e4
LT
1308 } else if (vm_flags & VM_SHARED) {
1309 error = shmem_zero_setup(vma);
1310 if (error)
1311 goto free_vma;
1312 }
1313
c9d0bf24
MD
1314 if (vma_wants_writenotify(vma)) {
1315 pgprot_t pprot = vma->vm_page_prot;
1316
1317 /* Can vma->vm_page_prot have changed??
1318 *
1319 * Answer: Yes, drivers may have changed it in their
1320 * f_op->mmap method.
1321 *
1322 * Ensures that vmas marked as uncached stay that way.
1323 */
1ddd439e 1324 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
c9d0bf24
MD
1325 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1326 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1327 }
d08b3851 1328
de33c8db
LT
1329 vma_link(mm, vma, prev, rb_link, rb_parent);
1330 file = vma->vm_file;
4d3d5b41
ON
1331
1332 /* Once vma denies write, undo our temporary denial count */
1333 if (correct_wcount)
1334 atomic_inc(&inode->i_writecount);
1335out:
cdd6c482 1336 perf_event_mmap(vma);
0a4a9391 1337
1da177e4 1338 mm->total_vm += len >> PAGE_SHIFT;
ab50b8ed 1339 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1da177e4 1340 if (vm_flags & VM_LOCKED) {
06f9d8c2
KM
1341 if (!mlock_vma_pages_range(vma, addr, addr + len))
1342 mm->locked_vm += (len >> PAGE_SHIFT);
ba470de4 1343 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
54cb8821 1344 make_pages_present(addr, addr + len);
1da177e4
LT
1345 return addr;
1346
1347unmap_and_free_vma:
1348 if (correct_wcount)
1349 atomic_inc(&inode->i_writecount);
1350 vma->vm_file = NULL;
1351 fput(file);
1352
1353 /* Undo any partial mapping done by a device driver. */
e0da382c
HD
1354 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1355 charged = 0;
1da177e4
LT
1356free_vma:
1357 kmem_cache_free(vm_area_cachep, vma);
1358unacct_error:
1359 if (charged)
1360 vm_unacct_memory(charged);
1361 return error;
1362}
1363
1da177e4
LT
1364/* Get an address range which is currently unmapped.
1365 * For shmat() with addr=0.
1366 *
1367 * Ugly calling convention alert:
1368 * Return value with the low bits set means error value,
1369 * ie
1370 * if (ret & ~PAGE_MASK)
1371 * error = ret;
1372 *
1373 * This function "knows" that -ENOMEM has the bits set.
1374 */
1375#ifndef HAVE_ARCH_UNMAPPED_AREA
1376unsigned long
1377arch_get_unmapped_area(struct file *filp, unsigned long addr,
1378 unsigned long len, unsigned long pgoff, unsigned long flags)
1379{
1380 struct mm_struct *mm = current->mm;
1381 struct vm_area_struct *vma;
1382 unsigned long start_addr;
1383
1384 if (len > TASK_SIZE)
1385 return -ENOMEM;
1386
06abdfb4
BH
1387 if (flags & MAP_FIXED)
1388 return addr;
1389
1da177e4
LT
1390 if (addr) {
1391 addr = PAGE_ALIGN(addr);
1392 vma = find_vma(mm, addr);
1393 if (TASK_SIZE - len >= addr &&
1394 (!vma || addr + len <= vma->vm_start))
1395 return addr;
1396 }
1363c3cd
WW
1397 if (len > mm->cached_hole_size) {
1398 start_addr = addr = mm->free_area_cache;
1399 } else {
1400 start_addr = addr = TASK_UNMAPPED_BASE;
1401 mm->cached_hole_size = 0;
1402 }
1da177e4
LT
1403
1404full_search:
1405 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1406 /* At this point: (!vma || addr < vma->vm_end). */
1407 if (TASK_SIZE - len < addr) {
1408 /*
1409 * Start a new search - just in case we missed
1410 * some holes.
1411 */
1412 if (start_addr != TASK_UNMAPPED_BASE) {
1363c3cd
WW
1413 addr = TASK_UNMAPPED_BASE;
1414 start_addr = addr;
1415 mm->cached_hole_size = 0;
1da177e4
LT
1416 goto full_search;
1417 }
1418 return -ENOMEM;
1419 }
1420 if (!vma || addr + len <= vma->vm_start) {
1421 /*
1422 * Remember the place where we stopped the search:
1423 */
1424 mm->free_area_cache = addr + len;
1425 return addr;
1426 }
1363c3cd
WW
1427 if (addr + mm->cached_hole_size < vma->vm_start)
1428 mm->cached_hole_size = vma->vm_start - addr;
1da177e4
LT
1429 addr = vma->vm_end;
1430 }
1431}
1432#endif
1433
1363c3cd 1434void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1da177e4
LT
1435{
1436 /*
1437 * Is this a new hole at the lowest possible address?
1438 */
1363c3cd
WW
1439 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1440 mm->free_area_cache = addr;
1441 mm->cached_hole_size = ~0UL;
1442 }
1da177e4
LT
1443}
1444
1445/*
1446 * This mmap-allocator allocates new areas top-down from below the
1447 * stack's low limit (the base):
1448 */
1449#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1450unsigned long
1451arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1452 const unsigned long len, const unsigned long pgoff,
1453 const unsigned long flags)
1454{
1455 struct vm_area_struct *vma;
1456 struct mm_struct *mm = current->mm;
1457 unsigned long addr = addr0;
1458
1459 /* requested length too big for entire address space */
1460 if (len > TASK_SIZE)
1461 return -ENOMEM;
1462
06abdfb4
BH
1463 if (flags & MAP_FIXED)
1464 return addr;
1465
1da177e4
LT
1466 /* requesting a specific address */
1467 if (addr) {
1468 addr = PAGE_ALIGN(addr);
1469 vma = find_vma(mm, addr);
1470 if (TASK_SIZE - len >= addr &&
1471 (!vma || addr + len <= vma->vm_start))
1472 return addr;
1473 }
1474
1363c3cd
WW
1475 /* check if free_area_cache is useful for us */
1476 if (len <= mm->cached_hole_size) {
1477 mm->cached_hole_size = 0;
1478 mm->free_area_cache = mm->mmap_base;
1479 }
1480
1da177e4
LT
1481 /* either no address requested or can't fit in requested address hole */
1482 addr = mm->free_area_cache;
1483
1484 /* make sure it can fit in the remaining address space */
49a43876 1485 if (addr > len) {
1da177e4
LT
1486 vma = find_vma(mm, addr-len);
1487 if (!vma || addr <= vma->vm_start)
1488 /* remember the address as a hint for next time */
1489 return (mm->free_area_cache = addr-len);
1490 }
1491
73219d17
CW
1492 if (mm->mmap_base < len)
1493 goto bottomup;
1494
1da177e4
LT
1495 addr = mm->mmap_base-len;
1496
1497 do {
1498 /*
1499 * Lookup failure means no vma is above this address,
1500 * else if new region fits below vma->vm_start,
1501 * return with success:
1502 */
1503 vma = find_vma(mm, addr);
1504 if (!vma || addr+len <= vma->vm_start)
1505 /* remember the address as a hint for next time */
1506 return (mm->free_area_cache = addr);
1507
1363c3cd
WW
1508 /* remember the largest hole we saw so far */
1509 if (addr + mm->cached_hole_size < vma->vm_start)
1510 mm->cached_hole_size = vma->vm_start - addr;
1511
1da177e4
LT
1512 /* try just below the current vma->vm_start */
1513 addr = vma->vm_start-len;
49a43876 1514 } while (len < vma->vm_start);
1da177e4 1515
73219d17 1516bottomup:
1da177e4
LT
1517 /*
1518 * A failed mmap() very likely causes application failure,
1519 * so fall back to the bottom-up function here. This scenario
1520 * can happen with large stack limits and large mmap()
1521 * allocations.
1522 */
1363c3cd
WW
1523 mm->cached_hole_size = ~0UL;
1524 mm->free_area_cache = TASK_UNMAPPED_BASE;
1da177e4
LT
1525 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1526 /*
1527 * Restore the topdown base:
1528 */
1529 mm->free_area_cache = mm->mmap_base;
1363c3cd 1530 mm->cached_hole_size = ~0UL;
1da177e4
LT
1531
1532 return addr;
1533}
1534#endif
1535
1363c3cd 1536void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1da177e4
LT
1537{
1538 /*
1539 * Is this a new hole at the highest possible address?
1540 */
1363c3cd
WW
1541 if (addr > mm->free_area_cache)
1542 mm->free_area_cache = addr;
1da177e4
LT
1543
1544 /* dont allow allocations above current base */
1363c3cd
WW
1545 if (mm->free_area_cache > mm->mmap_base)
1546 mm->free_area_cache = mm->mmap_base;
1da177e4
LT
1547}
1548
1549unsigned long
1550get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1551 unsigned long pgoff, unsigned long flags)
1552{
06abdfb4
BH
1553 unsigned long (*get_area)(struct file *, unsigned long,
1554 unsigned long, unsigned long, unsigned long);
1555
9206de95
AV
1556 unsigned long error = arch_mmap_check(addr, len, flags);
1557 if (error)
1558 return error;
1559
1560 /* Careful about overflows.. */
1561 if (len > TASK_SIZE)
1562 return -ENOMEM;
1563
06abdfb4
BH
1564 get_area = current->mm->get_unmapped_area;
1565 if (file && file->f_op && file->f_op->get_unmapped_area)
1566 get_area = file->f_op->get_unmapped_area;
1567 addr = get_area(file, addr, len, pgoff, flags);
1568 if (IS_ERR_VALUE(addr))
1569 return addr;
1da177e4 1570
07ab67c8
LT
1571 if (addr > TASK_SIZE - len)
1572 return -ENOMEM;
1573 if (addr & ~PAGE_MASK)
1574 return -EINVAL;
06abdfb4 1575
08e7d9b5 1576 return arch_rebalance_pgtables(addr, len);
1da177e4
LT
1577}
1578
1579EXPORT_SYMBOL(get_unmapped_area);
1580
1581/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
48aae425 1582struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1da177e4
LT
1583{
1584 struct vm_area_struct *vma = NULL;
1585
1586 if (mm) {
1587 /* Check the cache first. */
1588 /* (Cache hit rate is typically around 35%.) */
1589 vma = mm->mmap_cache;
1590 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1591 struct rb_node * rb_node;
1592
1593 rb_node = mm->mm_rb.rb_node;
1594 vma = NULL;
1595
1596 while (rb_node) {
1597 struct vm_area_struct * vma_tmp;
1598
1599 vma_tmp = rb_entry(rb_node,
1600 struct vm_area_struct, vm_rb);
1601
1602 if (vma_tmp->vm_end > addr) {
1603 vma = vma_tmp;
1604 if (vma_tmp->vm_start <= addr)
1605 break;
1606 rb_node = rb_node->rb_left;
1607 } else
1608 rb_node = rb_node->rb_right;
1609 }
1610 if (vma)
1611 mm->mmap_cache = vma;
1612 }
1613 }
1614 return vma;
1615}
1616
1617EXPORT_SYMBOL(find_vma);
1618
6bd4837d
KM
1619/*
1620 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1621 * Note: pprev is set to NULL when return value is NULL.
1622 */
1da177e4
LT
1623struct vm_area_struct *
1624find_vma_prev(struct mm_struct *mm, unsigned long addr,
1625 struct vm_area_struct **pprev)
1626{
6bd4837d 1627 struct vm_area_struct *vma;
1da177e4 1628
6bd4837d
KM
1629 vma = find_vma(mm, addr);
1630 *pprev = vma ? vma->vm_prev : NULL;
1631 return vma;
1da177e4
LT
1632}
1633
1634/*
1635 * Verify that the stack growth is acceptable and
1636 * update accounting. This is shared with both the
1637 * grow-up and grow-down cases.
1638 */
48aae425 1639static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1da177e4
LT
1640{
1641 struct mm_struct *mm = vma->vm_mm;
1642 struct rlimit *rlim = current->signal->rlim;
0d59a01b 1643 unsigned long new_start;
1da177e4
LT
1644
1645 /* address space limit tests */
119f657c 1646 if (!may_expand_vm(mm, grow))
1da177e4
LT
1647 return -ENOMEM;
1648
1649 /* Stack limit test */
59e99e5b 1650 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1da177e4
LT
1651 return -ENOMEM;
1652
1653 /* mlock limit tests */
1654 if (vma->vm_flags & VM_LOCKED) {
1655 unsigned long locked;
1656 unsigned long limit;
1657 locked = mm->locked_vm + grow;
59e99e5b
JS
1658 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1659 limit >>= PAGE_SHIFT;
1da177e4
LT
1660 if (locked > limit && !capable(CAP_IPC_LOCK))
1661 return -ENOMEM;
1662 }
1663
0d59a01b
AL
1664 /* Check to ensure the stack will not grow into a hugetlb-only region */
1665 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1666 vma->vm_end - size;
1667 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1668 return -EFAULT;
1669
1da177e4
LT
1670 /*
1671 * Overcommit.. This must be the final test, as it will
1672 * update security statistics.
1673 */
05fa199d 1674 if (security_vm_enough_memory_mm(mm, grow))
1da177e4
LT
1675 return -ENOMEM;
1676
1677 /* Ok, everything looks good - let it rip */
1678 mm->total_vm += grow;
1679 if (vma->vm_flags & VM_LOCKED)
1680 mm->locked_vm += grow;
ab50b8ed 1681 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1da177e4
LT
1682 return 0;
1683}
1684
46dea3d0 1685#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1da177e4 1686/*
46dea3d0
HD
1687 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1688 * vma is the last one with address > vma->vm_end. Have to extend vma.
1da177e4 1689 */
46dea3d0 1690int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1da177e4
LT
1691{
1692 int error;
1693
1694 if (!(vma->vm_flags & VM_GROWSUP))
1695 return -EFAULT;
1696
1697 /*
1698 * We must make sure the anon_vma is allocated
1699 * so that the anon_vma locking is not a noop.
1700 */
1701 if (unlikely(anon_vma_prepare(vma)))
1702 return -ENOMEM;
bb4a340e 1703 vma_lock_anon_vma(vma);
1da177e4
LT
1704
1705 /*
1706 * vma->vm_start/vm_end cannot change under us because the caller
1707 * is required to hold the mmap_sem in read mode. We need the
1708 * anon_vma lock to serialize against concurrent expand_stacks.
06b32f3a 1709 * Also guard against wrapping around to address 0.
1da177e4 1710 */
06b32f3a
HD
1711 if (address < PAGE_ALIGN(address+4))
1712 address = PAGE_ALIGN(address+4);
1713 else {
bb4a340e 1714 vma_unlock_anon_vma(vma);
06b32f3a
HD
1715 return -ENOMEM;
1716 }
1da177e4
LT
1717 error = 0;
1718
1719 /* Somebody else might have raced and expanded it already */
1720 if (address > vma->vm_end) {
1721 unsigned long size, grow;
1722
1723 size = address - vma->vm_start;
1724 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1725
42c36f63
HD
1726 error = -ENOMEM;
1727 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1728 error = acct_stack_growth(vma, size, grow);
1729 if (!error) {
1730 vma->vm_end = address;
1731 perf_event_mmap(vma);
1732 }
3af9e859 1733 }
1da177e4 1734 }
bb4a340e 1735 vma_unlock_anon_vma(vma);
b15d00b6 1736 khugepaged_enter_vma_merge(vma);
1da177e4
LT
1737 return error;
1738}
46dea3d0
HD
1739#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1740
1da177e4
LT
1741/*
1742 * vma is the first one with address < vma->vm_start. Have to extend vma.
1743 */
d05f3169 1744int expand_downwards(struct vm_area_struct *vma,
b6a2fea3 1745 unsigned long address)
1da177e4
LT
1746{
1747 int error;
1748
1749 /*
1750 * We must make sure the anon_vma is allocated
1751 * so that the anon_vma locking is not a noop.
1752 */
1753 if (unlikely(anon_vma_prepare(vma)))
1754 return -ENOMEM;
8869477a
EP
1755
1756 address &= PAGE_MASK;
88c3f7a8 1757 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
8869477a
EP
1758 if (error)
1759 return error;
1760
bb4a340e 1761 vma_lock_anon_vma(vma);
1da177e4
LT
1762
1763 /*
1764 * vma->vm_start/vm_end cannot change under us because the caller
1765 * is required to hold the mmap_sem in read mode. We need the
1766 * anon_vma lock to serialize against concurrent expand_stacks.
1767 */
1da177e4
LT
1768
1769 /* Somebody else might have raced and expanded it already */
1770 if (address < vma->vm_start) {
1771 unsigned long size, grow;
1772
1773 size = vma->vm_end - address;
1774 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1775
a626ca6a
LT
1776 error = -ENOMEM;
1777 if (grow <= vma->vm_pgoff) {
1778 error = acct_stack_growth(vma, size, grow);
1779 if (!error) {
1780 vma->vm_start = address;
1781 vma->vm_pgoff -= grow;
1782 perf_event_mmap(vma);
1783 }
1da177e4
LT
1784 }
1785 }
bb4a340e 1786 vma_unlock_anon_vma(vma);
b15d00b6 1787 khugepaged_enter_vma_merge(vma);
1da177e4
LT
1788 return error;
1789}
1790
b6a2fea3
OW
1791#ifdef CONFIG_STACK_GROWSUP
1792int expand_stack(struct vm_area_struct *vma, unsigned long address)
1793{
1794 return expand_upwards(vma, address);
1795}
1796
1797struct vm_area_struct *
1798find_extend_vma(struct mm_struct *mm, unsigned long addr)
1799{
1800 struct vm_area_struct *vma, *prev;
1801
1802 addr &= PAGE_MASK;
1803 vma = find_vma_prev(mm, addr, &prev);
1804 if (vma && (vma->vm_start <= addr))
1805 return vma;
1c127185 1806 if (!prev || expand_stack(prev, addr))
b6a2fea3 1807 return NULL;
ba470de4 1808 if (prev->vm_flags & VM_LOCKED) {
c58267c3 1809 mlock_vma_pages_range(prev, addr, prev->vm_end);
ba470de4 1810 }
b6a2fea3
OW
1811 return prev;
1812}
1813#else
1814int expand_stack(struct vm_area_struct *vma, unsigned long address)
1815{
1816 return expand_downwards(vma, address);
1817}
1818
1da177e4
LT
1819struct vm_area_struct *
1820find_extend_vma(struct mm_struct * mm, unsigned long addr)
1821{
1822 struct vm_area_struct * vma;
1823 unsigned long start;
1824
1825 addr &= PAGE_MASK;
1826 vma = find_vma(mm,addr);
1827 if (!vma)
1828 return NULL;
1829 if (vma->vm_start <= addr)
1830 return vma;
1831 if (!(vma->vm_flags & VM_GROWSDOWN))
1832 return NULL;
1833 start = vma->vm_start;
1834 if (expand_stack(vma, addr))
1835 return NULL;
ba470de4 1836 if (vma->vm_flags & VM_LOCKED) {
c58267c3 1837 mlock_vma_pages_range(vma, addr, start);
ba470de4 1838 }
1da177e4
LT
1839 return vma;
1840}
1841#endif
1842
1da177e4 1843/*
2c0b3814 1844 * Ok - we have the memory areas we should free on the vma list,
1da177e4 1845 * so release them, and do the vma updates.
2c0b3814
HD
1846 *
1847 * Called with the mm semaphore held.
1da177e4 1848 */
2c0b3814 1849static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 1850{
365e9c87
HD
1851 /* Update high watermark before we lower total_vm */
1852 update_hiwater_vm(mm);
1da177e4 1853 do {
2c0b3814
HD
1854 long nrpages = vma_pages(vma);
1855
1856 mm->total_vm -= nrpages;
2c0b3814 1857 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
a8fb5618 1858 vma = remove_vma(vma);
146425a3 1859 } while (vma);
1da177e4
LT
1860 validate_mm(mm);
1861}
1862
1863/*
1864 * Get rid of page table information in the indicated region.
1865 *
f10df686 1866 * Called with the mm semaphore held.
1da177e4
LT
1867 */
1868static void unmap_region(struct mm_struct *mm,
e0da382c
HD
1869 struct vm_area_struct *vma, struct vm_area_struct *prev,
1870 unsigned long start, unsigned long end)
1da177e4 1871{
e0da382c 1872 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
d16dfc55 1873 struct mmu_gather tlb;
1da177e4
LT
1874 unsigned long nr_accounted = 0;
1875
1876 lru_add_drain();
d16dfc55 1877 tlb_gather_mmu(&tlb, mm, 0);
365e9c87 1878 update_hiwater_rss(mm);
508034a3 1879 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1da177e4 1880 vm_unacct_memory(nr_accounted);
d16dfc55
PZ
1881 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1882 next ? next->vm_start : 0);
1883 tlb_finish_mmu(&tlb, start, end);
1da177e4
LT
1884}
1885
1886/*
1887 * Create a list of vma's touched by the unmap, removing them from the mm's
1888 * vma list as we go..
1889 */
1890static void
1891detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1892 struct vm_area_struct *prev, unsigned long end)
1893{
1894 struct vm_area_struct **insertion_point;
1895 struct vm_area_struct *tail_vma = NULL;
1363c3cd 1896 unsigned long addr;
1da177e4
LT
1897
1898 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
297c5eee 1899 vma->vm_prev = NULL;
1da177e4
LT
1900 do {
1901 rb_erase(&vma->vm_rb, &mm->mm_rb);
1902 mm->map_count--;
1903 tail_vma = vma;
1904 vma = vma->vm_next;
1905 } while (vma && vma->vm_start < end);
1906 *insertion_point = vma;
297c5eee
LT
1907 if (vma)
1908 vma->vm_prev = prev;
1da177e4 1909 tail_vma->vm_next = NULL;
1363c3cd
WW
1910 if (mm->unmap_area == arch_unmap_area)
1911 addr = prev ? prev->vm_end : mm->mmap_base;
1912 else
1913 addr = vma ? vma->vm_start : mm->mmap_base;
1914 mm->unmap_area(mm, addr);
1da177e4
LT
1915 mm->mmap_cache = NULL; /* Kill the cache. */
1916}
1917
1918/*
659ace58
KM
1919 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1920 * munmap path where it doesn't make sense to fail.
1da177e4 1921 */
659ace58 1922static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1da177e4
LT
1923 unsigned long addr, int new_below)
1924{
1925 struct mempolicy *pol;
1926 struct vm_area_struct *new;
5beb4930 1927 int err = -ENOMEM;
1da177e4 1928
a5516438
AK
1929 if (is_vm_hugetlb_page(vma) && (addr &
1930 ~(huge_page_mask(hstate_vma(vma)))))
1da177e4
LT
1931 return -EINVAL;
1932
e94b1766 1933 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1da177e4 1934 if (!new)
5beb4930 1935 goto out_err;
1da177e4
LT
1936
1937 /* most fields are the same, copy all, and then fixup */
1938 *new = *vma;
1939
5beb4930
RR
1940 INIT_LIST_HEAD(&new->anon_vma_chain);
1941
1da177e4
LT
1942 if (new_below)
1943 new->vm_end = addr;
1944 else {
1945 new->vm_start = addr;
1946 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1947 }
1948
846a16bf 1949 pol = mpol_dup(vma_policy(vma));
1da177e4 1950 if (IS_ERR(pol)) {
5beb4930
RR
1951 err = PTR_ERR(pol);
1952 goto out_free_vma;
1da177e4
LT
1953 }
1954 vma_set_policy(new, pol);
1955
5beb4930
RR
1956 if (anon_vma_clone(new, vma))
1957 goto out_free_mpol;
1958
925d1c40 1959 if (new->vm_file) {
1da177e4 1960 get_file(new->vm_file);
925d1c40
MH
1961 if (vma->vm_flags & VM_EXECUTABLE)
1962 added_exe_file_vma(mm);
1963 }
1da177e4
LT
1964
1965 if (new->vm_ops && new->vm_ops->open)
1966 new->vm_ops->open(new);
1967
1968 if (new_below)
5beb4930 1969 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1da177e4
LT
1970 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1971 else
5beb4930 1972 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1da177e4 1973
5beb4930
RR
1974 /* Success. */
1975 if (!err)
1976 return 0;
1977
1978 /* Clean everything up if vma_adjust failed. */
58927533
RR
1979 if (new->vm_ops && new->vm_ops->close)
1980 new->vm_ops->close(new);
5beb4930
RR
1981 if (new->vm_file) {
1982 if (vma->vm_flags & VM_EXECUTABLE)
1983 removed_exe_file_vma(mm);
1984 fput(new->vm_file);
1985 }
2aeadc30 1986 unlink_anon_vmas(new);
5beb4930
RR
1987 out_free_mpol:
1988 mpol_put(pol);
1989 out_free_vma:
1990 kmem_cache_free(vm_area_cachep, new);
1991 out_err:
1992 return err;
1da177e4
LT
1993}
1994
659ace58
KM
1995/*
1996 * Split a vma into two pieces at address 'addr', a new vma is allocated
1997 * either for the first part or the tail.
1998 */
1999int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2000 unsigned long addr, int new_below)
2001{
2002 if (mm->map_count >= sysctl_max_map_count)
2003 return -ENOMEM;
2004
2005 return __split_vma(mm, vma, addr, new_below);
2006}
2007
1da177e4
LT
2008/* Munmap is split into 2 main parts -- this part which finds
2009 * what needs doing, and the areas themselves, which do the
2010 * work. This now handles partial unmappings.
2011 * Jeremy Fitzhardinge <jeremy@goop.org>
2012 */
2013int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2014{
2015 unsigned long end;
146425a3 2016 struct vm_area_struct *vma, *prev, *last;
1da177e4
LT
2017
2018 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2019 return -EINVAL;
2020
2021 if ((len = PAGE_ALIGN(len)) == 0)
2022 return -EINVAL;
2023
2024 /* Find the first overlapping VMA */
9be34c9d 2025 vma = find_vma(mm, start);
146425a3 2026 if (!vma)
1da177e4 2027 return 0;
9be34c9d 2028 prev = vma->vm_prev;
146425a3 2029 /* we have start < vma->vm_end */
1da177e4
LT
2030
2031 /* if it doesn't overlap, we have nothing.. */
2032 end = start + len;
146425a3 2033 if (vma->vm_start >= end)
1da177e4
LT
2034 return 0;
2035
2036 /*
2037 * If we need to split any vma, do it now to save pain later.
2038 *
2039 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2040 * unmapped vm_area_struct will remain in use: so lower split_vma
2041 * places tmp vma above, and higher split_vma places tmp vma below.
2042 */
146425a3 2043 if (start > vma->vm_start) {
659ace58
KM
2044 int error;
2045
2046 /*
2047 * Make sure that map_count on return from munmap() will
2048 * not exceed its limit; but let map_count go just above
2049 * its limit temporarily, to help free resources as expected.
2050 */
2051 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2052 return -ENOMEM;
2053
2054 error = __split_vma(mm, vma, start, 0);
1da177e4
LT
2055 if (error)
2056 return error;
146425a3 2057 prev = vma;
1da177e4
LT
2058 }
2059
2060 /* Does it split the last one? */
2061 last = find_vma(mm, end);
2062 if (last && end > last->vm_start) {
659ace58 2063 int error = __split_vma(mm, last, end, 1);
1da177e4
LT
2064 if (error)
2065 return error;
2066 }
146425a3 2067 vma = prev? prev->vm_next: mm->mmap;
1da177e4 2068
ba470de4
RR
2069 /*
2070 * unlock any mlock()ed ranges before detaching vmas
2071 */
2072 if (mm->locked_vm) {
2073 struct vm_area_struct *tmp = vma;
2074 while (tmp && tmp->vm_start < end) {
2075 if (tmp->vm_flags & VM_LOCKED) {
2076 mm->locked_vm -= vma_pages(tmp);
2077 munlock_vma_pages_all(tmp);
2078 }
2079 tmp = tmp->vm_next;
2080 }
2081 }
2082
1da177e4
LT
2083 /*
2084 * Remove the vma's, and unmap the actual pages
2085 */
146425a3
HD
2086 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2087 unmap_region(mm, vma, prev, start, end);
1da177e4
LT
2088
2089 /* Fix up all other VM information */
2c0b3814 2090 remove_vma_list(mm, vma);
1da177e4
LT
2091
2092 return 0;
2093}
2094
2095EXPORT_SYMBOL(do_munmap);
2096
6a6160a7 2097SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1da177e4
LT
2098{
2099 int ret;
2100 struct mm_struct *mm = current->mm;
2101
2102 profile_munmap(addr);
2103
2104 down_write(&mm->mmap_sem);
2105 ret = do_munmap(mm, addr, len);
2106 up_write(&mm->mmap_sem);
2107 return ret;
2108}
2109
2110static inline void verify_mm_writelocked(struct mm_struct *mm)
2111{
a241ec65 2112#ifdef CONFIG_DEBUG_VM
1da177e4
LT
2113 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2114 WARN_ON(1);
2115 up_read(&mm->mmap_sem);
2116 }
2117#endif
2118}
2119
2120/*
2121 * this is really a simplified "do_mmap". it only handles
2122 * anonymous maps. eventually we may be able to do some
2123 * brk-specific accounting here.
2124 */
2125unsigned long do_brk(unsigned long addr, unsigned long len)
2126{
2127 struct mm_struct * mm = current->mm;
2128 struct vm_area_struct * vma, * prev;
2129 unsigned long flags;
2130 struct rb_node ** rb_link, * rb_parent;
2131 pgoff_t pgoff = addr >> PAGE_SHIFT;
3a459756 2132 int error;
1da177e4
LT
2133
2134 len = PAGE_ALIGN(len);
2135 if (!len)
2136 return addr;
2137
88c3f7a8 2138 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
5a211a5d
EP
2139 if (error)
2140 return error;
2141
3a459756
KK
2142 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2143
2c6a1016
AV
2144 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2145 if (error & ~PAGE_MASK)
3a459756
KK
2146 return error;
2147
1da177e4
LT
2148 /*
2149 * mlock MCL_FUTURE?
2150 */
2151 if (mm->def_flags & VM_LOCKED) {
2152 unsigned long locked, lock_limit;
93ea1d0a
CW
2153 locked = len >> PAGE_SHIFT;
2154 locked += mm->locked_vm;
59e99e5b 2155 lock_limit = rlimit(RLIMIT_MEMLOCK);
93ea1d0a 2156 lock_limit >>= PAGE_SHIFT;
1da177e4
LT
2157 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2158 return -EAGAIN;
2159 }
2160
2161 /*
2162 * mm->mmap_sem is required to protect against another thread
2163 * changing the mappings in case we sleep.
2164 */
2165 verify_mm_writelocked(mm);
2166
2167 /*
2168 * Clear old maps. this also does some error checking for us
2169 */
2170 munmap_back:
2171 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2172 if (vma && vma->vm_start < addr + len) {
2173 if (do_munmap(mm, addr, len))
2174 return -ENOMEM;
2175 goto munmap_back;
2176 }
2177
2178 /* Check against address space limits *after* clearing old maps... */
119f657c 2179 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1da177e4
LT
2180 return -ENOMEM;
2181
2182 if (mm->map_count > sysctl_max_map_count)
2183 return -ENOMEM;
2184
191c5424 2185 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
1da177e4
LT
2186 return -ENOMEM;
2187
1da177e4 2188 /* Can we just expand an old private anonymous mapping? */
ba470de4
RR
2189 vma = vma_merge(mm, prev, addr, addr + len, flags,
2190 NULL, NULL, pgoff, NULL);
2191 if (vma)
1da177e4
LT
2192 goto out;
2193
2194 /*
2195 * create a vma struct for an anonymous mapping
2196 */
c5e3b83e 2197 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1da177e4
LT
2198 if (!vma) {
2199 vm_unacct_memory(len >> PAGE_SHIFT);
2200 return -ENOMEM;
2201 }
1da177e4 2202
5beb4930 2203 INIT_LIST_HEAD(&vma->anon_vma_chain);
1da177e4
LT
2204 vma->vm_mm = mm;
2205 vma->vm_start = addr;
2206 vma->vm_end = addr + len;
2207 vma->vm_pgoff = pgoff;
2208 vma->vm_flags = flags;
3ed75eb8 2209 vma->vm_page_prot = vm_get_page_prot(flags);
1da177e4
LT
2210 vma_link(mm, vma, prev, rb_link, rb_parent);
2211out:
3af9e859 2212 perf_event_mmap(vma);
1da177e4
LT
2213 mm->total_vm += len >> PAGE_SHIFT;
2214 if (flags & VM_LOCKED) {
ba470de4
RR
2215 if (!mlock_vma_pages_range(vma, addr, addr + len))
2216 mm->locked_vm += (len >> PAGE_SHIFT);
1da177e4
LT
2217 }
2218 return addr;
2219}
2220
2221EXPORT_SYMBOL(do_brk);
2222
2223/* Release all mmaps. */
2224void exit_mmap(struct mm_struct *mm)
2225{
d16dfc55 2226 struct mmu_gather tlb;
ba470de4 2227 struct vm_area_struct *vma;
1da177e4 2228 unsigned long nr_accounted = 0;
ee39b37b 2229 unsigned long end;
1da177e4 2230
d6dd61c8 2231 /* mm's last user has gone, and its about to be pulled down */
cddb8a5c 2232 mmu_notifier_release(mm);
d6dd61c8 2233
ba470de4
RR
2234 if (mm->locked_vm) {
2235 vma = mm->mmap;
2236 while (vma) {
2237 if (vma->vm_flags & VM_LOCKED)
2238 munlock_vma_pages_all(vma);
2239 vma = vma->vm_next;
2240 }
2241 }
9480c53e
JF
2242
2243 arch_exit_mmap(mm);
2244
ba470de4 2245 vma = mm->mmap;
9480c53e
JF
2246 if (!vma) /* Can happen if dup_mmap() received an OOM */
2247 return;
2248
1da177e4 2249 lru_add_drain();
1da177e4 2250 flush_cache_mm(mm);
d16dfc55 2251 tlb_gather_mmu(&tlb, mm, 1);
901608d9 2252 /* update_hiwater_rss(mm) here? but nobody should be looking */
e0da382c 2253 /* Use -1 here to ensure all VMAs in the mm are unmapped */
508034a3 2254 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
1da177e4 2255 vm_unacct_memory(nr_accounted);
9ba69294 2256
d16dfc55
PZ
2257 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2258 tlb_finish_mmu(&tlb, 0, end);
1da177e4 2259
1da177e4 2260 /*
8f4f8c16
HD
2261 * Walk the list again, actually closing and freeing it,
2262 * with preemption enabled, without holding any MM locks.
1da177e4 2263 */
a8fb5618
HD
2264 while (vma)
2265 vma = remove_vma(vma);
e0da382c 2266
e2cdef8c 2267 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
1da177e4
LT
2268}
2269
2270/* Insert vm structure into process list sorted by address
2271 * and into the inode's i_mmap tree. If vm_file is non-NULL
3d48ae45 2272 * then i_mmap_mutex is taken here.
1da177e4
LT
2273 */
2274int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2275{
2276 struct vm_area_struct * __vma, * prev;
2277 struct rb_node ** rb_link, * rb_parent;
2278
2279 /*
2280 * The vm_pgoff of a purely anonymous vma should be irrelevant
2281 * until its first write fault, when page's anon_vma and index
2282 * are set. But now set the vm_pgoff it will almost certainly
2283 * end up with (unless mremap moves it elsewhere before that
2284 * first wfault), so /proc/pid/maps tells a consistent story.
2285 *
2286 * By setting it to reflect the virtual start address of the
2287 * vma, merges and splits can happen in a seamless way, just
2288 * using the existing file pgoff checks and manipulations.
2289 * Similarly in do_mmap_pgoff and in do_brk.
2290 */
2291 if (!vma->vm_file) {
2292 BUG_ON(vma->anon_vma);
2293 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2294 }
2295 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2296 if (__vma && __vma->vm_start < vma->vm_end)
2297 return -ENOMEM;
2fd4ef85 2298 if ((vma->vm_flags & VM_ACCOUNT) &&
34b4e4aa 2299 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2fd4ef85 2300 return -ENOMEM;
1da177e4
LT
2301 vma_link(mm, vma, prev, rb_link, rb_parent);
2302 return 0;
2303}
2304
2305/*
2306 * Copy the vma structure to a new location in the same mm,
2307 * prior to moving page table entries, to effect an mremap move.
2308 */
2309struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2310 unsigned long addr, unsigned long len, pgoff_t pgoff)
2311{
2312 struct vm_area_struct *vma = *vmap;
2313 unsigned long vma_start = vma->vm_start;
2314 struct mm_struct *mm = vma->vm_mm;
2315 struct vm_area_struct *new_vma, *prev;
2316 struct rb_node **rb_link, *rb_parent;
2317 struct mempolicy *pol;
948f017b 2318 bool faulted_in_anon_vma = true;
1da177e4
LT
2319
2320 /*
2321 * If anonymous vma has not yet been faulted, update new pgoff
2322 * to match new location, to increase its chance of merging.
2323 */
948f017b 2324 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
1da177e4 2325 pgoff = addr >> PAGE_SHIFT;
948f017b
AA
2326 faulted_in_anon_vma = false;
2327 }
1da177e4
LT
2328
2329 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2330 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2331 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2332 if (new_vma) {
2333 /*
2334 * Source vma may have been merged into new_vma
2335 */
948f017b
AA
2336 if (unlikely(vma_start >= new_vma->vm_start &&
2337 vma_start < new_vma->vm_end)) {
2338 /*
2339 * The only way we can get a vma_merge with
2340 * self during an mremap is if the vma hasn't
2341 * been faulted in yet and we were allowed to
2342 * reset the dst vma->vm_pgoff to the
2343 * destination address of the mremap to allow
2344 * the merge to happen. mremap must change the
2345 * vm_pgoff linearity between src and dst vmas
2346 * (in turn preventing a vma_merge) to be
2347 * safe. It is only safe to keep the vm_pgoff
2348 * linear if there are no pages mapped yet.
2349 */
2350 VM_BUG_ON(faulted_in_anon_vma);
1da177e4 2351 *vmap = new_vma;
948f017b
AA
2352 } else
2353 anon_vma_moveto_tail(new_vma);
1da177e4 2354 } else {
e94b1766 2355 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1da177e4
LT
2356 if (new_vma) {
2357 *new_vma = *vma;
846a16bf 2358 pol = mpol_dup(vma_policy(vma));
5beb4930
RR
2359 if (IS_ERR(pol))
2360 goto out_free_vma;
2361 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2362 if (anon_vma_clone(new_vma, vma))
2363 goto out_free_mempol;
1da177e4
LT
2364 vma_set_policy(new_vma, pol);
2365 new_vma->vm_start = addr;
2366 new_vma->vm_end = addr + len;
2367 new_vma->vm_pgoff = pgoff;
925d1c40 2368 if (new_vma->vm_file) {
1da177e4 2369 get_file(new_vma->vm_file);
925d1c40
MH
2370 if (vma->vm_flags & VM_EXECUTABLE)
2371 added_exe_file_vma(mm);
2372 }
1da177e4
LT
2373 if (new_vma->vm_ops && new_vma->vm_ops->open)
2374 new_vma->vm_ops->open(new_vma);
2375 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2376 }
2377 }
2378 return new_vma;
5beb4930
RR
2379
2380 out_free_mempol:
2381 mpol_put(pol);
2382 out_free_vma:
2383 kmem_cache_free(vm_area_cachep, new_vma);
2384 return NULL;
1da177e4 2385}
119f657c 2386
2387/*
2388 * Return true if the calling process may expand its vm space by the passed
2389 * number of pages
2390 */
2391int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2392{
2393 unsigned long cur = mm->total_vm; /* pages */
2394 unsigned long lim;
2395
59e99e5b 2396 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
119f657c 2397
2398 if (cur + npages > lim)
2399 return 0;
2400 return 1;
2401}
fa5dc22f
RM
2402
2403
b1d0e4f5
NP
2404static int special_mapping_fault(struct vm_area_struct *vma,
2405 struct vm_fault *vmf)
fa5dc22f 2406{
b1d0e4f5 2407 pgoff_t pgoff;
fa5dc22f
RM
2408 struct page **pages;
2409
b1d0e4f5
NP
2410 /*
2411 * special mappings have no vm_file, and in that case, the mm
2412 * uses vm_pgoff internally. So we have to subtract it from here.
2413 * We are allowed to do this because we are the mm; do not copy
2414 * this code into drivers!
2415 */
2416 pgoff = vmf->pgoff - vma->vm_pgoff;
fa5dc22f 2417
b1d0e4f5
NP
2418 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2419 pgoff--;
fa5dc22f
RM
2420
2421 if (*pages) {
2422 struct page *page = *pages;
2423 get_page(page);
b1d0e4f5
NP
2424 vmf->page = page;
2425 return 0;
fa5dc22f
RM
2426 }
2427
b1d0e4f5 2428 return VM_FAULT_SIGBUS;
fa5dc22f
RM
2429}
2430
2431/*
2432 * Having a close hook prevents vma merging regardless of flags.
2433 */
2434static void special_mapping_close(struct vm_area_struct *vma)
2435{
2436}
2437
f0f37e2f 2438static const struct vm_operations_struct special_mapping_vmops = {
fa5dc22f 2439 .close = special_mapping_close,
b1d0e4f5 2440 .fault = special_mapping_fault,
fa5dc22f
RM
2441};
2442
2443/*
2444 * Called with mm->mmap_sem held for writing.
2445 * Insert a new vma covering the given region, with the given flags.
2446 * Its pages are supplied by the given array of struct page *.
2447 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2448 * The region past the last page supplied will always produce SIGBUS.
2449 * The array pointer and the pages it points to are assumed to stay alive
2450 * for as long as this mapping might exist.
2451 */
2452int install_special_mapping(struct mm_struct *mm,
2453 unsigned long addr, unsigned long len,
2454 unsigned long vm_flags, struct page **pages)
2455{
462e635e 2456 int ret;
fa5dc22f
RM
2457 struct vm_area_struct *vma;
2458
2459 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2460 if (unlikely(vma == NULL))
2461 return -ENOMEM;
2462
5beb4930 2463 INIT_LIST_HEAD(&vma->anon_vma_chain);
fa5dc22f
RM
2464 vma->vm_mm = mm;
2465 vma->vm_start = addr;
2466 vma->vm_end = addr + len;
2467
2f98735c 2468 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
3ed75eb8 2469 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
fa5dc22f
RM
2470
2471 vma->vm_ops = &special_mapping_vmops;
2472 vma->vm_private_data = pages;
2473
462e635e
TO
2474 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2475 if (ret)
2476 goto out;
2477
2478 ret = insert_vm_struct(mm, vma);
2479 if (ret)
2480 goto out;
fa5dc22f
RM
2481
2482 mm->total_vm += len >> PAGE_SHIFT;
2483
cdd6c482 2484 perf_event_mmap(vma);
089dd79d 2485
fa5dc22f 2486 return 0;
462e635e
TO
2487
2488out:
2489 kmem_cache_free(vm_area_cachep, vma);
2490 return ret;
fa5dc22f 2491}
7906d00c
AA
2492
2493static DEFINE_MUTEX(mm_all_locks_mutex);
2494
454ed842 2495static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
7906d00c 2496{
012f1800 2497 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
7906d00c
AA
2498 /*
2499 * The LSB of head.next can't change from under us
2500 * because we hold the mm_all_locks_mutex.
2501 */
2b575eb6 2502 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
7906d00c
AA
2503 /*
2504 * We can safely modify head.next after taking the
2b575eb6 2505 * anon_vma->root->mutex. If some other vma in this mm shares
7906d00c
AA
2506 * the same anon_vma we won't take it again.
2507 *
2508 * No need of atomic instructions here, head.next
2509 * can't change from under us thanks to the
2b575eb6 2510 * anon_vma->root->mutex.
7906d00c
AA
2511 */
2512 if (__test_and_set_bit(0, (unsigned long *)
012f1800 2513 &anon_vma->root->head.next))
7906d00c
AA
2514 BUG();
2515 }
2516}
2517
454ed842 2518static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
7906d00c
AA
2519{
2520 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2521 /*
2522 * AS_MM_ALL_LOCKS can't change from under us because
2523 * we hold the mm_all_locks_mutex.
2524 *
2525 * Operations on ->flags have to be atomic because
2526 * even if AS_MM_ALL_LOCKS is stable thanks to the
2527 * mm_all_locks_mutex, there may be other cpus
2528 * changing other bitflags in parallel to us.
2529 */
2530 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2531 BUG();
3d48ae45 2532 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
7906d00c
AA
2533 }
2534}
2535
2536/*
2537 * This operation locks against the VM for all pte/vma/mm related
2538 * operations that could ever happen on a certain mm. This includes
2539 * vmtruncate, try_to_unmap, and all page faults.
2540 *
2541 * The caller must take the mmap_sem in write mode before calling
2542 * mm_take_all_locks(). The caller isn't allowed to release the
2543 * mmap_sem until mm_drop_all_locks() returns.
2544 *
2545 * mmap_sem in write mode is required in order to block all operations
2546 * that could modify pagetables and free pages without need of
2547 * altering the vma layout (for example populate_range() with
2548 * nonlinear vmas). It's also needed in write mode to avoid new
2549 * anon_vmas to be associated with existing vmas.
2550 *
2551 * A single task can't take more than one mm_take_all_locks() in a row
2552 * or it would deadlock.
2553 *
2554 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2555 * mapping->flags avoid to take the same lock twice, if more than one
2556 * vma in this mm is backed by the same anon_vma or address_space.
2557 *
2558 * We can take all the locks in random order because the VM code
2b575eb6 2559 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
7906d00c
AA
2560 * takes more than one of them in a row. Secondly we're protected
2561 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2562 *
2563 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2564 * that may have to take thousand of locks.
2565 *
2566 * mm_take_all_locks() can fail if it's interrupted by signals.
2567 */
2568int mm_take_all_locks(struct mm_struct *mm)
2569{
2570 struct vm_area_struct *vma;
5beb4930 2571 struct anon_vma_chain *avc;
7906d00c
AA
2572
2573 BUG_ON(down_read_trylock(&mm->mmap_sem));
2574
2575 mutex_lock(&mm_all_locks_mutex);
2576
2577 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2578 if (signal_pending(current))
2579 goto out_unlock;
7906d00c 2580 if (vma->vm_file && vma->vm_file->f_mapping)
454ed842 2581 vm_lock_mapping(mm, vma->vm_file->f_mapping);
7906d00c 2582 }
7cd5a02f
PZ
2583
2584 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2585 if (signal_pending(current))
2586 goto out_unlock;
2587 if (vma->anon_vma)
5beb4930
RR
2588 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2589 vm_lock_anon_vma(mm, avc->anon_vma);
7906d00c 2590 }
7cd5a02f 2591
584cff54 2592 return 0;
7906d00c
AA
2593
2594out_unlock:
584cff54
KC
2595 mm_drop_all_locks(mm);
2596 return -EINTR;
7906d00c
AA
2597}
2598
2599static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2600{
012f1800 2601 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
7906d00c
AA
2602 /*
2603 * The LSB of head.next can't change to 0 from under
2604 * us because we hold the mm_all_locks_mutex.
2605 *
2606 * We must however clear the bitflag before unlocking
2607 * the vma so the users using the anon_vma->head will
2608 * never see our bitflag.
2609 *
2610 * No need of atomic instructions here, head.next
2611 * can't change from under us until we release the
2b575eb6 2612 * anon_vma->root->mutex.
7906d00c
AA
2613 */
2614 if (!__test_and_clear_bit(0, (unsigned long *)
012f1800 2615 &anon_vma->root->head.next))
7906d00c 2616 BUG();
cba48b98 2617 anon_vma_unlock(anon_vma);
7906d00c
AA
2618 }
2619}
2620
2621static void vm_unlock_mapping(struct address_space *mapping)
2622{
2623 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2624 /*
2625 * AS_MM_ALL_LOCKS can't change to 0 from under us
2626 * because we hold the mm_all_locks_mutex.
2627 */
3d48ae45 2628 mutex_unlock(&mapping->i_mmap_mutex);
7906d00c
AA
2629 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2630 &mapping->flags))
2631 BUG();
2632 }
2633}
2634
2635/*
2636 * The mmap_sem cannot be released by the caller until
2637 * mm_drop_all_locks() returns.
2638 */
2639void mm_drop_all_locks(struct mm_struct *mm)
2640{
2641 struct vm_area_struct *vma;
5beb4930 2642 struct anon_vma_chain *avc;
7906d00c
AA
2643
2644 BUG_ON(down_read_trylock(&mm->mmap_sem));
2645 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2646
2647 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2648 if (vma->anon_vma)
5beb4930
RR
2649 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2650 vm_unlock_anon_vma(avc->anon_vma);
7906d00c
AA
2651 if (vma->vm_file && vma->vm_file->f_mapping)
2652 vm_unlock_mapping(vma->vm_file->f_mapping);
2653 }
2654
2655 mutex_unlock(&mm_all_locks_mutex);
2656}
8feae131
DH
2657
2658/*
2659 * initialise the VMA slab
2660 */
2661void __init mmap_init(void)
2662{
00a62ce9
KM
2663 int ret;
2664
2665 ret = percpu_counter_init(&vm_committed_as, 0);
2666 VM_BUG_ON(ret);
8feae131 2667}
This page took 0.755545 seconds and 5 git commands to generate.