blob: 1abb9185a686d2787125c7af61120b68e726ccb5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
Alan Cox046c6882009-01-05 14:06:29 +00006 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 */
8
9#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070010#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#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>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/init.h>
19#include <linux/file.h>
20#include <linux/fs.h>
21#include <linux/personality.h>
22#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050023#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#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>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070030#include <linux/mmu_notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/uaccess.h>
33#include <asm/cacheflush.h>
34#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020035#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Jan Beulich42b77722008-07-23 21:27:10 -070037#include "internal.h"
38
Kirill Korotaev3a459752006-09-07 14:17:04 +040039#ifndef arch_mmap_check
40#define arch_mmap_check(addr, len, flags) (0)
41#endif
42
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080043#ifndef arch_rebalance_pgtables
44#define arch_rebalance_pgtables(addr, len) (addr)
45#endif
46
Hugh Dickinse0da3822005-04-19 13:29:15 -070047static void unmap_region(struct mm_struct *mm,
48 struct vm_area_struct *vma, struct vm_area_struct *prev,
49 unsigned long start, unsigned long end);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * WARNING: the debugging will use recursive algorithms so never enable this
53 * unless you know what you are doing.
54 */
55#undef DEBUG_MM_RB
56
57/* description of effects of mapping type and prot in current implementation.
58 * this is due to the limited x86 page protection hardware. The expected
59 * behavior is in parens:
60 *
61 * map_type prot
62 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
63 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
64 * w: (no) no w: (no) no w: (yes) yes w: (no) no
65 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
66 *
67 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
68 * w: (no) no w: (no) no w: (copy) copy w: (no) no
69 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
70 *
71 */
72pgprot_t protection_map[16] = {
73 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
74 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
75};
76
Hugh Dickins804af2c2006-07-26 21:39:49 +010077pgprot_t vm_get_page_prot(unsigned long vm_flags)
78{
Dave Kleikampb845f312008-07-08 00:28:51 +100079 return __pgprot(pgprot_val(protection_map[vm_flags &
80 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
81 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010082}
83EXPORT_SYMBOL(vm_get_page_prot);
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
86int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070087int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Alan Cox80119ef2008-05-23 13:04:31 -070088atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90/*
91 * Check that a process has enough memory to allocate a new virtual
92 * mapping. 0 means there is enough memory for the allocation to
93 * succeed and -ENOMEM implies there is not.
94 *
95 * We currently support three overcommit policies, which are set via the
96 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
97 *
98 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
99 * Additional code 2002 Jul 20 by Robert Love.
100 *
101 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
102 *
103 * Note this is a helper function intended to be used by LSMs which
104 * wish to use this logic.
105 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700106int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107{
108 unsigned long free, allowed;
109
110 vm_acct_memory(pages);
111
112 /*
113 * Sometimes we want to use more memory than we have
114 */
115 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
116 return 0;
117
118 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
119 unsigned long n;
120
Christoph Lameter347ce432006-06-30 01:55:35 -0700121 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 free += nr_swap_pages;
123
124 /*
125 * Any slabs which are created with the
126 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
127 * which are reclaimable, under pressure. The dentry
128 * cache and most inode caches should fall into this
129 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700130 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 /*
133 * Leave the last 3% for root
134 */
135 if (!cap_sys_admin)
136 free -= free / 32;
137
138 if (free > pages)
139 return 0;
140
141 /*
142 * nr_free_pages() is very expensive on large systems,
143 * only call if we're about to fail.
144 */
145 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700146
147 /*
148 * Leave reserved pages. The pages are not for anonymous pages.
149 */
150 if (n <= totalreserve_pages)
151 goto error;
152 else
153 n -= totalreserve_pages;
154
155 /*
156 * Leave the last 3% for root
157 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 if (!cap_sys_admin)
159 n -= n / 32;
160 free += n;
161
162 if (free > pages)
163 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700164
165 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 }
167
168 allowed = (totalram_pages - hugetlb_total_pages())
169 * sysctl_overcommit_ratio / 100;
170 /*
171 * Leave the last 3% for root
172 */
173 if (!cap_sys_admin)
174 allowed -= allowed / 32;
175 allowed += total_swap_pages;
176
177 /* Don't let a single process grow too big:
178 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700179 if (mm)
180 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Simon Derr2f60f8d2005-08-04 19:52:03 -0700182 /*
183 * cast `allowed' as a signed long because vm_committed_space
184 * sometimes has a negative value
185 */
Alan Cox80119ef2008-05-23 13:04:31 -0700186 if (atomic_long_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700188error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 vm_unacct_memory(pages);
190
191 return -ENOMEM;
192}
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/*
195 * Requires inode->i_mapping->i_mmap_lock
196 */
197static void __remove_shared_vm_struct(struct vm_area_struct *vma,
198 struct file *file, struct address_space *mapping)
199{
200 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800201 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 if (vma->vm_flags & VM_SHARED)
203 mapping->i_mmap_writable--;
204
205 flush_dcache_mmap_lock(mapping);
206 if (unlikely(vma->vm_flags & VM_NONLINEAR))
207 list_del_init(&vma->shared.vm_set.list);
208 else
209 vma_prio_tree_remove(vma, &mapping->i_mmap);
210 flush_dcache_mmap_unlock(mapping);
211}
212
213/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700214 * Unlink a file-based vm structure from its prio_tree, to hide
215 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700217void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
219 struct file *file = vma->vm_file;
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 if (file) {
222 struct address_space *mapping = file->f_mapping;
223 spin_lock(&mapping->i_mmap_lock);
224 __remove_shared_vm_struct(vma, file, mapping);
225 spin_unlock(&mapping->i_mmap_lock);
226 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700227}
228
229/*
230 * Close a vm structure and free it, returning the next.
231 */
232static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
233{
234 struct vm_area_struct *next = vma->vm_next;
235
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700236 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 if (vma->vm_ops && vma->vm_ops->close)
238 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700239 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700240 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700241 if (vma->vm_flags & VM_EXECUTABLE)
242 removed_exe_file_vma(vma->vm_mm);
243 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700244 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700246 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100249SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
251 unsigned long rlim, retval;
252 unsigned long newbrk, oldbrk;
253 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700254 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256 down_write(&mm->mmap_sem);
257
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700258#ifdef CONFIG_COMPAT_BRK
259 min_brk = mm->end_code;
260#else
261 min_brk = mm->start_brk;
262#endif
263 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700265
266 /*
267 * Check against rlimit here. If this check is done later after the test
268 * of oldbrk with newbrk then it can escape the test and let the data
269 * segment grow beyond its set limit the in case where the limit is
270 * not page aligned -Ram Gupta
271 */
272 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100273 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
274 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700275 goto out;
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 newbrk = PAGE_ALIGN(brk);
278 oldbrk = PAGE_ALIGN(mm->brk);
279 if (oldbrk == newbrk)
280 goto set_brk;
281
282 /* Always allow shrinking brk. */
283 if (brk <= mm->brk) {
284 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
285 goto set_brk;
286 goto out;
287 }
288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 /* Check against existing mmap mappings. */
290 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
291 goto out;
292
293 /* Ok, looks good - let it rip. */
294 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
295 goto out;
296set_brk:
297 mm->brk = brk;
298out:
299 retval = mm->brk;
300 up_write(&mm->mmap_sem);
301 return retval;
302}
303
304#ifdef DEBUG_MM_RB
305static int browse_rb(struct rb_root *root)
306{
307 int i = 0, j;
308 struct rb_node *nd, *pn = NULL;
309 unsigned long prev = 0, pend = 0;
310
311 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
312 struct vm_area_struct *vma;
313 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
314 if (vma->vm_start < prev)
315 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
316 if (vma->vm_start < pend)
317 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
318 if (vma->vm_start > vma->vm_end)
319 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
320 i++;
321 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800322 prev = vma->vm_start;
323 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 }
325 j = 0;
326 for (nd = pn; nd; nd = rb_prev(nd)) {
327 j++;
328 }
329 if (i != j)
330 printk("backwards %d, forwards %d\n", j, i), i = 0;
331 return i;
332}
333
334void validate_mm(struct mm_struct *mm)
335{
336 int bug = 0;
337 int i = 0;
338 struct vm_area_struct *tmp = mm->mmap;
339 while (tmp) {
340 tmp = tmp->vm_next;
341 i++;
342 }
343 if (i != mm->map_count)
344 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
345 i = browse_rb(&mm->mm_rb);
346 if (i != mm->map_count)
347 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200348 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349}
350#else
351#define validate_mm(mm) do { } while (0)
352#endif
353
354static struct vm_area_struct *
355find_vma_prepare(struct mm_struct *mm, unsigned long addr,
356 struct vm_area_struct **pprev, struct rb_node ***rb_link,
357 struct rb_node ** rb_parent)
358{
359 struct vm_area_struct * vma;
360 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
361
362 __rb_link = &mm->mm_rb.rb_node;
363 rb_prev = __rb_parent = NULL;
364 vma = NULL;
365
366 while (*__rb_link) {
367 struct vm_area_struct *vma_tmp;
368
369 __rb_parent = *__rb_link;
370 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
371
372 if (vma_tmp->vm_end > addr) {
373 vma = vma_tmp;
374 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700375 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 __rb_link = &__rb_parent->rb_left;
377 } else {
378 rb_prev = __rb_parent;
379 __rb_link = &__rb_parent->rb_right;
380 }
381 }
382
383 *pprev = NULL;
384 if (rb_prev)
385 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
386 *rb_link = __rb_link;
387 *rb_parent = __rb_parent;
388 return vma;
389}
390
391static inline void
392__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
393 struct vm_area_struct *prev, struct rb_node *rb_parent)
394{
395 if (prev) {
396 vma->vm_next = prev->vm_next;
397 prev->vm_next = vma;
398 } else {
399 mm->mmap = vma;
400 if (rb_parent)
401 vma->vm_next = rb_entry(rb_parent,
402 struct vm_area_struct, vm_rb);
403 else
404 vma->vm_next = NULL;
405 }
406}
407
408void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
409 struct rb_node **rb_link, struct rb_node *rb_parent)
410{
411 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
412 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
413}
414
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700415static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416{
ZhenwenXu48aae422009-01-06 14:40:21 -0800417 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 file = vma->vm_file;
420 if (file) {
421 struct address_space *mapping = file->f_mapping;
422
423 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800424 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 if (vma->vm_flags & VM_SHARED)
426 mapping->i_mmap_writable++;
427
428 flush_dcache_mmap_lock(mapping);
429 if (unlikely(vma->vm_flags & VM_NONLINEAR))
430 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
431 else
432 vma_prio_tree_insert(vma, &mapping->i_mmap);
433 flush_dcache_mmap_unlock(mapping);
434 }
435}
436
437static void
438__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
439 struct vm_area_struct *prev, struct rb_node **rb_link,
440 struct rb_node *rb_parent)
441{
442 __vma_link_list(mm, vma, prev, rb_parent);
443 __vma_link_rb(mm, vma, rb_link, rb_parent);
444 __anon_vma_link(vma);
445}
446
447static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
448 struct vm_area_struct *prev, struct rb_node **rb_link,
449 struct rb_node *rb_parent)
450{
451 struct address_space *mapping = NULL;
452
453 if (vma->vm_file)
454 mapping = vma->vm_file->f_mapping;
455
456 if (mapping) {
457 spin_lock(&mapping->i_mmap_lock);
458 vma->vm_truncate_count = mapping->truncate_count;
459 }
460 anon_vma_lock(vma);
461
462 __vma_link(mm, vma, prev, rb_link, rb_parent);
463 __vma_link_file(vma);
464
465 anon_vma_unlock(vma);
466 if (mapping)
467 spin_unlock(&mapping->i_mmap_lock);
468
469 mm->map_count++;
470 validate_mm(mm);
471}
472
473/*
474 * Helper for vma_adjust in the split_vma insert case:
475 * insert vm structure into list and rbtree and anon_vma,
476 * but it has already been inserted into prio_tree earlier.
477 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800478static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
ZhenwenXu48aae422009-01-06 14:40:21 -0800480 struct vm_area_struct *__vma, *prev;
481 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200484 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 __vma_link(mm, vma, prev, rb_link, rb_parent);
486 mm->map_count++;
487}
488
489static inline void
490__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
491 struct vm_area_struct *prev)
492{
493 prev->vm_next = vma->vm_next;
494 rb_erase(&vma->vm_rb, &mm->mm_rb);
495 if (mm->mmap_cache == vma)
496 mm->mmap_cache = prev;
497}
498
499/*
500 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
501 * is already present in an i_mmap tree without adjusting the tree.
502 * The following helper function should be used when such adjustments
503 * are necessary. The "insert" vma (if any) is to be inserted
504 * before we drop the necessary locks.
505 */
506void vma_adjust(struct vm_area_struct *vma, unsigned long start,
507 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
508{
509 struct mm_struct *mm = vma->vm_mm;
510 struct vm_area_struct *next = vma->vm_next;
511 struct vm_area_struct *importer = NULL;
512 struct address_space *mapping = NULL;
513 struct prio_tree_root *root = NULL;
514 struct file *file = vma->vm_file;
515 struct anon_vma *anon_vma = NULL;
516 long adjust_next = 0;
517 int remove_next = 0;
518
519 if (next && !insert) {
520 if (end >= next->vm_end) {
521 /*
522 * vma expands, overlapping all the next, and
523 * perhaps the one after too (mprotect case 6).
524 */
525again: remove_next = 1 + (end > next->vm_end);
526 end = next->vm_end;
527 anon_vma = next->anon_vma;
528 importer = vma;
529 } else if (end > next->vm_start) {
530 /*
531 * vma expands, overlapping part of the next:
532 * mprotect case 5 shifting the boundary up.
533 */
534 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
535 anon_vma = next->anon_vma;
536 importer = vma;
537 } else if (end < vma->vm_end) {
538 /*
539 * vma shrinks, and !insert tells it's not
540 * split_vma inserting another: so it must be
541 * mprotect case 4 shifting the boundary down.
542 */
543 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
544 anon_vma = next->anon_vma;
545 importer = next;
546 }
547 }
548
549 if (file) {
550 mapping = file->f_mapping;
551 if (!(vma->vm_flags & VM_NONLINEAR))
552 root = &mapping->i_mmap;
553 spin_lock(&mapping->i_mmap_lock);
554 if (importer &&
555 vma->vm_truncate_count != next->vm_truncate_count) {
556 /*
557 * unmap_mapping_range might be in progress:
558 * ensure that the expanding vma is rescanned.
559 */
560 importer->vm_truncate_count = 0;
561 }
562 if (insert) {
563 insert->vm_truncate_count = vma->vm_truncate_count;
564 /*
565 * Put into prio_tree now, so instantiated pages
566 * are visible to arm/parisc __flush_dcache_page
567 * throughout; but we cannot insert into address
568 * space until vma start or end is updated.
569 */
570 __vma_link_file(insert);
571 }
572 }
573
574 /*
575 * When changing only vma->vm_end, we don't really need
576 * anon_vma lock: but is that case worth optimizing out?
577 */
578 if (vma->anon_vma)
579 anon_vma = vma->anon_vma;
580 if (anon_vma) {
581 spin_lock(&anon_vma->lock);
582 /*
583 * Easily overlooked: when mprotect shifts the boundary,
584 * make sure the expanding vma has anon_vma set if the
585 * shrinking vma had, to cover any anon pages imported.
586 */
587 if (importer && !importer->anon_vma) {
588 importer->anon_vma = anon_vma;
589 __anon_vma_link(importer);
590 }
591 }
592
593 if (root) {
594 flush_dcache_mmap_lock(mapping);
595 vma_prio_tree_remove(vma, root);
596 if (adjust_next)
597 vma_prio_tree_remove(next, root);
598 }
599
600 vma->vm_start = start;
601 vma->vm_end = end;
602 vma->vm_pgoff = pgoff;
603 if (adjust_next) {
604 next->vm_start += adjust_next << PAGE_SHIFT;
605 next->vm_pgoff += adjust_next;
606 }
607
608 if (root) {
609 if (adjust_next)
610 vma_prio_tree_insert(next, root);
611 vma_prio_tree_insert(vma, root);
612 flush_dcache_mmap_unlock(mapping);
613 }
614
615 if (remove_next) {
616 /*
617 * vma_merge has merged next into vma, and needs
618 * us to remove next before dropping the locks.
619 */
620 __vma_unlink(mm, next, vma);
621 if (file)
622 __remove_shared_vm_struct(next, file, mapping);
623 if (next->anon_vma)
624 __anon_vma_merge(vma, next);
625 } else if (insert) {
626 /*
627 * split_vma has split insert from vma, and needs
628 * us to insert it before dropping the locks
629 * (it may either follow vma or precede it).
630 */
631 __insert_vm_struct(mm, insert);
632 }
633
634 if (anon_vma)
635 spin_unlock(&anon_vma->lock);
636 if (mapping)
637 spin_unlock(&mapping->i_mmap_lock);
638
639 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700640 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700642 if (next->vm_flags & VM_EXECUTABLE)
643 removed_exe_file_vma(mm);
644 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700646 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 kmem_cache_free(vm_area_cachep, next);
648 /*
649 * In mprotect's case 6 (see comments on vma_merge),
650 * we must remove another next too. It would clutter
651 * up the code too much to do both in one go.
652 */
653 if (remove_next == 2) {
654 next = vma->vm_next;
655 goto again;
656 }
657 }
658
659 validate_mm(mm);
660}
661
Linus Torvalds33bfad52009-01-30 11:37:22 -0800662/* Flags that can be inherited from an existing mapping when merging */
663#define VM_MERGEABLE_FLAGS (VM_CAN_NONLINEAR)
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665/*
666 * If the vma has a ->close operation then the driver probably needs to release
667 * per-vma resources, so we don't attempt to merge those.
668 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669static inline int is_mergeable_vma(struct vm_area_struct *vma,
670 struct file *file, unsigned long vm_flags)
671{
Linus Torvalds33bfad52009-01-30 11:37:22 -0800672 if ((vma->vm_flags ^ vm_flags) & ~VM_MERGEABLE_FLAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 return 0;
674 if (vma->vm_file != file)
675 return 0;
676 if (vma->vm_ops && vma->vm_ops->close)
677 return 0;
678 return 1;
679}
680
681static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
682 struct anon_vma *anon_vma2)
683{
684 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
685}
686
687/*
688 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
689 * in front of (at a lower virtual address and file offset than) the vma.
690 *
691 * We cannot merge two vmas if they have differently assigned (non-NULL)
692 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
693 *
694 * We don't check here for the merged mmap wrapping around the end of pagecache
695 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
696 * wrap, nor mmaps which cover the final page at index -1UL.
697 */
698static int
699can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
700 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
701{
702 if (is_mergeable_vma(vma, file, vm_flags) &&
703 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
704 if (vma->vm_pgoff == vm_pgoff)
705 return 1;
706 }
707 return 0;
708}
709
710/*
711 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
712 * beyond (at a higher virtual address and file offset than) the vma.
713 *
714 * We cannot merge two vmas if they have differently assigned (non-NULL)
715 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
716 */
717static int
718can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
719 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
720{
721 if (is_mergeable_vma(vma, file, vm_flags) &&
722 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
723 pgoff_t vm_pglen;
724 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
725 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
726 return 1;
727 }
728 return 0;
729}
730
731/*
732 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
733 * whether that can be merged with its predecessor or its successor.
734 * Or both (it neatly fills a hole).
735 *
736 * In most cases - when called for mmap, brk or mremap - [addr,end) is
737 * certain not to be mapped by the time vma_merge is called; but when
738 * called for mprotect, it is certain to be already mapped (either at
739 * an offset within prev, or at the start of next), and the flags of
740 * this area are about to be changed to vm_flags - and the no-change
741 * case has already been eliminated.
742 *
743 * The following mprotect cases have to be considered, where AAAA is
744 * the area passed down from mprotect_fixup, never extending beyond one
745 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
746 *
747 * AAAA AAAA AAAA AAAA
748 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
749 * cannot merge might become might become might become
750 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
751 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
752 * mremap move: PPPPNNNNNNNN 8
753 * AAAA
754 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
755 * might become case 1 below case 2 below case 3 below
756 *
757 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
758 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
759 */
760struct vm_area_struct *vma_merge(struct mm_struct *mm,
761 struct vm_area_struct *prev, unsigned long addr,
762 unsigned long end, unsigned long vm_flags,
763 struct anon_vma *anon_vma, struct file *file,
764 pgoff_t pgoff, struct mempolicy *policy)
765{
766 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
767 struct vm_area_struct *area, *next;
768
769 /*
770 * We later require that vma->vm_flags == vm_flags,
771 * so this tests vma->vm_flags & VM_SPECIAL, too.
772 */
773 if (vm_flags & VM_SPECIAL)
774 return NULL;
775
776 if (prev)
777 next = prev->vm_next;
778 else
779 next = mm->mmap;
780 area = next;
781 if (next && next->vm_end == end) /* cases 6, 7, 8 */
782 next = next->vm_next;
783
784 /*
785 * Can it merge with the predecessor?
786 */
787 if (prev && prev->vm_end == addr &&
788 mpol_equal(vma_policy(prev), policy) &&
789 can_vma_merge_after(prev, vm_flags,
790 anon_vma, file, pgoff)) {
791 /*
792 * OK, it can. Can we now merge in the successor as well?
793 */
794 if (next && end == next->vm_start &&
795 mpol_equal(policy, vma_policy(next)) &&
796 can_vma_merge_before(next, vm_flags,
797 anon_vma, file, pgoff+pglen) &&
798 is_mergeable_anon_vma(prev->anon_vma,
799 next->anon_vma)) {
800 /* cases 1, 6 */
801 vma_adjust(prev, prev->vm_start,
802 next->vm_end, prev->vm_pgoff, NULL);
803 } else /* cases 2, 5, 7 */
804 vma_adjust(prev, prev->vm_start,
805 end, prev->vm_pgoff, NULL);
806 return prev;
807 }
808
809 /*
810 * Can this new request be merged in front of next?
811 */
812 if (next && end == next->vm_start &&
813 mpol_equal(policy, vma_policy(next)) &&
814 can_vma_merge_before(next, vm_flags,
815 anon_vma, file, pgoff+pglen)) {
816 if (prev && addr < prev->vm_end) /* case 4 */
817 vma_adjust(prev, prev->vm_start,
818 addr, prev->vm_pgoff, NULL);
819 else /* cases 3, 8 */
820 vma_adjust(area, addr, next->vm_end,
821 next->vm_pgoff - pglen, NULL);
822 return area;
823 }
824
825 return NULL;
826}
827
828/*
829 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
830 * neighbouring vmas for a suitable anon_vma, before it goes off
831 * to allocate a new anon_vma. It checks because a repetitive
832 * sequence of mprotects and faults may otherwise lead to distinct
833 * anon_vmas being allocated, preventing vma merge in subsequent
834 * mprotect.
835 */
836struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
837{
838 struct vm_area_struct *near;
839 unsigned long vm_flags;
840
841 near = vma->vm_next;
842 if (!near)
843 goto try_prev;
844
845 /*
846 * Since only mprotect tries to remerge vmas, match flags
847 * which might be mprotected into each other later on.
848 * Neither mlock nor madvise tries to remerge at present,
849 * so leave their flags as obstructing a merge.
850 */
851 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
852 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
853
854 if (near->anon_vma && vma->vm_end == near->vm_start &&
855 mpol_equal(vma_policy(vma), vma_policy(near)) &&
856 can_vma_merge_before(near, vm_flags,
857 NULL, vma->vm_file, vma->vm_pgoff +
858 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
859 return near->anon_vma;
860try_prev:
861 /*
862 * It is potentially slow to have to call find_vma_prev here.
863 * But it's only on the first write fault on the vma, not
864 * every time, and we could devise a way to avoid it later
865 * (e.g. stash info in next's anon_vma_node when assigning
866 * an anon_vma, or when trying vma_merge). Another time.
867 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200868 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 if (!near)
870 goto none;
871
872 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
873 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
874
875 if (near->anon_vma && near->vm_end == vma->vm_start &&
876 mpol_equal(vma_policy(near), vma_policy(vma)) &&
877 can_vma_merge_after(near, vm_flags,
878 NULL, vma->vm_file, vma->vm_pgoff))
879 return near->anon_vma;
880none:
881 /*
882 * There's no absolute need to look only at touching neighbours:
883 * we could search further afield for "compatible" anon_vmas.
884 * But it would probably just be a waste of time searching,
885 * or lead to too many vmas hanging off the same anon_vma.
886 * We're trying to allow mprotect remerging later on,
887 * not trying to minimize memory used for anon_vmas.
888 */
889 return NULL;
890}
891
892#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700893void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 struct file *file, long pages)
895{
896 const unsigned long stack_flags
897 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (file) {
900 mm->shared_vm += pages;
901 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
902 mm->exec_vm += pages;
903 } else if (flags & stack_flags)
904 mm->stack_vm += pages;
905 if (flags & (VM_RESERVED|VM_IO))
906 mm->reserved_vm += pages;
907}
908#endif /* CONFIG_PROC_FS */
909
910/*
911 * The caller must hold down_write(current->mm->mmap_sem).
912 */
913
ZhenwenXu48aae422009-01-06 14:40:21 -0800914unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 unsigned long len, unsigned long prot,
916 unsigned long flags, unsigned long pgoff)
917{
918 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 struct inode *inode;
920 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700922 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 /*
925 * Does the application expect PROT_READ to imply PROT_EXEC?
926 *
927 * (the exception is when the underlying filesystem is noexec
928 * mounted, in which case we dont add PROT_EXEC.)
929 */
930 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800931 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 prot |= PROT_EXEC;
933
934 if (!len)
935 return -EINVAL;
936
Eric Paris7cd94142007-11-26 18:47:40 -0500937 if (!(flags & MAP_FIXED))
938 addr = round_hint_to_min(addr);
939
Kirill Korotaev3a459752006-09-07 14:17:04 +0400940 error = arch_mmap_check(addr, len, flags);
941 if (error)
942 return error;
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 /* Careful about overflows.. */
945 len = PAGE_ALIGN(len);
946 if (!len || len > TASK_SIZE)
947 return -ENOMEM;
948
949 /* offset overflow? */
950 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
951 return -EOVERFLOW;
952
953 /* Too many mappings? */
954 if (mm->map_count > sysctl_max_map_count)
955 return -ENOMEM;
956
957 /* Obtain the address to map to. we verify (or select) it and ensure
958 * that it represents a valid section of the address space.
959 */
960 addr = get_unmapped_area(file, addr, len, pgoff, flags);
961 if (addr & ~PAGE_MASK)
962 return addr;
963
964 /* Do simple checking here so the lower-level routines won't have
965 * to. we assume access permissions have been handled by the open
966 * of the memory object, so we don't do any here.
967 */
968 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
969 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
970
971 if (flags & MAP_LOCKED) {
972 if (!can_do_mlock())
973 return -EPERM;
974 vm_flags |= VM_LOCKED;
975 }
Rik van Rielba470de2008-10-18 20:26:50 -0700976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 /* mlock MCL_FUTURE? */
978 if (vm_flags & VM_LOCKED) {
979 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700980 locked = len >> PAGE_SHIFT;
981 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700983 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
985 return -EAGAIN;
986 }
987
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800988 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 if (file) {
991 switch (flags & MAP_TYPE) {
992 case MAP_SHARED:
993 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
994 return -EACCES;
995
996 /*
997 * Make sure we don't allow writing to an append-only
998 * file..
999 */
1000 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1001 return -EACCES;
1002
1003 /*
1004 * Make sure there are no mandatory locks on the file.
1005 */
1006 if (locks_verify_locked(inode))
1007 return -EAGAIN;
1008
1009 vm_flags |= VM_SHARED | VM_MAYSHARE;
1010 if (!(file->f_mode & FMODE_WRITE))
1011 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1012
1013 /* fall through */
1014 case MAP_PRIVATE:
1015 if (!(file->f_mode & FMODE_READ))
1016 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001017 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001018 if (vm_flags & VM_EXEC)
1019 return -EPERM;
1020 vm_flags &= ~VM_MAYEXEC;
1021 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001022
1023 if (!file->f_op || !file->f_op->mmap)
1024 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 break;
1026
1027 default:
1028 return -EINVAL;
1029 }
1030 } else {
1031 switch (flags & MAP_TYPE) {
1032 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001033 /*
1034 * Ignore pgoff.
1035 */
1036 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 vm_flags |= VM_SHARED | VM_MAYSHARE;
1038 break;
1039 case MAP_PRIVATE:
1040 /*
1041 * Set pgoff according to addr for anon_vma.
1042 */
1043 pgoff = addr >> PAGE_SHIFT;
1044 break;
1045 default:
1046 return -EINVAL;
1047 }
1048 }
1049
Eric Parised032182007-06-28 15:55:21 -04001050 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 if (error)
1052 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05001053 error = ima_file_mmap(file, prot);
1054 if (error)
1055 return error;
Eric Parised032182007-06-28 15:55:21 -04001056
Mel Gorman5a6fe122009-02-10 14:02:27 +00001057 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001058}
1059EXPORT_SYMBOL(do_mmap_pgoff);
1060
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001061/*
1062 * Some shared mappigns will want the pages marked read-only
1063 * to track write events. If so, we'll downgrade vm_page_prot
1064 * to the private version (using protection_map[] without the
1065 * VM_SHARED bit).
1066 */
1067int vma_wants_writenotify(struct vm_area_struct *vma)
1068{
1069 unsigned int vm_flags = vma->vm_flags;
1070
1071 /* If it was private or non-writable, the write bit is already clear */
1072 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1073 return 0;
1074
1075 /* The backer wishes to know when pages are first written to? */
1076 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1077 return 1;
1078
1079 /* The open routine did something to the protections already? */
1080 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001081 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001082 return 0;
1083
1084 /* Specialty mapping? */
1085 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1086 return 0;
1087
1088 /* Can the mapping track the dirty pages? */
1089 return vma->vm_file && vma->vm_file->f_mapping &&
1090 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1091}
1092
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001093/*
1094 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001095 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001096 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001097static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001098{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001099 /*
1100 * hugetlb has its own accounting separate from the core VM
1101 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1102 */
1103 if (file && is_file_hugepages(file))
1104 return 0;
1105
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001106 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1107}
1108
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001109unsigned long mmap_region(struct file *file, unsigned long addr,
1110 unsigned long len, unsigned long flags,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001111 unsigned int vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001112{
1113 struct mm_struct *mm = current->mm;
1114 struct vm_area_struct *vma, *prev;
1115 int correct_wcount = 0;
1116 int error;
1117 struct rb_node **rb_link, *rb_parent;
1118 unsigned long charged = 0;
1119 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1120
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 /* Clear old maps */
1122 error = -ENOMEM;
1123munmap_back:
1124 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1125 if (vma && vma->vm_start < addr + len) {
1126 if (do_munmap(mm, addr, len))
1127 return -ENOMEM;
1128 goto munmap_back;
1129 }
1130
1131 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001132 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return -ENOMEM;
1134
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001135 /*
1136 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001137 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001138 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001139 if ((flags & MAP_NORESERVE)) {
1140 /* We honor MAP_NORESERVE if allowed to overcommit */
1141 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1142 vm_flags |= VM_NORESERVE;
1143
1144 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1145 if (file && is_file_hugepages(file))
1146 vm_flags |= VM_NORESERVE;
1147 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001148
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001149 /*
1150 * Private writable mapping: check memory availability
1151 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001152 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001153 charged = len >> PAGE_SHIFT;
1154 if (security_vm_enough_memory(charged))
1155 return -ENOMEM;
1156 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 }
1158
1159 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001160 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001162 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1163 if (vma)
1164 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
1166 /*
1167 * Determine the object being mapped and call the appropriate
1168 * specific mapper. the address has already been validated, but
1169 * not unmapped, but the maps are removed from the list.
1170 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001171 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (!vma) {
1173 error = -ENOMEM;
1174 goto unacct_error;
1175 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
1177 vma->vm_mm = mm;
1178 vma->vm_start = addr;
1179 vma->vm_end = addr + len;
1180 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001181 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 vma->vm_pgoff = pgoff;
1183
1184 if (file) {
1185 error = -EINVAL;
1186 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1187 goto free_vma;
1188 if (vm_flags & VM_DENYWRITE) {
1189 error = deny_write_access(file);
1190 if (error)
1191 goto free_vma;
1192 correct_wcount = 1;
1193 }
1194 vma->vm_file = file;
1195 get_file(file);
1196 error = file->f_op->mmap(file, vma);
1197 if (error)
1198 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001199 if (vm_flags & VM_EXECUTABLE)
1200 added_exe_file_vma(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 } else if (vm_flags & VM_SHARED) {
1202 error = shmem_zero_setup(vma);
1203 if (error)
1204 goto free_vma;
1205 }
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 /* Can addr have changed??
1208 *
1209 * Answer: Yes, several device drivers can do it in their
1210 * f_op->mmap method. -DaveM
1211 */
1212 addr = vma->vm_start;
1213 pgoff = vma->vm_pgoff;
1214 vm_flags = vma->vm_flags;
1215
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001216 if (vma_wants_writenotify(vma))
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001217 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001218
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001219 vma_link(mm, vma, prev, rb_link, rb_parent);
1220 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001221
1222 /* Once vma denies write, undo our temporary denial count */
1223 if (correct_wcount)
1224 atomic_inc(&inode->i_writecount);
1225out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001227 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (vm_flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07001229 /*
1230 * makes pages present; downgrades, drops, reacquires mmap_sem
1231 */
1232 long nr_pages = mlock_vma_pages_range(vma, addr, addr + len);
1233 if (nr_pages < 0)
1234 return nr_pages; /* vma gone! */
1235 mm->locked_vm += (len >> PAGE_SHIFT) - nr_pages;
1236 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001237 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 return addr;
1239
1240unmap_and_free_vma:
1241 if (correct_wcount)
1242 atomic_inc(&inode->i_writecount);
1243 vma->vm_file = NULL;
1244 fput(file);
1245
1246 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001247 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1248 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249free_vma:
1250 kmem_cache_free(vm_area_cachep, vma);
1251unacct_error:
1252 if (charged)
1253 vm_unacct_memory(charged);
1254 return error;
1255}
1256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257/* Get an address range which is currently unmapped.
1258 * For shmat() with addr=0.
1259 *
1260 * Ugly calling convention alert:
1261 * Return value with the low bits set means error value,
1262 * ie
1263 * if (ret & ~PAGE_MASK)
1264 * error = ret;
1265 *
1266 * This function "knows" that -ENOMEM has the bits set.
1267 */
1268#ifndef HAVE_ARCH_UNMAPPED_AREA
1269unsigned long
1270arch_get_unmapped_area(struct file *filp, unsigned long addr,
1271 unsigned long len, unsigned long pgoff, unsigned long flags)
1272{
1273 struct mm_struct *mm = current->mm;
1274 struct vm_area_struct *vma;
1275 unsigned long start_addr;
1276
1277 if (len > TASK_SIZE)
1278 return -ENOMEM;
1279
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001280 if (flags & MAP_FIXED)
1281 return addr;
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if (addr) {
1284 addr = PAGE_ALIGN(addr);
1285 vma = find_vma(mm, addr);
1286 if (TASK_SIZE - len >= addr &&
1287 (!vma || addr + len <= vma->vm_start))
1288 return addr;
1289 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001290 if (len > mm->cached_hole_size) {
1291 start_addr = addr = mm->free_area_cache;
1292 } else {
1293 start_addr = addr = TASK_UNMAPPED_BASE;
1294 mm->cached_hole_size = 0;
1295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297full_search:
1298 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1299 /* At this point: (!vma || addr < vma->vm_end). */
1300 if (TASK_SIZE - len < addr) {
1301 /*
1302 * Start a new search - just in case we missed
1303 * some holes.
1304 */
1305 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001306 addr = TASK_UNMAPPED_BASE;
1307 start_addr = addr;
1308 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 goto full_search;
1310 }
1311 return -ENOMEM;
1312 }
1313 if (!vma || addr + len <= vma->vm_start) {
1314 /*
1315 * Remember the place where we stopped the search:
1316 */
1317 mm->free_area_cache = addr + len;
1318 return addr;
1319 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001320 if (addr + mm->cached_hole_size < vma->vm_start)
1321 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 addr = vma->vm_end;
1323 }
1324}
1325#endif
1326
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001327void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
1329 /*
1330 * Is this a new hole at the lowest possible address?
1331 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001332 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1333 mm->free_area_cache = addr;
1334 mm->cached_hole_size = ~0UL;
1335 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
1337
1338/*
1339 * This mmap-allocator allocates new areas top-down from below the
1340 * stack's low limit (the base):
1341 */
1342#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1343unsigned long
1344arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1345 const unsigned long len, const unsigned long pgoff,
1346 const unsigned long flags)
1347{
1348 struct vm_area_struct *vma;
1349 struct mm_struct *mm = current->mm;
1350 unsigned long addr = addr0;
1351
1352 /* requested length too big for entire address space */
1353 if (len > TASK_SIZE)
1354 return -ENOMEM;
1355
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001356 if (flags & MAP_FIXED)
1357 return addr;
1358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 /* requesting a specific address */
1360 if (addr) {
1361 addr = PAGE_ALIGN(addr);
1362 vma = find_vma(mm, addr);
1363 if (TASK_SIZE - len >= addr &&
1364 (!vma || addr + len <= vma->vm_start))
1365 return addr;
1366 }
1367
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001368 /* check if free_area_cache is useful for us */
1369 if (len <= mm->cached_hole_size) {
1370 mm->cached_hole_size = 0;
1371 mm->free_area_cache = mm->mmap_base;
1372 }
1373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 /* either no address requested or can't fit in requested address hole */
1375 addr = mm->free_area_cache;
1376
1377 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001378 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 vma = find_vma(mm, addr-len);
1380 if (!vma || addr <= vma->vm_start)
1381 /* remember the address as a hint for next time */
1382 return (mm->free_area_cache = addr-len);
1383 }
1384
Chris Wright73219d12005-06-21 17:14:52 -07001385 if (mm->mmap_base < len)
1386 goto bottomup;
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 addr = mm->mmap_base-len;
1389
1390 do {
1391 /*
1392 * Lookup failure means no vma is above this address,
1393 * else if new region fits below vma->vm_start,
1394 * return with success:
1395 */
1396 vma = find_vma(mm, addr);
1397 if (!vma || addr+len <= vma->vm_start)
1398 /* remember the address as a hint for next time */
1399 return (mm->free_area_cache = addr);
1400
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001401 /* remember the largest hole we saw so far */
1402 if (addr + mm->cached_hole_size < vma->vm_start)
1403 mm->cached_hole_size = vma->vm_start - addr;
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 /* try just below the current vma->vm_start */
1406 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001407 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Chris Wright73219d12005-06-21 17:14:52 -07001409bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 /*
1411 * A failed mmap() very likely causes application failure,
1412 * so fall back to the bottom-up function here. This scenario
1413 * can happen with large stack limits and large mmap()
1414 * allocations.
1415 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001416 mm->cached_hole_size = ~0UL;
1417 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1419 /*
1420 * Restore the topdown base:
1421 */
1422 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001423 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425 return addr;
1426}
1427#endif
1428
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001429void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
1431 /*
1432 * Is this a new hole at the highest possible address?
1433 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001434 if (addr > mm->free_area_cache)
1435 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
1437 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001438 if (mm->free_area_cache > mm->mmap_base)
1439 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440}
1441
1442unsigned long
1443get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1444 unsigned long pgoff, unsigned long flags)
1445{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001446 unsigned long (*get_area)(struct file *, unsigned long,
1447 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001449 get_area = current->mm->get_unmapped_area;
1450 if (file && file->f_op && file->f_op->get_unmapped_area)
1451 get_area = file->f_op->get_unmapped_area;
1452 addr = get_area(file, addr, len, pgoff, flags);
1453 if (IS_ERR_VALUE(addr))
1454 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001456 if (addr > TASK_SIZE - len)
1457 return -ENOMEM;
1458 if (addr & ~PAGE_MASK)
1459 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001460
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001461 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
1463
1464EXPORT_SYMBOL(get_unmapped_area);
1465
1466/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001467struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
1469 struct vm_area_struct *vma = NULL;
1470
1471 if (mm) {
1472 /* Check the cache first. */
1473 /* (Cache hit rate is typically around 35%.) */
1474 vma = mm->mmap_cache;
1475 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1476 struct rb_node * rb_node;
1477
1478 rb_node = mm->mm_rb.rb_node;
1479 vma = NULL;
1480
1481 while (rb_node) {
1482 struct vm_area_struct * vma_tmp;
1483
1484 vma_tmp = rb_entry(rb_node,
1485 struct vm_area_struct, vm_rb);
1486
1487 if (vma_tmp->vm_end > addr) {
1488 vma = vma_tmp;
1489 if (vma_tmp->vm_start <= addr)
1490 break;
1491 rb_node = rb_node->rb_left;
1492 } else
1493 rb_node = rb_node->rb_right;
1494 }
1495 if (vma)
1496 mm->mmap_cache = vma;
1497 }
1498 }
1499 return vma;
1500}
1501
1502EXPORT_SYMBOL(find_vma);
1503
1504/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1505struct vm_area_struct *
1506find_vma_prev(struct mm_struct *mm, unsigned long addr,
1507 struct vm_area_struct **pprev)
1508{
1509 struct vm_area_struct *vma = NULL, *prev = NULL;
ZhenwenXu48aae422009-01-06 14:40:21 -08001510 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 if (!mm)
1512 goto out;
1513
1514 /* Guard against addr being lower than the first VMA */
1515 vma = mm->mmap;
1516
1517 /* Go through the RB tree quickly. */
1518 rb_node = mm->mm_rb.rb_node;
1519
1520 while (rb_node) {
1521 struct vm_area_struct *vma_tmp;
1522 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1523
1524 if (addr < vma_tmp->vm_end) {
1525 rb_node = rb_node->rb_left;
1526 } else {
1527 prev = vma_tmp;
1528 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1529 break;
1530 rb_node = rb_node->rb_right;
1531 }
1532 }
1533
1534out:
1535 *pprev = prev;
1536 return prev ? prev->vm_next : vma;
1537}
1538
1539/*
1540 * Verify that the stack growth is acceptable and
1541 * update accounting. This is shared with both the
1542 * grow-up and grow-down cases.
1543 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001544static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545{
1546 struct mm_struct *mm = vma->vm_mm;
1547 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001548 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001551 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 return -ENOMEM;
1553
1554 /* Stack limit test */
1555 if (size > rlim[RLIMIT_STACK].rlim_cur)
1556 return -ENOMEM;
1557
1558 /* mlock limit tests */
1559 if (vma->vm_flags & VM_LOCKED) {
1560 unsigned long locked;
1561 unsigned long limit;
1562 locked = mm->locked_vm + grow;
1563 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1564 if (locked > limit && !capable(CAP_IPC_LOCK))
1565 return -ENOMEM;
1566 }
1567
Adam Litke0d59a012007-01-30 14:35:39 -08001568 /* Check to ensure the stack will not grow into a hugetlb-only region */
1569 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1570 vma->vm_end - size;
1571 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1572 return -EFAULT;
1573
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 /*
1575 * Overcommit.. This must be the final test, as it will
1576 * update security statistics.
1577 */
1578 if (security_vm_enough_memory(grow))
1579 return -ENOMEM;
1580
1581 /* Ok, everything looks good - let it rip */
1582 mm->total_vm += grow;
1583 if (vma->vm_flags & VM_LOCKED)
1584 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001585 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 return 0;
1587}
1588
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001589#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001591 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1592 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001594#ifndef CONFIG_IA64
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001595static
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001596#endif
1597int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598{
1599 int error;
1600
1601 if (!(vma->vm_flags & VM_GROWSUP))
1602 return -EFAULT;
1603
1604 /*
1605 * We must make sure the anon_vma is allocated
1606 * so that the anon_vma locking is not a noop.
1607 */
1608 if (unlikely(anon_vma_prepare(vma)))
1609 return -ENOMEM;
1610 anon_vma_lock(vma);
1611
1612 /*
1613 * vma->vm_start/vm_end cannot change under us because the caller
1614 * is required to hold the mmap_sem in read mode. We need the
1615 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001616 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 */
Helge Deller06b32f32006-12-19 19:28:33 +01001618 if (address < PAGE_ALIGN(address+4))
1619 address = PAGE_ALIGN(address+4);
1620 else {
1621 anon_vma_unlock(vma);
1622 return -ENOMEM;
1623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 error = 0;
1625
1626 /* Somebody else might have raced and expanded it already */
1627 if (address > vma->vm_end) {
1628 unsigned long size, grow;
1629
1630 size = address - vma->vm_start;
1631 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1632
1633 error = acct_stack_growth(vma, size, grow);
1634 if (!error)
1635 vma->vm_end = address;
1636 }
1637 anon_vma_unlock(vma);
1638 return error;
1639}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001640#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1641
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642/*
1643 * vma is the first one with address < vma->vm_start. Have to extend vma.
1644 */
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001645static int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001646 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 int error;
1649
1650 /*
1651 * We must make sure the anon_vma is allocated
1652 * so that the anon_vma locking is not a noop.
1653 */
1654 if (unlikely(anon_vma_prepare(vma)))
1655 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001656
1657 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001658 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001659 if (error)
1660 return error;
1661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 anon_vma_lock(vma);
1663
1664 /*
1665 * vma->vm_start/vm_end cannot change under us because the caller
1666 * is required to hold the mmap_sem in read mode. We need the
1667 * anon_vma lock to serialize against concurrent expand_stacks.
1668 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
1670 /* Somebody else might have raced and expanded it already */
1671 if (address < vma->vm_start) {
1672 unsigned long size, grow;
1673
1674 size = vma->vm_end - address;
1675 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1676
1677 error = acct_stack_growth(vma, size, grow);
1678 if (!error) {
1679 vma->vm_start = address;
1680 vma->vm_pgoff -= grow;
1681 }
1682 }
1683 anon_vma_unlock(vma);
1684 return error;
1685}
1686
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001687int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1688{
1689 return expand_downwards(vma, address);
1690}
1691
1692#ifdef CONFIG_STACK_GROWSUP
1693int expand_stack(struct vm_area_struct *vma, unsigned long address)
1694{
1695 return expand_upwards(vma, address);
1696}
1697
1698struct vm_area_struct *
1699find_extend_vma(struct mm_struct *mm, unsigned long addr)
1700{
1701 struct vm_area_struct *vma, *prev;
1702
1703 addr &= PAGE_MASK;
1704 vma = find_vma_prev(mm, addr, &prev);
1705 if (vma && (vma->vm_start <= addr))
1706 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001707 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001708 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001709 if (prev->vm_flags & VM_LOCKED) {
1710 if (mlock_vma_pages_range(prev, addr, prev->vm_end) < 0)
1711 return NULL; /* vma gone! */
1712 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001713 return prev;
1714}
1715#else
1716int expand_stack(struct vm_area_struct *vma, unsigned long address)
1717{
1718 return expand_downwards(vma, address);
1719}
1720
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721struct vm_area_struct *
1722find_extend_vma(struct mm_struct * mm, unsigned long addr)
1723{
1724 struct vm_area_struct * vma;
1725 unsigned long start;
1726
1727 addr &= PAGE_MASK;
1728 vma = find_vma(mm,addr);
1729 if (!vma)
1730 return NULL;
1731 if (vma->vm_start <= addr)
1732 return vma;
1733 if (!(vma->vm_flags & VM_GROWSDOWN))
1734 return NULL;
1735 start = vma->vm_start;
1736 if (expand_stack(vma, addr))
1737 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001738 if (vma->vm_flags & VM_LOCKED) {
1739 if (mlock_vma_pages_range(vma, addr, start) < 0)
1740 return NULL; /* vma gone! */
1741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 return vma;
1743}
1744#endif
1745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001747 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001749 *
1750 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001752static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001754 /* Update high watermark before we lower total_vm */
1755 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001757 long nrpages = vma_pages(vma);
1758
1759 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001760 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001761 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001762 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 validate_mm(mm);
1764}
1765
1766/*
1767 * Get rid of page table information in the indicated region.
1768 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001769 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 */
1771static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001772 struct vm_area_struct *vma, struct vm_area_struct *prev,
1773 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001775 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 struct mmu_gather *tlb;
1777 unsigned long nr_accounted = 0;
1778
1779 lru_add_drain();
1780 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001781 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001782 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 vm_unacct_memory(nr_accounted);
Jan Beulich42b77722008-07-23 21:27:10 -07001784 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001785 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 tlb_finish_mmu(tlb, start, end);
1787}
1788
1789/*
1790 * Create a list of vma's touched by the unmap, removing them from the mm's
1791 * vma list as we go..
1792 */
1793static void
1794detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1795 struct vm_area_struct *prev, unsigned long end)
1796{
1797 struct vm_area_struct **insertion_point;
1798 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001799 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
1801 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1802 do {
1803 rb_erase(&vma->vm_rb, &mm->mm_rb);
1804 mm->map_count--;
1805 tail_vma = vma;
1806 vma = vma->vm_next;
1807 } while (vma && vma->vm_start < end);
1808 *insertion_point = vma;
1809 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001810 if (mm->unmap_area == arch_unmap_area)
1811 addr = prev ? prev->vm_end : mm->mmap_base;
1812 else
1813 addr = vma ? vma->vm_start : mm->mmap_base;
1814 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 mm->mmap_cache = NULL; /* Kill the cache. */
1816}
1817
1818/*
1819 * Split a vma into two pieces at address 'addr', a new vma is allocated
Michael Opdenacker59c51592007-05-09 08:57:56 +02001820 * either for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 */
1822int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1823 unsigned long addr, int new_below)
1824{
1825 struct mempolicy *pol;
1826 struct vm_area_struct *new;
1827
Andi Kleena5516432008-07-23 21:27:41 -07001828 if (is_vm_hugetlb_page(vma) && (addr &
1829 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 return -EINVAL;
1831
1832 if (mm->map_count >= sysctl_max_map_count)
1833 return -ENOMEM;
1834
Christoph Lametere94b1762006-12-06 20:33:17 -08001835 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (!new)
1837 return -ENOMEM;
1838
1839 /* most fields are the same, copy all, and then fixup */
1840 *new = *vma;
1841
1842 if (new_below)
1843 new->vm_end = addr;
1844 else {
1845 new->vm_start = addr;
1846 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1847 }
1848
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001849 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 if (IS_ERR(pol)) {
1851 kmem_cache_free(vm_area_cachep, new);
1852 return PTR_ERR(pol);
1853 }
1854 vma_set_policy(new, pol);
1855
Matt Helsley925d1c42008-04-29 01:01:36 -07001856 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001858 if (vma->vm_flags & VM_EXECUTABLE)
1859 added_exe_file_vma(mm);
1860 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
1862 if (new->vm_ops && new->vm_ops->open)
1863 new->vm_ops->open(new);
1864
1865 if (new_below)
1866 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1867 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1868 else
1869 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1870
1871 return 0;
1872}
1873
1874/* Munmap is split into 2 main parts -- this part which finds
1875 * what needs doing, and the areas themselves, which do the
1876 * work. This now handles partial unmappings.
1877 * Jeremy Fitzhardinge <jeremy@goop.org>
1878 */
1879int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1880{
1881 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001882 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
1884 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1885 return -EINVAL;
1886
1887 if ((len = PAGE_ALIGN(len)) == 0)
1888 return -EINVAL;
1889
1890 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001891 vma = find_vma_prev(mm, start, &prev);
1892 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001894 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896 /* if it doesn't overlap, we have nothing.. */
1897 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001898 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 return 0;
1900
1901 /*
1902 * If we need to split any vma, do it now to save pain later.
1903 *
1904 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1905 * unmapped vm_area_struct will remain in use: so lower split_vma
1906 * places tmp vma above, and higher split_vma places tmp vma below.
1907 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001908 if (start > vma->vm_start) {
1909 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 if (error)
1911 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001912 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 }
1914
1915 /* Does it split the last one? */
1916 last = find_vma(mm, end);
1917 if (last && end > last->vm_start) {
1918 int error = split_vma(mm, last, end, 1);
1919 if (error)
1920 return error;
1921 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001922 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 /*
Rik van Rielba470de2008-10-18 20:26:50 -07001925 * unlock any mlock()ed ranges before detaching vmas
1926 */
1927 if (mm->locked_vm) {
1928 struct vm_area_struct *tmp = vma;
1929 while (tmp && tmp->vm_start < end) {
1930 if (tmp->vm_flags & VM_LOCKED) {
1931 mm->locked_vm -= vma_pages(tmp);
1932 munlock_vma_pages_all(tmp);
1933 }
1934 tmp = tmp->vm_next;
1935 }
1936 }
1937
1938 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 * Remove the vma's, and unmap the actual pages
1940 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001941 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1942 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001945 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
1947 return 0;
1948}
1949
1950EXPORT_SYMBOL(do_munmap);
1951
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001952SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953{
1954 int ret;
1955 struct mm_struct *mm = current->mm;
1956
1957 profile_munmap(addr);
1958
1959 down_write(&mm->mmap_sem);
1960 ret = do_munmap(mm, addr, len);
1961 up_write(&mm->mmap_sem);
1962 return ret;
1963}
1964
1965static inline void verify_mm_writelocked(struct mm_struct *mm)
1966{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001967#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1969 WARN_ON(1);
1970 up_read(&mm->mmap_sem);
1971 }
1972#endif
1973}
1974
1975/*
1976 * this is really a simplified "do_mmap". it only handles
1977 * anonymous maps. eventually we may be able to do some
1978 * brk-specific accounting here.
1979 */
1980unsigned long do_brk(unsigned long addr, unsigned long len)
1981{
1982 struct mm_struct * mm = current->mm;
1983 struct vm_area_struct * vma, * prev;
1984 unsigned long flags;
1985 struct rb_node ** rb_link, * rb_parent;
1986 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04001987 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
1989 len = PAGE_ALIGN(len);
1990 if (!len)
1991 return addr;
1992
1993 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1994 return -EINVAL;
1995
Hugh Dickinscd2579d2006-11-14 13:43:38 +00001996 if (is_hugepage_only_range(mm, addr, len))
1997 return -EINVAL;
David Gibsoncb07c9a2006-11-14 02:03:38 -08001998
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001999 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002000 if (error)
2001 return error;
2002
Kirill Korotaev3a459752006-09-07 14:17:04 +04002003 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2004
2005 error = arch_mmap_check(addr, len, flags);
2006 if (error)
2007 return error;
2008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 /*
2010 * mlock MCL_FUTURE?
2011 */
2012 if (mm->def_flags & VM_LOCKED) {
2013 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002014 locked = len >> PAGE_SHIFT;
2015 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07002017 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2019 return -EAGAIN;
2020 }
2021
2022 /*
2023 * mm->mmap_sem is required to protect against another thread
2024 * changing the mappings in case we sleep.
2025 */
2026 verify_mm_writelocked(mm);
2027
2028 /*
2029 * Clear old maps. this also does some error checking for us
2030 */
2031 munmap_back:
2032 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2033 if (vma && vma->vm_start < addr + len) {
2034 if (do_munmap(mm, addr, len))
2035 return -ENOMEM;
2036 goto munmap_back;
2037 }
2038
2039 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002040 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 return -ENOMEM;
2042
2043 if (mm->map_count > sysctl_max_map_count)
2044 return -ENOMEM;
2045
2046 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2047 return -ENOMEM;
2048
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002050 vma = vma_merge(mm, prev, addr, addr + len, flags,
2051 NULL, NULL, pgoff, NULL);
2052 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 goto out;
2054
2055 /*
2056 * create a vma struct for an anonymous mapping
2057 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002058 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 if (!vma) {
2060 vm_unacct_memory(len >> PAGE_SHIFT);
2061 return -ENOMEM;
2062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
2064 vma->vm_mm = mm;
2065 vma->vm_start = addr;
2066 vma->vm_end = addr + len;
2067 vma->vm_pgoff = pgoff;
2068 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002069 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 vma_link(mm, vma, prev, rb_link, rb_parent);
2071out:
2072 mm->total_vm += len >> PAGE_SHIFT;
2073 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002074 if (!mlock_vma_pages_range(vma, addr, addr + len))
2075 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 }
2077 return addr;
2078}
2079
2080EXPORT_SYMBOL(do_brk);
2081
2082/* Release all mmaps. */
2083void exit_mmap(struct mm_struct *mm)
2084{
2085 struct mmu_gather *tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002086 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002088 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002090 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002091 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002092
Rik van Rielba470de2008-10-18 20:26:50 -07002093 if (mm->locked_vm) {
2094 vma = mm->mmap;
2095 while (vma) {
2096 if (vma->vm_flags & VM_LOCKED)
2097 munlock_vma_pages_all(vma);
2098 vma = vma->vm_next;
2099 }
2100 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002101
2102 arch_exit_mmap(mm);
2103
Rik van Rielba470de2008-10-18 20:26:50 -07002104 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002105 if (!vma) /* Can happen if dup_mmap() received an OOM */
2106 return;
2107
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002110 tlb = tlb_gather_mmu(mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002111 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002112 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002113 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 vm_unacct_memory(nr_accounted);
Jan Beulich42b77722008-07-23 21:27:10 -07002115 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07002116 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002119 * Walk the list again, actually closing and freeing it,
2120 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002122 while (vma)
2123 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002124
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002125 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126}
2127
2128/* Insert vm structure into process list sorted by address
2129 * and into the inode's i_mmap tree. If vm_file is non-NULL
2130 * then i_mmap_lock is taken here.
2131 */
2132int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2133{
2134 struct vm_area_struct * __vma, * prev;
2135 struct rb_node ** rb_link, * rb_parent;
2136
2137 /*
2138 * The vm_pgoff of a purely anonymous vma should be irrelevant
2139 * until its first write fault, when page's anon_vma and index
2140 * are set. But now set the vm_pgoff it will almost certainly
2141 * end up with (unless mremap moves it elsewhere before that
2142 * first wfault), so /proc/pid/maps tells a consistent story.
2143 *
2144 * By setting it to reflect the virtual start address of the
2145 * vma, merges and splits can happen in a seamless way, just
2146 * using the existing file pgoff checks and manipulations.
2147 * Similarly in do_mmap_pgoff and in do_brk.
2148 */
2149 if (!vma->vm_file) {
2150 BUG_ON(vma->anon_vma);
2151 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2152 }
2153 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2154 if (__vma && __vma->vm_start < vma->vm_end)
2155 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002156 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002157 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002158 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 vma_link(mm, vma, prev, rb_link, rb_parent);
2160 return 0;
2161}
2162
2163/*
2164 * Copy the vma structure to a new location in the same mm,
2165 * prior to moving page table entries, to effect an mremap move.
2166 */
2167struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2168 unsigned long addr, unsigned long len, pgoff_t pgoff)
2169{
2170 struct vm_area_struct *vma = *vmap;
2171 unsigned long vma_start = vma->vm_start;
2172 struct mm_struct *mm = vma->vm_mm;
2173 struct vm_area_struct *new_vma, *prev;
2174 struct rb_node **rb_link, *rb_parent;
2175 struct mempolicy *pol;
2176
2177 /*
2178 * If anonymous vma has not yet been faulted, update new pgoff
2179 * to match new location, to increase its chance of merging.
2180 */
2181 if (!vma->vm_file && !vma->anon_vma)
2182 pgoff = addr >> PAGE_SHIFT;
2183
2184 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2185 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2186 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2187 if (new_vma) {
2188 /*
2189 * Source vma may have been merged into new_vma
2190 */
2191 if (vma_start >= new_vma->vm_start &&
2192 vma_start < new_vma->vm_end)
2193 *vmap = new_vma;
2194 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002195 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 if (new_vma) {
2197 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002198 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 if (IS_ERR(pol)) {
2200 kmem_cache_free(vm_area_cachep, new_vma);
2201 return NULL;
2202 }
2203 vma_set_policy(new_vma, pol);
2204 new_vma->vm_start = addr;
2205 new_vma->vm_end = addr + len;
2206 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002207 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002209 if (vma->vm_flags & VM_EXECUTABLE)
2210 added_exe_file_vma(mm);
2211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 if (new_vma->vm_ops && new_vma->vm_ops->open)
2213 new_vma->vm_ops->open(new_vma);
2214 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2215 }
2216 }
2217 return new_vma;
2218}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002219
2220/*
2221 * Return true if the calling process may expand its vm space by the passed
2222 * number of pages
2223 */
2224int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2225{
2226 unsigned long cur = mm->total_vm; /* pages */
2227 unsigned long lim;
2228
2229 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2230
2231 if (cur + npages > lim)
2232 return 0;
2233 return 1;
2234}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002235
2236
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002237static int special_mapping_fault(struct vm_area_struct *vma,
2238 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002239{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002240 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002241 struct page **pages;
2242
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002243 /*
2244 * special mappings have no vm_file, and in that case, the mm
2245 * uses vm_pgoff internally. So we have to subtract it from here.
2246 * We are allowed to do this because we are the mm; do not copy
2247 * this code into drivers!
2248 */
2249 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002250
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002251 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2252 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002253
2254 if (*pages) {
2255 struct page *page = *pages;
2256 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002257 vmf->page = page;
2258 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002259 }
2260
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002261 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002262}
2263
2264/*
2265 * Having a close hook prevents vma merging regardless of flags.
2266 */
2267static void special_mapping_close(struct vm_area_struct *vma)
2268{
2269}
2270
2271static struct vm_operations_struct special_mapping_vmops = {
2272 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002273 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002274};
2275
2276/*
2277 * Called with mm->mmap_sem held for writing.
2278 * Insert a new vma covering the given region, with the given flags.
2279 * Its pages are supplied by the given array of struct page *.
2280 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2281 * The region past the last page supplied will always produce SIGBUS.
2282 * The array pointer and the pages it points to are assumed to stay alive
2283 * for as long as this mapping might exist.
2284 */
2285int install_special_mapping(struct mm_struct *mm,
2286 unsigned long addr, unsigned long len,
2287 unsigned long vm_flags, struct page **pages)
2288{
2289 struct vm_area_struct *vma;
2290
2291 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2292 if (unlikely(vma == NULL))
2293 return -ENOMEM;
2294
2295 vma->vm_mm = mm;
2296 vma->vm_start = addr;
2297 vma->vm_end = addr + len;
2298
Nick Piggin2f987352008-02-02 03:08:53 +01002299 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002300 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002301
2302 vma->vm_ops = &special_mapping_vmops;
2303 vma->vm_private_data = pages;
2304
2305 if (unlikely(insert_vm_struct(mm, vma))) {
2306 kmem_cache_free(vm_area_cachep, vma);
2307 return -ENOMEM;
2308 }
2309
2310 mm->total_vm += len >> PAGE_SHIFT;
2311
2312 return 0;
2313}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002314
2315static DEFINE_MUTEX(mm_all_locks_mutex);
2316
Peter Zijlstra454ed842008-08-11 09:30:25 +02002317static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002318{
2319 if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2320 /*
2321 * The LSB of head.next can't change from under us
2322 * because we hold the mm_all_locks_mutex.
2323 */
Peter Zijlstra454ed842008-08-11 09:30:25 +02002324 spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002325 /*
2326 * We can safely modify head.next after taking the
2327 * anon_vma->lock. If some other vma in this mm shares
2328 * the same anon_vma we won't take it again.
2329 *
2330 * No need of atomic instructions here, head.next
2331 * can't change from under us thanks to the
2332 * anon_vma->lock.
2333 */
2334 if (__test_and_set_bit(0, (unsigned long *)
2335 &anon_vma->head.next))
2336 BUG();
2337 }
2338}
2339
Peter Zijlstra454ed842008-08-11 09:30:25 +02002340static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002341{
2342 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2343 /*
2344 * AS_MM_ALL_LOCKS can't change from under us because
2345 * we hold the mm_all_locks_mutex.
2346 *
2347 * Operations on ->flags have to be atomic because
2348 * even if AS_MM_ALL_LOCKS is stable thanks to the
2349 * mm_all_locks_mutex, there may be other cpus
2350 * changing other bitflags in parallel to us.
2351 */
2352 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2353 BUG();
Peter Zijlstra454ed842008-08-11 09:30:25 +02002354 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002355 }
2356}
2357
2358/*
2359 * This operation locks against the VM for all pte/vma/mm related
2360 * operations that could ever happen on a certain mm. This includes
2361 * vmtruncate, try_to_unmap, and all page faults.
2362 *
2363 * The caller must take the mmap_sem in write mode before calling
2364 * mm_take_all_locks(). The caller isn't allowed to release the
2365 * mmap_sem until mm_drop_all_locks() returns.
2366 *
2367 * mmap_sem in write mode is required in order to block all operations
2368 * that could modify pagetables and free pages without need of
2369 * altering the vma layout (for example populate_range() with
2370 * nonlinear vmas). It's also needed in write mode to avoid new
2371 * anon_vmas to be associated with existing vmas.
2372 *
2373 * A single task can't take more than one mm_take_all_locks() in a row
2374 * or it would deadlock.
2375 *
2376 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2377 * mapping->flags avoid to take the same lock twice, if more than one
2378 * vma in this mm is backed by the same anon_vma or address_space.
2379 *
2380 * We can take all the locks in random order because the VM code
2381 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2382 * takes more than one of them in a row. Secondly we're protected
2383 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2384 *
2385 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2386 * that may have to take thousand of locks.
2387 *
2388 * mm_take_all_locks() can fail if it's interrupted by signals.
2389 */
2390int mm_take_all_locks(struct mm_struct *mm)
2391{
2392 struct vm_area_struct *vma;
2393 int ret = -EINTR;
2394
2395 BUG_ON(down_read_trylock(&mm->mmap_sem));
2396
2397 mutex_lock(&mm_all_locks_mutex);
2398
2399 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2400 if (signal_pending(current))
2401 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002402 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002403 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002404 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002405
2406 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2407 if (signal_pending(current))
2408 goto out_unlock;
2409 if (vma->anon_vma)
2410 vm_lock_anon_vma(mm, vma->anon_vma);
2411 }
2412
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002413 ret = 0;
2414
2415out_unlock:
2416 if (ret)
2417 mm_drop_all_locks(mm);
2418
2419 return ret;
2420}
2421
2422static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2423{
2424 if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2425 /*
2426 * The LSB of head.next can't change to 0 from under
2427 * us because we hold the mm_all_locks_mutex.
2428 *
2429 * We must however clear the bitflag before unlocking
2430 * the vma so the users using the anon_vma->head will
2431 * never see our bitflag.
2432 *
2433 * No need of atomic instructions here, head.next
2434 * can't change from under us until we release the
2435 * anon_vma->lock.
2436 */
2437 if (!__test_and_clear_bit(0, (unsigned long *)
2438 &anon_vma->head.next))
2439 BUG();
2440 spin_unlock(&anon_vma->lock);
2441 }
2442}
2443
2444static void vm_unlock_mapping(struct address_space *mapping)
2445{
2446 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2447 /*
2448 * AS_MM_ALL_LOCKS can't change to 0 from under us
2449 * because we hold the mm_all_locks_mutex.
2450 */
2451 spin_unlock(&mapping->i_mmap_lock);
2452 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2453 &mapping->flags))
2454 BUG();
2455 }
2456}
2457
2458/*
2459 * The mmap_sem cannot be released by the caller until
2460 * mm_drop_all_locks() returns.
2461 */
2462void mm_drop_all_locks(struct mm_struct *mm)
2463{
2464 struct vm_area_struct *vma;
2465
2466 BUG_ON(down_read_trylock(&mm->mmap_sem));
2467 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2468
2469 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2470 if (vma->anon_vma)
2471 vm_unlock_anon_vma(vma->anon_vma);
2472 if (vma->vm_file && vma->vm_file->f_mapping)
2473 vm_unlock_mapping(vma->vm_file->f_mapping);
2474 }
2475
2476 mutex_unlock(&mm_all_locks_mutex);
2477}
David Howells8feae132009-01-08 12:04:47 +00002478
2479/*
2480 * initialise the VMA slab
2481 */
2482void __init mmap_init(void)
2483{
2484 vm_area_cachep = kmem_cache_create("vm_area_struct",
2485 sizeof(struct vm_area_struct), 0,
2486 SLAB_PANIC, NULL);
2487}