blob: 40d49986e7149ec415959d49a6ad7308f5af368f [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>
23#include <linux/hugetlb.h>
24#include <linux/profile.h>
25#include <linux/module.h>
26#include <linux/mount.h>
27#include <linux/mempolicy.h>
28#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070029#include <linux/mmu_notifier.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020030#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040031#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080032#include <linux/khugepaged.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34#include <asm/uaccess.h>
35#include <asm/cacheflush.h>
36#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020037#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Jan Beulich42b77722008-07-23 21:27:10 -070039#include "internal.h"
40
Kirill Korotaev3a459752006-09-07 14:17:04 +040041#ifndef arch_mmap_check
42#define arch_mmap_check(addr, len, flags) (0)
43#endif
44
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080045#ifndef arch_rebalance_pgtables
46#define arch_rebalance_pgtables(addr, len) (addr)
47#endif
48
Hugh Dickinse0da3822005-04-19 13:29:15 -070049static void unmap_region(struct mm_struct *mm,
50 struct vm_area_struct *vma, struct vm_area_struct *prev,
51 unsigned long start, unsigned long end);
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053/*
54 * WARNING: the debugging will use recursive algorithms so never enable this
55 * unless you know what you are doing.
56 */
57#undef DEBUG_MM_RB
58
59/* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
62 *
63 * map_type prot
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
68 *
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 *
73 */
74pgprot_t protection_map[16] = {
75 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
76 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
77};
78
Hugh Dickins804af2c2006-07-26 21:39:49 +010079pgprot_t vm_get_page_prot(unsigned long vm_flags)
80{
Dave Kleikampb845f312008-07-08 00:28:51 +100081 return __pgprot(pgprot_val(protection_map[vm_flags &
82 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
83 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010084}
85EXPORT_SYMBOL(vm_get_page_prot);
86
Shaohua Li34679d72011-05-24 17:11:18 -070087int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
88int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070089int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Shaohua Li34679d72011-05-24 17:11:18 -070090/*
91 * Make sure vm_committed_as in one cacheline and not cacheline shared with
92 * other variables. It can be updated by several CPUs frequently.
93 */
94struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96/*
97 * Check that a process has enough memory to allocate a new virtual
98 * mapping. 0 means there is enough memory for the allocation to
99 * succeed and -ENOMEM implies there is not.
100 *
101 * We currently support three overcommit policies, which are set via the
102 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
103 *
104 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
105 * Additional code 2002 Jul 20 by Robert Love.
106 *
107 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
108 *
109 * Note this is a helper function intended to be used by LSMs which
110 * wish to use this logic.
111 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700112int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
114 unsigned long free, allowed;
115
116 vm_acct_memory(pages);
117
118 /*
119 * Sometimes we want to use more memory than we have
120 */
121 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
122 return 0;
123
124 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
125 unsigned long n;
126
Christoph Lameter347ce432006-06-30 01:55:35 -0700127 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 free += nr_swap_pages;
129
130 /*
131 * Any slabs which are created with the
132 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
133 * which are reclaimable, under pressure. The dentry
134 * cache and most inode caches should fall into this
135 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700136 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138 /*
139 * Leave the last 3% for root
140 */
141 if (!cap_sys_admin)
142 free -= free / 32;
143
144 if (free > pages)
145 return 0;
146
147 /*
148 * nr_free_pages() is very expensive on large systems,
149 * only call if we're about to fail.
150 */
151 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700152
153 /*
154 * Leave reserved pages. The pages are not for anonymous pages.
155 */
156 if (n <= totalreserve_pages)
157 goto error;
158 else
159 n -= totalreserve_pages;
160
161 /*
162 * Leave the last 3% for root
163 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 if (!cap_sys_admin)
165 n -= n / 32;
166 free += n;
167
168 if (free > pages)
169 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700170
171 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
173
174 allowed = (totalram_pages - hugetlb_total_pages())
175 * sysctl_overcommit_ratio / 100;
176 /*
177 * Leave the last 3% for root
178 */
179 if (!cap_sys_admin)
180 allowed -= allowed / 32;
181 allowed += total_swap_pages;
182
183 /* Don't let a single process grow too big:
184 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700185 if (mm)
186 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700188 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700190error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 vm_unacct_memory(pages);
192
193 return -ENOMEM;
194}
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196/*
197 * Requires inode->i_mapping->i_mmap_lock
198 */
199static void __remove_shared_vm_struct(struct vm_area_struct *vma,
200 struct file *file, struct address_space *mapping)
201{
202 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800203 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 if (vma->vm_flags & VM_SHARED)
205 mapping->i_mmap_writable--;
206
207 flush_dcache_mmap_lock(mapping);
208 if (unlikely(vma->vm_flags & VM_NONLINEAR))
209 list_del_init(&vma->shared.vm_set.list);
210 else
211 vma_prio_tree_remove(vma, &mapping->i_mmap);
212 flush_dcache_mmap_unlock(mapping);
213}
214
215/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700216 * Unlink a file-based vm structure from its prio_tree, to hide
217 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700219void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
221 struct file *file = vma->vm_file;
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 if (file) {
224 struct address_space *mapping = file->f_mapping;
225 spin_lock(&mapping->i_mmap_lock);
226 __remove_shared_vm_struct(vma, file, mapping);
227 spin_unlock(&mapping->i_mmap_lock);
228 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700229}
230
231/*
232 * Close a vm structure and free it, returning the next.
233 */
234static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
235{
236 struct vm_area_struct *next = vma->vm_next;
237
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700238 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 if (vma->vm_ops && vma->vm_ops->close)
240 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700241 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700242 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700243 if (vma->vm_flags & VM_EXECUTABLE)
244 removed_exe_file_vma(vma->vm_mm);
245 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700246 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700248 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100251SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
253 unsigned long rlim, retval;
254 unsigned long newbrk, oldbrk;
255 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700256 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 down_write(&mm->mmap_sem);
259
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700260#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800261 /*
262 * CONFIG_COMPAT_BRK can still be overridden by setting
263 * randomize_va_space to 2, which will still cause mm->start_brk
264 * to be arbitrarily shifted
265 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700266 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800267 min_brk = mm->start_brk;
268 else
269 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700270#else
271 min_brk = mm->start_brk;
272#endif
273 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700275
276 /*
277 * Check against rlimit here. If this check is done later after the test
278 * of oldbrk with newbrk then it can escape the test and let the data
279 * segment grow beyond its set limit the in case where the limit is
280 * not page aligned -Ram Gupta
281 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800282 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100283 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
284 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700285 goto out;
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 newbrk = PAGE_ALIGN(brk);
288 oldbrk = PAGE_ALIGN(mm->brk);
289 if (oldbrk == newbrk)
290 goto set_brk;
291
292 /* Always allow shrinking brk. */
293 if (brk <= mm->brk) {
294 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
295 goto set_brk;
296 goto out;
297 }
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 /* Check against existing mmap mappings. */
300 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
301 goto out;
302
303 /* Ok, looks good - let it rip. */
304 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
305 goto out;
306set_brk:
307 mm->brk = brk;
308out:
309 retval = mm->brk;
310 up_write(&mm->mmap_sem);
311 return retval;
312}
313
314#ifdef DEBUG_MM_RB
315static int browse_rb(struct rb_root *root)
316{
317 int i = 0, j;
318 struct rb_node *nd, *pn = NULL;
319 unsigned long prev = 0, pend = 0;
320
321 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
322 struct vm_area_struct *vma;
323 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
324 if (vma->vm_start < prev)
325 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
326 if (vma->vm_start < pend)
327 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
328 if (vma->vm_start > vma->vm_end)
329 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
330 i++;
331 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800332 prev = vma->vm_start;
333 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
335 j = 0;
336 for (nd = pn; nd; nd = rb_prev(nd)) {
337 j++;
338 }
339 if (i != j)
340 printk("backwards %d, forwards %d\n", j, i), i = 0;
341 return i;
342}
343
344void validate_mm(struct mm_struct *mm)
345{
346 int bug = 0;
347 int i = 0;
348 struct vm_area_struct *tmp = mm->mmap;
349 while (tmp) {
350 tmp = tmp->vm_next;
351 i++;
352 }
353 if (i != mm->map_count)
354 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
355 i = browse_rb(&mm->mm_rb);
356 if (i != mm->map_count)
357 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200358 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359}
360#else
361#define validate_mm(mm) do { } while (0)
362#endif
363
364static struct vm_area_struct *
365find_vma_prepare(struct mm_struct *mm, unsigned long addr,
366 struct vm_area_struct **pprev, struct rb_node ***rb_link,
367 struct rb_node ** rb_parent)
368{
369 struct vm_area_struct * vma;
370 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
371
372 __rb_link = &mm->mm_rb.rb_node;
373 rb_prev = __rb_parent = NULL;
374 vma = NULL;
375
376 while (*__rb_link) {
377 struct vm_area_struct *vma_tmp;
378
379 __rb_parent = *__rb_link;
380 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
381
382 if (vma_tmp->vm_end > addr) {
383 vma = vma_tmp;
384 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700385 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 __rb_link = &__rb_parent->rb_left;
387 } else {
388 rb_prev = __rb_parent;
389 __rb_link = &__rb_parent->rb_right;
390 }
391 }
392
393 *pprev = NULL;
394 if (rb_prev)
395 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
396 *rb_link = __rb_link;
397 *rb_parent = __rb_parent;
398 return vma;
399}
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
402 struct rb_node **rb_link, struct rb_node *rb_parent)
403{
404 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
405 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
406}
407
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700408static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
ZhenwenXu48aae422009-01-06 14:40:21 -0800410 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412 file = vma->vm_file;
413 if (file) {
414 struct address_space *mapping = file->f_mapping;
415
416 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800417 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 if (vma->vm_flags & VM_SHARED)
419 mapping->i_mmap_writable++;
420
421 flush_dcache_mmap_lock(mapping);
422 if (unlikely(vma->vm_flags & VM_NONLINEAR))
423 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
424 else
425 vma_prio_tree_insert(vma, &mapping->i_mmap);
426 flush_dcache_mmap_unlock(mapping);
427 }
428}
429
430static void
431__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
432 struct vm_area_struct *prev, struct rb_node **rb_link,
433 struct rb_node *rb_parent)
434{
435 __vma_link_list(mm, vma, prev, rb_parent);
436 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
439static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
440 struct vm_area_struct *prev, struct rb_node **rb_link,
441 struct rb_node *rb_parent)
442{
443 struct address_space *mapping = NULL;
444
445 if (vma->vm_file)
446 mapping = vma->vm_file->f_mapping;
447
448 if (mapping) {
449 spin_lock(&mapping->i_mmap_lock);
450 vma->vm_truncate_count = mapping->truncate_count;
451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
453 __vma_link(mm, vma, prev, rb_link, rb_parent);
454 __vma_link_file(vma);
455
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 if (mapping)
457 spin_unlock(&mapping->i_mmap_lock);
458
459 mm->map_count++;
460 validate_mm(mm);
461}
462
463/*
464 * Helper for vma_adjust in the split_vma insert case:
465 * insert vm structure into list and rbtree and anon_vma,
466 * but it has already been inserted into prio_tree earlier.
467 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800468static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
ZhenwenXu48aae422009-01-06 14:40:21 -0800470 struct vm_area_struct *__vma, *prev;
471 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200474 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 __vma_link(mm, vma, prev, rb_link, rb_parent);
476 mm->map_count++;
477}
478
479static inline void
480__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
481 struct vm_area_struct *prev)
482{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700483 struct vm_area_struct *next = vma->vm_next;
484
485 prev->vm_next = next;
486 if (next)
487 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 rb_erase(&vma->vm_rb, &mm->mm_rb);
489 if (mm->mmap_cache == vma)
490 mm->mmap_cache = prev;
491}
492
493/*
494 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
495 * is already present in an i_mmap tree without adjusting the tree.
496 * The following helper function should be used when such adjustments
497 * are necessary. The "insert" vma (if any) is to be inserted
498 * before we drop the necessary locks.
499 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800500int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
502{
503 struct mm_struct *mm = vma->vm_mm;
504 struct vm_area_struct *next = vma->vm_next;
505 struct vm_area_struct *importer = NULL;
506 struct address_space *mapping = NULL;
507 struct prio_tree_root *root = NULL;
Rik van Riel012f1802010-08-09 17:18:40 -0700508 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 long adjust_next = 0;
511 int remove_next = 0;
512
513 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700514 struct vm_area_struct *exporter = NULL;
515
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 if (end >= next->vm_end) {
517 /*
518 * vma expands, overlapping all the next, and
519 * perhaps the one after too (mprotect case 6).
520 */
521again: remove_next = 1 + (end > next->vm_end);
522 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700523 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 importer = vma;
525 } else if (end > next->vm_start) {
526 /*
527 * vma expands, overlapping part of the next:
528 * mprotect case 5 shifting the boundary up.
529 */
530 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700531 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 importer = vma;
533 } else if (end < vma->vm_end) {
534 /*
535 * vma shrinks, and !insert tells it's not
536 * split_vma inserting another: so it must be
537 * mprotect case 4 shifting the boundary down.
538 */
539 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700540 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 importer = next;
542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
Rik van Riel5beb4932010-03-05 13:42:07 -0800544 /*
545 * Easily overlooked: when mprotect shifts the boundary,
546 * make sure the expanding vma has anon_vma set if the
547 * shrinking vma had, to cover any anon pages imported.
548 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700549 if (exporter && exporter->anon_vma && !importer->anon_vma) {
550 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800551 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700552 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800553 }
554 }
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 if (file) {
557 mapping = file->f_mapping;
558 if (!(vma->vm_flags & VM_NONLINEAR))
559 root = &mapping->i_mmap;
560 spin_lock(&mapping->i_mmap_lock);
561 if (importer &&
562 vma->vm_truncate_count != next->vm_truncate_count) {
563 /*
564 * unmap_mapping_range might be in progress:
565 * ensure that the expanding vma is rescanned.
566 */
567 importer->vm_truncate_count = 0;
568 }
569 if (insert) {
570 insert->vm_truncate_count = vma->vm_truncate_count;
571 /*
572 * Put into prio_tree now, so instantiated pages
573 * are visible to arm/parisc __flush_dcache_page
574 * throughout; but we cannot insert into address
575 * space until vma start or end is updated.
576 */
577 __vma_link_file(insert);
578 }
579 }
580
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800581 vma_adjust_trans_huge(vma, start, end, adjust_next);
582
Rik van Riel012f1802010-08-09 17:18:40 -0700583 /*
584 * When changing only vma->vm_end, we don't really need anon_vma
585 * lock. This is a fairly rare case by itself, but the anon_vma
586 * lock may be shared between many sibling processes. Skipping
587 * the lock for brk adjustments makes a difference sometimes.
588 */
Shaohua Li5f70b962011-05-24 17:11:19 -0700589 if (vma->anon_vma && (importer || start != vma->vm_start)) {
Rik van Riel012f1802010-08-09 17:18:40 -0700590 anon_vma = vma->anon_vma;
591 anon_vma_lock(anon_vma);
592 }
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 if (root) {
595 flush_dcache_mmap_lock(mapping);
596 vma_prio_tree_remove(vma, root);
597 if (adjust_next)
598 vma_prio_tree_remove(next, root);
599 }
600
601 vma->vm_start = start;
602 vma->vm_end = end;
603 vma->vm_pgoff = pgoff;
604 if (adjust_next) {
605 next->vm_start += adjust_next << PAGE_SHIFT;
606 next->vm_pgoff += adjust_next;
607 }
608
609 if (root) {
610 if (adjust_next)
611 vma_prio_tree_insert(next, root);
612 vma_prio_tree_insert(vma, root);
613 flush_dcache_mmap_unlock(mapping);
614 }
615
616 if (remove_next) {
617 /*
618 * vma_merge has merged next into vma, and needs
619 * us to remove next before dropping the locks.
620 */
621 __vma_unlink(mm, next, vma);
622 if (file)
623 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 } else if (insert) {
625 /*
626 * split_vma has split insert from vma, and needs
627 * us to insert it before dropping the locks
628 * (it may either follow vma or precede it).
629 */
630 __insert_vm_struct(mm, insert);
631 }
632
Rik van Riel012f1802010-08-09 17:18:40 -0700633 if (anon_vma)
634 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 if (mapping)
636 spin_unlock(&mapping->i_mmap_lock);
637
638 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700639 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700641 if (next->vm_flags & VM_EXECUTABLE)
642 removed_exe_file_vma(mm);
643 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800644 if (next->anon_vma)
645 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700647 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 kmem_cache_free(vm_area_cachep, next);
649 /*
650 * In mprotect's case 6 (see comments on vma_merge),
651 * we must remove another next too. It would clutter
652 * up the code too much to do both in one go.
653 */
654 if (remove_next == 2) {
655 next = vma->vm_next;
656 goto again;
657 }
658 }
659
660 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800661
662 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665/*
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{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700672 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
673 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 return 0;
675 if (vma->vm_file != file)
676 return 0;
677 if (vma->vm_ops && vma->vm_ops->close)
678 return 0;
679 return 1;
680}
681
682static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700683 struct anon_vma *anon_vma2,
684 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
Shaohua Li965f55d2011-05-24 17:11:20 -0700686 /*
687 * The list_is_singular() test is to avoid merging VMA cloned from
688 * parents. This can improve scalability caused by anon_vma lock.
689 */
690 if ((!anon_vma1 || !anon_vma2) && (!vma ||
691 list_is_singular(&vma->anon_vma_chain)))
692 return 1;
693 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
696/*
697 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
698 * in front of (at a lower virtual address and file offset than) the vma.
699 *
700 * We cannot merge two vmas if they have differently assigned (non-NULL)
701 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
702 *
703 * We don't check here for the merged mmap wrapping around the end of pagecache
704 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
705 * wrap, nor mmaps which cover the final page at index -1UL.
706 */
707static int
708can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
709 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
710{
711 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700712 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (vma->vm_pgoff == vm_pgoff)
714 return 1;
715 }
716 return 0;
717}
718
719/*
720 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
721 * beyond (at a higher virtual address and file offset than) the vma.
722 *
723 * We cannot merge two vmas if they have differently assigned (non-NULL)
724 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
725 */
726static int
727can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
728 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
729{
730 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700731 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 pgoff_t vm_pglen;
733 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
734 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
735 return 1;
736 }
737 return 0;
738}
739
740/*
741 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
742 * whether that can be merged with its predecessor or its successor.
743 * Or both (it neatly fills a hole).
744 *
745 * In most cases - when called for mmap, brk or mremap - [addr,end) is
746 * certain not to be mapped by the time vma_merge is called; but when
747 * called for mprotect, it is certain to be already mapped (either at
748 * an offset within prev, or at the start of next), and the flags of
749 * this area are about to be changed to vm_flags - and the no-change
750 * case has already been eliminated.
751 *
752 * The following mprotect cases have to be considered, where AAAA is
753 * the area passed down from mprotect_fixup, never extending beyond one
754 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
755 *
756 * AAAA AAAA AAAA AAAA
757 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
758 * cannot merge might become might become might become
759 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
760 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
761 * mremap move: PPPPNNNNNNNN 8
762 * AAAA
763 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
764 * might become case 1 below case 2 below case 3 below
765 *
766 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
767 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
768 */
769struct vm_area_struct *vma_merge(struct mm_struct *mm,
770 struct vm_area_struct *prev, unsigned long addr,
771 unsigned long end, unsigned long vm_flags,
772 struct anon_vma *anon_vma, struct file *file,
773 pgoff_t pgoff, struct mempolicy *policy)
774{
775 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
776 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800777 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779 /*
780 * We later require that vma->vm_flags == vm_flags,
781 * so this tests vma->vm_flags & VM_SPECIAL, too.
782 */
783 if (vm_flags & VM_SPECIAL)
784 return NULL;
785
786 if (prev)
787 next = prev->vm_next;
788 else
789 next = mm->mmap;
790 area = next;
791 if (next && next->vm_end == end) /* cases 6, 7, 8 */
792 next = next->vm_next;
793
794 /*
795 * Can it merge with the predecessor?
796 */
797 if (prev && prev->vm_end == addr &&
798 mpol_equal(vma_policy(prev), policy) &&
799 can_vma_merge_after(prev, vm_flags,
800 anon_vma, file, pgoff)) {
801 /*
802 * OK, it can. Can we now merge in the successor as well?
803 */
804 if (next && end == next->vm_start &&
805 mpol_equal(policy, vma_policy(next)) &&
806 can_vma_merge_before(next, vm_flags,
807 anon_vma, file, pgoff+pglen) &&
808 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700809 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800811 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 next->vm_end, prev->vm_pgoff, NULL);
813 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800814 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800816 if (err)
817 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800818 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return prev;
820 }
821
822 /*
823 * Can this new request be merged in front of next?
824 */
825 if (next && end == next->vm_start &&
826 mpol_equal(policy, vma_policy(next)) &&
827 can_vma_merge_before(next, vm_flags,
828 anon_vma, file, pgoff+pglen)) {
829 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800830 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 addr, prev->vm_pgoff, NULL);
832 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800833 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800835 if (err)
836 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800837 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 return area;
839 }
840
841 return NULL;
842}
843
844/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700845 * Rough compatbility check to quickly see if it's even worth looking
846 * at sharing an anon_vma.
847 *
848 * They need to have the same vm_file, and the flags can only differ
849 * in things that mprotect may change.
850 *
851 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
852 * we can merge the two vma's. For example, we refuse to merge a vma if
853 * there is a vm_ops->close() function, because that indicates that the
854 * driver is doing some kind of reference counting. But that doesn't
855 * really matter for the anon_vma sharing case.
856 */
857static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
858{
859 return a->vm_end == b->vm_start &&
860 mpol_equal(vma_policy(a), vma_policy(b)) &&
861 a->vm_file == b->vm_file &&
862 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
863 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
864}
865
866/*
867 * Do some basic sanity checking to see if we can re-use the anon_vma
868 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
869 * the same as 'old', the other will be the new one that is trying
870 * to share the anon_vma.
871 *
872 * NOTE! This runs with mm_sem held for reading, so it is possible that
873 * the anon_vma of 'old' is concurrently in the process of being set up
874 * by another page fault trying to merge _that_. But that's ok: if it
875 * is being set up, that automatically means that it will be a singleton
876 * acceptable for merging, so we can do all of this optimistically. But
877 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
878 *
879 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
880 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
881 * is to return an anon_vma that is "complex" due to having gone through
882 * a fork).
883 *
884 * We also make sure that the two vma's are compatible (adjacent,
885 * and with the same memory policies). That's all stable, even with just
886 * a read lock on the mm_sem.
887 */
888static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
889{
890 if (anon_vma_compatible(a, b)) {
891 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
892
893 if (anon_vma && list_is_singular(&old->anon_vma_chain))
894 return anon_vma;
895 }
896 return NULL;
897}
898
899/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
901 * neighbouring vmas for a suitable anon_vma, before it goes off
902 * to allocate a new anon_vma. It checks because a repetitive
903 * sequence of mprotects and faults may otherwise lead to distinct
904 * anon_vmas being allocated, preventing vma merge in subsequent
905 * mprotect.
906 */
907struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
908{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700909 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 near = vma->vm_next;
913 if (!near)
914 goto try_prev;
915
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700916 anon_vma = reusable_anon_vma(near, vma, near);
917 if (anon_vma)
918 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919try_prev:
920 /*
921 * It is potentially slow to have to call find_vma_prev here.
922 * But it's only on the first write fault on the vma, not
923 * every time, and we could devise a way to avoid it later
924 * (e.g. stash info in next's anon_vma_node when assigning
925 * an anon_vma, or when trying vma_merge). Another time.
926 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200927 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (!near)
929 goto none;
930
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700931 anon_vma = reusable_anon_vma(near, near, vma);
932 if (anon_vma)
933 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934none:
935 /*
936 * There's no absolute need to look only at touching neighbours:
937 * we could search further afield for "compatible" anon_vmas.
938 * But it would probably just be a waste of time searching,
939 * or lead to too many vmas hanging off the same anon_vma.
940 * We're trying to allow mprotect remerging later on,
941 * not trying to minimize memory used for anon_vmas.
942 */
943 return NULL;
944}
945
946#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700947void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 struct file *file, long pages)
949{
950 const unsigned long stack_flags
951 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (file) {
954 mm->shared_vm += pages;
955 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
956 mm->exec_vm += pages;
957 } else if (flags & stack_flags)
958 mm->stack_vm += pages;
959 if (flags & (VM_RESERVED|VM_IO))
960 mm->reserved_vm += pages;
961}
962#endif /* CONFIG_PROC_FS */
963
964/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700965 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 */
967
ZhenwenXu48aae422009-01-06 14:40:21 -0800968unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 unsigned long len, unsigned long prot,
970 unsigned long flags, unsigned long pgoff)
971{
972 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 struct inode *inode;
974 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700976 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /*
979 * Does the application expect PROT_READ to imply PROT_EXEC?
980 *
981 * (the exception is when the underlying filesystem is noexec
982 * mounted, in which case we dont add PROT_EXEC.)
983 */
984 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800985 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 prot |= PROT_EXEC;
987
988 if (!len)
989 return -EINVAL;
990
Eric Paris7cd94142007-11-26 18:47:40 -0500991 if (!(flags & MAP_FIXED))
992 addr = round_hint_to_min(addr);
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 /* Careful about overflows.. */
995 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500996 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 return -ENOMEM;
998
999 /* offset overflow? */
1000 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1001 return -EOVERFLOW;
1002
1003 /* Too many mappings? */
1004 if (mm->map_count > sysctl_max_map_count)
1005 return -ENOMEM;
1006
1007 /* Obtain the address to map to. we verify (or select) it and ensure
1008 * that it represents a valid section of the address space.
1009 */
1010 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1011 if (addr & ~PAGE_MASK)
1012 return addr;
1013
1014 /* Do simple checking here so the lower-level routines won't have
1015 * to. we assume access permissions have been handled by the open
1016 * of the memory object, so we don't do any here.
1017 */
1018 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1019 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1020
Huang Shijiecdf7b342009-09-21 17:03:36 -07001021 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 if (!can_do_mlock())
1023 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 /* mlock MCL_FUTURE? */
1026 if (vm_flags & VM_LOCKED) {
1027 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001028 locked = len >> PAGE_SHIFT;
1029 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001030 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001031 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1033 return -EAGAIN;
1034 }
1035
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001036 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 if (file) {
1039 switch (flags & MAP_TYPE) {
1040 case MAP_SHARED:
1041 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1042 return -EACCES;
1043
1044 /*
1045 * Make sure we don't allow writing to an append-only
1046 * file..
1047 */
1048 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1049 return -EACCES;
1050
1051 /*
1052 * Make sure there are no mandatory locks on the file.
1053 */
1054 if (locks_verify_locked(inode))
1055 return -EAGAIN;
1056
1057 vm_flags |= VM_SHARED | VM_MAYSHARE;
1058 if (!(file->f_mode & FMODE_WRITE))
1059 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1060
1061 /* fall through */
1062 case MAP_PRIVATE:
1063 if (!(file->f_mode & FMODE_READ))
1064 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001065 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001066 if (vm_flags & VM_EXEC)
1067 return -EPERM;
1068 vm_flags &= ~VM_MAYEXEC;
1069 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001070
1071 if (!file->f_op || !file->f_op->mmap)
1072 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 break;
1074
1075 default:
1076 return -EINVAL;
1077 }
1078 } else {
1079 switch (flags & MAP_TYPE) {
1080 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001081 /*
1082 * Ignore pgoff.
1083 */
1084 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 vm_flags |= VM_SHARED | VM_MAYSHARE;
1086 break;
1087 case MAP_PRIVATE:
1088 /*
1089 * Set pgoff according to addr for anon_vma.
1090 */
1091 pgoff = addr >> PAGE_SHIFT;
1092 break;
1093 default:
1094 return -EINVAL;
1095 }
1096 }
1097
Eric Parised032182007-06-28 15:55:21 -04001098 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 if (error)
1100 return error;
Eric Parised032182007-06-28 15:55:21 -04001101
Mel Gorman5a6fe122009-02-10 14:02:27 +00001102 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001103}
1104EXPORT_SYMBOL(do_mmap_pgoff);
1105
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001106SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1107 unsigned long, prot, unsigned long, flags,
1108 unsigned long, fd, unsigned long, pgoff)
1109{
1110 struct file *file = NULL;
1111 unsigned long retval = -EBADF;
1112
1113 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001114 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001115 if (unlikely(flags & MAP_HUGETLB))
1116 return -EINVAL;
1117 file = fget(fd);
1118 if (!file)
1119 goto out;
1120 } else if (flags & MAP_HUGETLB) {
1121 struct user_struct *user = NULL;
1122 /*
1123 * VM_NORESERVE is used because the reservations will be
1124 * taken when vm_ops->mmap() is called
1125 * A dummy user value is used because we are not locking
1126 * memory so no accounting is necessary
1127 */
1128 len = ALIGN(len, huge_page_size(&default_hstate));
1129 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1130 &user, HUGETLB_ANONHUGE_INODE);
1131 if (IS_ERR(file))
1132 return PTR_ERR(file);
1133 }
1134
1135 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1136
1137 down_write(&current->mm->mmap_sem);
1138 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1139 up_write(&current->mm->mmap_sem);
1140
1141 if (file)
1142 fput(file);
1143out:
1144 return retval;
1145}
1146
Christoph Hellwiga4679372010-03-10 15:21:15 -08001147#ifdef __ARCH_WANT_SYS_OLD_MMAP
1148struct mmap_arg_struct {
1149 unsigned long addr;
1150 unsigned long len;
1151 unsigned long prot;
1152 unsigned long flags;
1153 unsigned long fd;
1154 unsigned long offset;
1155};
1156
1157SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1158{
1159 struct mmap_arg_struct a;
1160
1161 if (copy_from_user(&a, arg, sizeof(a)))
1162 return -EFAULT;
1163 if (a.offset & ~PAGE_MASK)
1164 return -EINVAL;
1165
1166 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1167 a.offset >> PAGE_SHIFT);
1168}
1169#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1170
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001171/*
1172 * Some shared mappigns will want the pages marked read-only
1173 * to track write events. If so, we'll downgrade vm_page_prot
1174 * to the private version (using protection_map[] without the
1175 * VM_SHARED bit).
1176 */
1177int vma_wants_writenotify(struct vm_area_struct *vma)
1178{
1179 unsigned int vm_flags = vma->vm_flags;
1180
1181 /* If it was private or non-writable, the write bit is already clear */
1182 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1183 return 0;
1184
1185 /* The backer wishes to know when pages are first written to? */
1186 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1187 return 1;
1188
1189 /* The open routine did something to the protections already? */
1190 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001191 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001192 return 0;
1193
1194 /* Specialty mapping? */
1195 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1196 return 0;
1197
1198 /* Can the mapping track the dirty pages? */
1199 return vma->vm_file && vma->vm_file->f_mapping &&
1200 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1201}
1202
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001203/*
1204 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001205 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001206 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001207static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001208{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001209 /*
1210 * hugetlb has its own accounting separate from the core VM
1211 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1212 */
1213 if (file && is_file_hugepages(file))
1214 return 0;
1215
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001216 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1217}
1218
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001219unsigned long mmap_region(struct file *file, unsigned long addr,
1220 unsigned long len, unsigned long flags,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001221 unsigned int vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001222{
1223 struct mm_struct *mm = current->mm;
1224 struct vm_area_struct *vma, *prev;
1225 int correct_wcount = 0;
1226 int error;
1227 struct rb_node **rb_link, *rb_parent;
1228 unsigned long charged = 0;
1229 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 /* Clear old maps */
1232 error = -ENOMEM;
1233munmap_back:
1234 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1235 if (vma && vma->vm_start < addr + len) {
1236 if (do_munmap(mm, addr, len))
1237 return -ENOMEM;
1238 goto munmap_back;
1239 }
1240
1241 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001242 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 return -ENOMEM;
1244
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001245 /*
1246 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001247 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001248 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001249 if ((flags & MAP_NORESERVE)) {
1250 /* We honor MAP_NORESERVE if allowed to overcommit */
1251 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1252 vm_flags |= VM_NORESERVE;
1253
1254 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1255 if (file && is_file_hugepages(file))
1256 vm_flags |= VM_NORESERVE;
1257 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001258
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001259 /*
1260 * Private writable mapping: check memory availability
1261 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001262 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001263 charged = len >> PAGE_SHIFT;
1264 if (security_vm_enough_memory(charged))
1265 return -ENOMEM;
1266 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
1268
1269 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001270 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001272 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1273 if (vma)
1274 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
1276 /*
1277 * Determine the object being mapped and call the appropriate
1278 * specific mapper. the address has already been validated, but
1279 * not unmapped, but the maps are removed from the list.
1280 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001281 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (!vma) {
1283 error = -ENOMEM;
1284 goto unacct_error;
1285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
1287 vma->vm_mm = mm;
1288 vma->vm_start = addr;
1289 vma->vm_end = addr + len;
1290 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001291 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001293 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
1295 if (file) {
1296 error = -EINVAL;
1297 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1298 goto free_vma;
1299 if (vm_flags & VM_DENYWRITE) {
1300 error = deny_write_access(file);
1301 if (error)
1302 goto free_vma;
1303 correct_wcount = 1;
1304 }
1305 vma->vm_file = file;
1306 get_file(file);
1307 error = file->f_op->mmap(file, vma);
1308 if (error)
1309 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001310 if (vm_flags & VM_EXECUTABLE)
1311 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001312
1313 /* Can addr have changed??
1314 *
1315 * Answer: Yes, several device drivers can do it in their
1316 * f_op->mmap method. -DaveM
1317 */
1318 addr = vma->vm_start;
1319 pgoff = vma->vm_pgoff;
1320 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 } else if (vm_flags & VM_SHARED) {
1322 error = shmem_zero_setup(vma);
1323 if (error)
1324 goto free_vma;
1325 }
1326
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001327 if (vma_wants_writenotify(vma)) {
1328 pgprot_t pprot = vma->vm_page_prot;
1329
1330 /* Can vma->vm_page_prot have changed??
1331 *
1332 * Answer: Yes, drivers may have changed it in their
1333 * f_op->mmap method.
1334 *
1335 * Ensures that vmas marked as uncached stay that way.
1336 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001337 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001338 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1339 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1340 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001341
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001342 vma_link(mm, vma, prev, rb_link, rb_parent);
1343 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001344
1345 /* Once vma denies write, undo our temporary denial count */
1346 if (correct_wcount)
1347 atomic_inc(&inode->i_writecount);
1348out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001349 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001352 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001354 if (!mlock_vma_pages_range(vma, addr, addr + len))
1355 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001356 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001357 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 return addr;
1359
1360unmap_and_free_vma:
1361 if (correct_wcount)
1362 atomic_inc(&inode->i_writecount);
1363 vma->vm_file = NULL;
1364 fput(file);
1365
1366 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001367 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1368 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369free_vma:
1370 kmem_cache_free(vm_area_cachep, vma);
1371unacct_error:
1372 if (charged)
1373 vm_unacct_memory(charged);
1374 return error;
1375}
1376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377/* Get an address range which is currently unmapped.
1378 * For shmat() with addr=0.
1379 *
1380 * Ugly calling convention alert:
1381 * Return value with the low bits set means error value,
1382 * ie
1383 * if (ret & ~PAGE_MASK)
1384 * error = ret;
1385 *
1386 * This function "knows" that -ENOMEM has the bits set.
1387 */
1388#ifndef HAVE_ARCH_UNMAPPED_AREA
1389unsigned long
1390arch_get_unmapped_area(struct file *filp, unsigned long addr,
1391 unsigned long len, unsigned long pgoff, unsigned long flags)
1392{
1393 struct mm_struct *mm = current->mm;
1394 struct vm_area_struct *vma;
1395 unsigned long start_addr;
1396
1397 if (len > TASK_SIZE)
1398 return -ENOMEM;
1399
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001400 if (flags & MAP_FIXED)
1401 return addr;
1402
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 if (addr) {
1404 addr = PAGE_ALIGN(addr);
1405 vma = find_vma(mm, addr);
1406 if (TASK_SIZE - len >= addr &&
1407 (!vma || addr + len <= vma->vm_start))
1408 return addr;
1409 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001410 if (len > mm->cached_hole_size) {
1411 start_addr = addr = mm->free_area_cache;
1412 } else {
1413 start_addr = addr = TASK_UNMAPPED_BASE;
1414 mm->cached_hole_size = 0;
1415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
1417full_search:
1418 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1419 /* At this point: (!vma || addr < vma->vm_end). */
1420 if (TASK_SIZE - len < addr) {
1421 /*
1422 * Start a new search - just in case we missed
1423 * some holes.
1424 */
1425 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001426 addr = TASK_UNMAPPED_BASE;
1427 start_addr = addr;
1428 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 goto full_search;
1430 }
1431 return -ENOMEM;
1432 }
1433 if (!vma || addr + len <= vma->vm_start) {
1434 /*
1435 * Remember the place where we stopped the search:
1436 */
1437 mm->free_area_cache = addr + len;
1438 return addr;
1439 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001440 if (addr + mm->cached_hole_size < vma->vm_start)
1441 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 addr = vma->vm_end;
1443 }
1444}
1445#endif
1446
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001447void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448{
1449 /*
1450 * Is this a new hole at the lowest possible address?
1451 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001452 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1453 mm->free_area_cache = addr;
1454 mm->cached_hole_size = ~0UL;
1455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
1458/*
1459 * This mmap-allocator allocates new areas top-down from below the
1460 * stack's low limit (the base):
1461 */
1462#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1463unsigned long
1464arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1465 const unsigned long len, const unsigned long pgoff,
1466 const unsigned long flags)
1467{
1468 struct vm_area_struct *vma;
1469 struct mm_struct *mm = current->mm;
1470 unsigned long addr = addr0;
1471
1472 /* requested length too big for entire address space */
1473 if (len > TASK_SIZE)
1474 return -ENOMEM;
1475
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001476 if (flags & MAP_FIXED)
1477 return addr;
1478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 /* requesting a specific address */
1480 if (addr) {
1481 addr = PAGE_ALIGN(addr);
1482 vma = find_vma(mm, addr);
1483 if (TASK_SIZE - len >= addr &&
1484 (!vma || addr + len <= vma->vm_start))
1485 return addr;
1486 }
1487
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001488 /* check if free_area_cache is useful for us */
1489 if (len <= mm->cached_hole_size) {
1490 mm->cached_hole_size = 0;
1491 mm->free_area_cache = mm->mmap_base;
1492 }
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 /* either no address requested or can't fit in requested address hole */
1495 addr = mm->free_area_cache;
1496
1497 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001498 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 vma = find_vma(mm, addr-len);
1500 if (!vma || addr <= vma->vm_start)
1501 /* remember the address as a hint for next time */
1502 return (mm->free_area_cache = addr-len);
1503 }
1504
Chris Wright73219d12005-06-21 17:14:52 -07001505 if (mm->mmap_base < len)
1506 goto bottomup;
1507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 addr = mm->mmap_base-len;
1509
1510 do {
1511 /*
1512 * Lookup failure means no vma is above this address,
1513 * else if new region fits below vma->vm_start,
1514 * return with success:
1515 */
1516 vma = find_vma(mm, addr);
1517 if (!vma || addr+len <= vma->vm_start)
1518 /* remember the address as a hint for next time */
1519 return (mm->free_area_cache = addr);
1520
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001521 /* remember the largest hole we saw so far */
1522 if (addr + mm->cached_hole_size < vma->vm_start)
1523 mm->cached_hole_size = vma->vm_start - addr;
1524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 /* try just below the current vma->vm_start */
1526 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001527 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
Chris Wright73219d12005-06-21 17:14:52 -07001529bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 /*
1531 * A failed mmap() very likely causes application failure,
1532 * so fall back to the bottom-up function here. This scenario
1533 * can happen with large stack limits and large mmap()
1534 * allocations.
1535 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001536 mm->cached_hole_size = ~0UL;
1537 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1539 /*
1540 * Restore the topdown base:
1541 */
1542 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001543 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545 return addr;
1546}
1547#endif
1548
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001549void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
1551 /*
1552 * Is this a new hole at the highest possible address?
1553 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001554 if (addr > mm->free_area_cache)
1555 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
1557 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001558 if (mm->free_area_cache > mm->mmap_base)
1559 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560}
1561
1562unsigned long
1563get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1564 unsigned long pgoff, unsigned long flags)
1565{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001566 unsigned long (*get_area)(struct file *, unsigned long,
1567 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Al Viro9206de92009-12-03 15:23:11 -05001569 unsigned long error = arch_mmap_check(addr, len, flags);
1570 if (error)
1571 return error;
1572
1573 /* Careful about overflows.. */
1574 if (len > TASK_SIZE)
1575 return -ENOMEM;
1576
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001577 get_area = current->mm->get_unmapped_area;
1578 if (file && file->f_op && file->f_op->get_unmapped_area)
1579 get_area = file->f_op->get_unmapped_area;
1580 addr = get_area(file, addr, len, pgoff, flags);
1581 if (IS_ERR_VALUE(addr))
1582 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001584 if (addr > TASK_SIZE - len)
1585 return -ENOMEM;
1586 if (addr & ~PAGE_MASK)
1587 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001588
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001589 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590}
1591
1592EXPORT_SYMBOL(get_unmapped_area);
1593
1594/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001595struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
1597 struct vm_area_struct *vma = NULL;
1598
1599 if (mm) {
1600 /* Check the cache first. */
1601 /* (Cache hit rate is typically around 35%.) */
1602 vma = mm->mmap_cache;
1603 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1604 struct rb_node * rb_node;
1605
1606 rb_node = mm->mm_rb.rb_node;
1607 vma = NULL;
1608
1609 while (rb_node) {
1610 struct vm_area_struct * vma_tmp;
1611
1612 vma_tmp = rb_entry(rb_node,
1613 struct vm_area_struct, vm_rb);
1614
1615 if (vma_tmp->vm_end > addr) {
1616 vma = vma_tmp;
1617 if (vma_tmp->vm_start <= addr)
1618 break;
1619 rb_node = rb_node->rb_left;
1620 } else
1621 rb_node = rb_node->rb_right;
1622 }
1623 if (vma)
1624 mm->mmap_cache = vma;
1625 }
1626 }
1627 return vma;
1628}
1629
1630EXPORT_SYMBOL(find_vma);
1631
1632/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1633struct vm_area_struct *
1634find_vma_prev(struct mm_struct *mm, unsigned long addr,
1635 struct vm_area_struct **pprev)
1636{
1637 struct vm_area_struct *vma = NULL, *prev = NULL;
ZhenwenXu48aae422009-01-06 14:40:21 -08001638 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 if (!mm)
1640 goto out;
1641
1642 /* Guard against addr being lower than the first VMA */
1643 vma = mm->mmap;
1644
1645 /* Go through the RB tree quickly. */
1646 rb_node = mm->mm_rb.rb_node;
1647
1648 while (rb_node) {
1649 struct vm_area_struct *vma_tmp;
1650 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1651
1652 if (addr < vma_tmp->vm_end) {
1653 rb_node = rb_node->rb_left;
1654 } else {
1655 prev = vma_tmp;
1656 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1657 break;
1658 rb_node = rb_node->rb_right;
1659 }
1660 }
1661
1662out:
1663 *pprev = prev;
1664 return prev ? prev->vm_next : vma;
1665}
1666
1667/*
1668 * Verify that the stack growth is acceptable and
1669 * update accounting. This is shared with both the
1670 * grow-up and grow-down cases.
1671 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001672static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
1674 struct mm_struct *mm = vma->vm_mm;
1675 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001676 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
1678 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001679 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 return -ENOMEM;
1681
1682 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001683 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 return -ENOMEM;
1685
1686 /* mlock limit tests */
1687 if (vma->vm_flags & VM_LOCKED) {
1688 unsigned long locked;
1689 unsigned long limit;
1690 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001691 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1692 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 if (locked > limit && !capable(CAP_IPC_LOCK))
1694 return -ENOMEM;
1695 }
1696
Adam Litke0d59a012007-01-30 14:35:39 -08001697 /* Check to ensure the stack will not grow into a hugetlb-only region */
1698 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1699 vma->vm_end - size;
1700 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1701 return -EFAULT;
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 /*
1704 * Overcommit.. This must be the final test, as it will
1705 * update security statistics.
1706 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001707 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 return -ENOMEM;
1709
1710 /* Ok, everything looks good - let it rip */
1711 mm->total_vm += grow;
1712 if (vma->vm_flags & VM_LOCKED)
1713 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001714 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 return 0;
1716}
1717
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001718#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001720 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1721 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001723int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724{
1725 int error;
1726
1727 if (!(vma->vm_flags & VM_GROWSUP))
1728 return -EFAULT;
1729
1730 /*
1731 * We must make sure the anon_vma is allocated
1732 * so that the anon_vma locking is not a noop.
1733 */
1734 if (unlikely(anon_vma_prepare(vma)))
1735 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001736 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
1738 /*
1739 * vma->vm_start/vm_end cannot change under us because the caller
1740 * is required to hold the mmap_sem in read mode. We need the
1741 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001742 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 */
Helge Deller06b32f32006-12-19 19:28:33 +01001744 if (address < PAGE_ALIGN(address+4))
1745 address = PAGE_ALIGN(address+4);
1746 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001747 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001748 return -ENOMEM;
1749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 error = 0;
1751
1752 /* Somebody else might have raced and expanded it already */
1753 if (address > vma->vm_end) {
1754 unsigned long size, grow;
1755
1756 size = address - vma->vm_start;
1757 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1758
Hugh Dickins42c36f62011-05-09 17:44:42 -07001759 error = -ENOMEM;
1760 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1761 error = acct_stack_growth(vma, size, grow);
1762 if (!error) {
1763 vma->vm_end = address;
1764 perf_event_mmap(vma);
1765 }
Eric B Munson3af9e852010-05-18 15:30:49 +01001766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001768 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001769 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 return error;
1771}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001772#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774/*
1775 * vma is the first one with address < vma->vm_start. Have to extend vma.
1776 */
Michal Hockod05f3162011-05-24 17:11:44 -07001777int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001778 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779{
1780 int error;
1781
1782 /*
1783 * We must make sure the anon_vma is allocated
1784 * so that the anon_vma locking is not a noop.
1785 */
1786 if (unlikely(anon_vma_prepare(vma)))
1787 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001788
1789 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001790 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001791 if (error)
1792 return error;
1793
Rik van Rielbb4a3402010-08-09 17:18:37 -07001794 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
1796 /*
1797 * vma->vm_start/vm_end cannot change under us because the caller
1798 * is required to hold the mmap_sem in read mode. We need the
1799 * anon_vma lock to serialize against concurrent expand_stacks.
1800 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
1802 /* Somebody else might have raced and expanded it already */
1803 if (address < vma->vm_start) {
1804 unsigned long size, grow;
1805
1806 size = vma->vm_end - address;
1807 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1808
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001809 error = -ENOMEM;
1810 if (grow <= vma->vm_pgoff) {
1811 error = acct_stack_growth(vma, size, grow);
1812 if (!error) {
1813 vma->vm_start = address;
1814 vma->vm_pgoff -= grow;
1815 perf_event_mmap(vma);
1816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 }
1818 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001819 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001820 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 return error;
1822}
1823
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001824#ifdef CONFIG_STACK_GROWSUP
1825int expand_stack(struct vm_area_struct *vma, unsigned long address)
1826{
1827 return expand_upwards(vma, address);
1828}
1829
1830struct vm_area_struct *
1831find_extend_vma(struct mm_struct *mm, unsigned long addr)
1832{
1833 struct vm_area_struct *vma, *prev;
1834
1835 addr &= PAGE_MASK;
1836 vma = find_vma_prev(mm, addr, &prev);
1837 if (vma && (vma->vm_start <= addr))
1838 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001839 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001840 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001841 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001842 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001843 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001844 return prev;
1845}
1846#else
1847int expand_stack(struct vm_area_struct *vma, unsigned long address)
1848{
1849 return expand_downwards(vma, address);
1850}
1851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852struct vm_area_struct *
1853find_extend_vma(struct mm_struct * mm, unsigned long addr)
1854{
1855 struct vm_area_struct * vma;
1856 unsigned long start;
1857
1858 addr &= PAGE_MASK;
1859 vma = find_vma(mm,addr);
1860 if (!vma)
1861 return NULL;
1862 if (vma->vm_start <= addr)
1863 return vma;
1864 if (!(vma->vm_flags & VM_GROWSDOWN))
1865 return NULL;
1866 start = vma->vm_start;
1867 if (expand_stack(vma, addr))
1868 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001869 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001870 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return vma;
1873}
1874#endif
1875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001877 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001879 *
1880 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001882static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001884 /* Update high watermark before we lower total_vm */
1885 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001887 long nrpages = vma_pages(vma);
1888
1889 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001890 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001891 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001892 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 validate_mm(mm);
1894}
1895
1896/*
1897 * Get rid of page table information in the indicated region.
1898 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001899 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 */
1901static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001902 struct vm_area_struct *vma, struct vm_area_struct *prev,
1903 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001905 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001906 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 unsigned long nr_accounted = 0;
1908
1909 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001910 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001911 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001912 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 vm_unacct_memory(nr_accounted);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001914 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1915 next ? next->vm_start : 0);
1916 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917}
1918
1919/*
1920 * Create a list of vma's touched by the unmap, removing them from the mm's
1921 * vma list as we go..
1922 */
1923static void
1924detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1925 struct vm_area_struct *prev, unsigned long end)
1926{
1927 struct vm_area_struct **insertion_point;
1928 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001929 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
1931 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001932 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 do {
1934 rb_erase(&vma->vm_rb, &mm->mm_rb);
1935 mm->map_count--;
1936 tail_vma = vma;
1937 vma = vma->vm_next;
1938 } while (vma && vma->vm_start < end);
1939 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001940 if (vma)
1941 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001943 if (mm->unmap_area == arch_unmap_area)
1944 addr = prev ? prev->vm_end : mm->mmap_base;
1945 else
1946 addr = vma ? vma->vm_start : mm->mmap_base;
1947 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 mm->mmap_cache = NULL; /* Kill the cache. */
1949}
1950
1951/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001952 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1953 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001955static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 unsigned long addr, int new_below)
1957{
1958 struct mempolicy *pol;
1959 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001960 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
Andi Kleena5516432008-07-23 21:27:41 -07001962 if (is_vm_hugetlb_page(vma) && (addr &
1963 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 return -EINVAL;
1965
Christoph Lametere94b1762006-12-06 20:33:17 -08001966 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001968 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
1970 /* most fields are the same, copy all, and then fixup */
1971 *new = *vma;
1972
Rik van Riel5beb4932010-03-05 13:42:07 -08001973 INIT_LIST_HEAD(&new->anon_vma_chain);
1974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 if (new_below)
1976 new->vm_end = addr;
1977 else {
1978 new->vm_start = addr;
1979 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1980 }
1981
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001982 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001984 err = PTR_ERR(pol);
1985 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 }
1987 vma_set_policy(new, pol);
1988
Rik van Riel5beb4932010-03-05 13:42:07 -08001989 if (anon_vma_clone(new, vma))
1990 goto out_free_mpol;
1991
Matt Helsley925d1c42008-04-29 01:01:36 -07001992 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001994 if (vma->vm_flags & VM_EXECUTABLE)
1995 added_exe_file_vma(mm);
1996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
1998 if (new->vm_ops && new->vm_ops->open)
1999 new->vm_ops->open(new);
2000
2001 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002002 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2004 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002005 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Rik van Riel5beb4932010-03-05 13:42:07 -08002007 /* Success. */
2008 if (!err)
2009 return 0;
2010
2011 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002012 if (new->vm_ops && new->vm_ops->close)
2013 new->vm_ops->close(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002014 if (new->vm_file) {
2015 if (vma->vm_flags & VM_EXECUTABLE)
2016 removed_exe_file_vma(mm);
2017 fput(new->vm_file);
2018 }
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002019 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002020 out_free_mpol:
2021 mpol_put(pol);
2022 out_free_vma:
2023 kmem_cache_free(vm_area_cachep, new);
2024 out_err:
2025 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026}
2027
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002028/*
2029 * Split a vma into two pieces at address 'addr', a new vma is allocated
2030 * either for the first part or the tail.
2031 */
2032int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2033 unsigned long addr, int new_below)
2034{
2035 if (mm->map_count >= sysctl_max_map_count)
2036 return -ENOMEM;
2037
2038 return __split_vma(mm, vma, addr, new_below);
2039}
2040
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041/* Munmap is split into 2 main parts -- this part which finds
2042 * what needs doing, and the areas themselves, which do the
2043 * work. This now handles partial unmappings.
2044 * Jeremy Fitzhardinge <jeremy@goop.org>
2045 */
2046int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2047{
2048 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002049 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
2051 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2052 return -EINVAL;
2053
2054 if ((len = PAGE_ALIGN(len)) == 0)
2055 return -EINVAL;
2056
2057 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07002058 vma = find_vma_prev(mm, start, &prev);
2059 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07002061 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
2063 /* if it doesn't overlap, we have nothing.. */
2064 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002065 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 return 0;
2067
2068 /*
2069 * If we need to split any vma, do it now to save pain later.
2070 *
2071 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2072 * unmapped vm_area_struct will remain in use: so lower split_vma
2073 * places tmp vma above, and higher split_vma places tmp vma below.
2074 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002075 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002076 int error;
2077
2078 /*
2079 * Make sure that map_count on return from munmap() will
2080 * not exceed its limit; but let map_count go just above
2081 * its limit temporarily, to help free resources as expected.
2082 */
2083 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2084 return -ENOMEM;
2085
2086 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 if (error)
2088 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002089 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 }
2091
2092 /* Does it split the last one? */
2093 last = find_vma(mm, end);
2094 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002095 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 if (error)
2097 return error;
2098 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002099 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
2101 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002102 * unlock any mlock()ed ranges before detaching vmas
2103 */
2104 if (mm->locked_vm) {
2105 struct vm_area_struct *tmp = vma;
2106 while (tmp && tmp->vm_start < end) {
2107 if (tmp->vm_flags & VM_LOCKED) {
2108 mm->locked_vm -= vma_pages(tmp);
2109 munlock_vma_pages_all(tmp);
2110 }
2111 tmp = tmp->vm_next;
2112 }
2113 }
2114
2115 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 * Remove the vma's, and unmap the actual pages
2117 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002118 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2119 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120
2121 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002122 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
2124 return 0;
2125}
2126
2127EXPORT_SYMBOL(do_munmap);
2128
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002129SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130{
2131 int ret;
2132 struct mm_struct *mm = current->mm;
2133
2134 profile_munmap(addr);
2135
2136 down_write(&mm->mmap_sem);
2137 ret = do_munmap(mm, addr, len);
2138 up_write(&mm->mmap_sem);
2139 return ret;
2140}
2141
2142static inline void verify_mm_writelocked(struct mm_struct *mm)
2143{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002144#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2146 WARN_ON(1);
2147 up_read(&mm->mmap_sem);
2148 }
2149#endif
2150}
2151
2152/*
2153 * this is really a simplified "do_mmap". it only handles
2154 * anonymous maps. eventually we may be able to do some
2155 * brk-specific accounting here.
2156 */
2157unsigned long do_brk(unsigned long addr, unsigned long len)
2158{
2159 struct mm_struct * mm = current->mm;
2160 struct vm_area_struct * vma, * prev;
2161 unsigned long flags;
2162 struct rb_node ** rb_link, * rb_parent;
2163 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002164 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
2166 len = PAGE_ALIGN(len);
2167 if (!len)
2168 return addr;
2169
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002170 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002171 if (error)
2172 return error;
2173
Kirill Korotaev3a459752006-09-07 14:17:04 +04002174 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2175
Al Viro2c6a10162009-12-03 19:40:46 -05002176 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2177 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002178 return error;
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 /*
2181 * mlock MCL_FUTURE?
2182 */
2183 if (mm->def_flags & VM_LOCKED) {
2184 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002185 locked = len >> PAGE_SHIFT;
2186 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002187 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002188 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2190 return -EAGAIN;
2191 }
2192
2193 /*
2194 * mm->mmap_sem is required to protect against another thread
2195 * changing the mappings in case we sleep.
2196 */
2197 verify_mm_writelocked(mm);
2198
2199 /*
2200 * Clear old maps. this also does some error checking for us
2201 */
2202 munmap_back:
2203 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2204 if (vma && vma->vm_start < addr + len) {
2205 if (do_munmap(mm, addr, len))
2206 return -ENOMEM;
2207 goto munmap_back;
2208 }
2209
2210 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002211 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 return -ENOMEM;
2213
2214 if (mm->map_count > sysctl_max_map_count)
2215 return -ENOMEM;
2216
2217 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2218 return -ENOMEM;
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002221 vma = vma_merge(mm, prev, addr, addr + len, flags,
2222 NULL, NULL, pgoff, NULL);
2223 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 goto out;
2225
2226 /*
2227 * create a vma struct for an anonymous mapping
2228 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002229 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 if (!vma) {
2231 vm_unacct_memory(len >> PAGE_SHIFT);
2232 return -ENOMEM;
2233 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234
Rik van Riel5beb4932010-03-05 13:42:07 -08002235 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 vma->vm_mm = mm;
2237 vma->vm_start = addr;
2238 vma->vm_end = addr + len;
2239 vma->vm_pgoff = pgoff;
2240 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002241 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 vma_link(mm, vma, prev, rb_link, rb_parent);
2243out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002244 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 mm->total_vm += len >> PAGE_SHIFT;
2246 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002247 if (!mlock_vma_pages_range(vma, addr, addr + len))
2248 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 }
2250 return addr;
2251}
2252
2253EXPORT_SYMBOL(do_brk);
2254
2255/* Release all mmaps. */
2256void exit_mmap(struct mm_struct *mm)
2257{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002258 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002259 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002261 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002263 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002264 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002265
Rik van Rielba470de2008-10-18 20:26:50 -07002266 if (mm->locked_vm) {
2267 vma = mm->mmap;
2268 while (vma) {
2269 if (vma->vm_flags & VM_LOCKED)
2270 munlock_vma_pages_all(vma);
2271 vma = vma->vm_next;
2272 }
2273 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002274
2275 arch_exit_mmap(mm);
2276
Rik van Rielba470de2008-10-18 20:26:50 -07002277 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002278 if (!vma) /* Can happen if dup_mmap() received an OOM */
2279 return;
2280
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 flush_cache_mm(mm);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002283 tlb_gather_mmu(&tlb, mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002284 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002285 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002286 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 vm_unacct_memory(nr_accounted);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002288
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002289 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2290 tlb_finish_mmu(&tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002293 * Walk the list again, actually closing and freeing it,
2294 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002296 while (vma)
2297 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002298
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002299 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300}
2301
2302/* Insert vm structure into process list sorted by address
2303 * and into the inode's i_mmap tree. If vm_file is non-NULL
2304 * then i_mmap_lock is taken here.
2305 */
2306int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2307{
2308 struct vm_area_struct * __vma, * prev;
2309 struct rb_node ** rb_link, * rb_parent;
2310
2311 /*
2312 * The vm_pgoff of a purely anonymous vma should be irrelevant
2313 * until its first write fault, when page's anon_vma and index
2314 * are set. But now set the vm_pgoff it will almost certainly
2315 * end up with (unless mremap moves it elsewhere before that
2316 * first wfault), so /proc/pid/maps tells a consistent story.
2317 *
2318 * By setting it to reflect the virtual start address of the
2319 * vma, merges and splits can happen in a seamless way, just
2320 * using the existing file pgoff checks and manipulations.
2321 * Similarly in do_mmap_pgoff and in do_brk.
2322 */
2323 if (!vma->vm_file) {
2324 BUG_ON(vma->anon_vma);
2325 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2326 }
2327 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2328 if (__vma && __vma->vm_start < vma->vm_end)
2329 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002330 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002331 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002332 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 vma_link(mm, vma, prev, rb_link, rb_parent);
2334 return 0;
2335}
2336
2337/*
2338 * Copy the vma structure to a new location in the same mm,
2339 * prior to moving page table entries, to effect an mremap move.
2340 */
2341struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2342 unsigned long addr, unsigned long len, pgoff_t pgoff)
2343{
2344 struct vm_area_struct *vma = *vmap;
2345 unsigned long vma_start = vma->vm_start;
2346 struct mm_struct *mm = vma->vm_mm;
2347 struct vm_area_struct *new_vma, *prev;
2348 struct rb_node **rb_link, *rb_parent;
2349 struct mempolicy *pol;
2350
2351 /*
2352 * If anonymous vma has not yet been faulted, update new pgoff
2353 * to match new location, to increase its chance of merging.
2354 */
2355 if (!vma->vm_file && !vma->anon_vma)
2356 pgoff = addr >> PAGE_SHIFT;
2357
2358 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2359 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2360 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2361 if (new_vma) {
2362 /*
2363 * Source vma may have been merged into new_vma
2364 */
2365 if (vma_start >= new_vma->vm_start &&
2366 vma_start < new_vma->vm_end)
2367 *vmap = new_vma;
2368 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002369 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 if (new_vma) {
2371 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002372 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002373 if (IS_ERR(pol))
2374 goto out_free_vma;
2375 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2376 if (anon_vma_clone(new_vma, vma))
2377 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 vma_set_policy(new_vma, pol);
2379 new_vma->vm_start = addr;
2380 new_vma->vm_end = addr + len;
2381 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002382 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002384 if (vma->vm_flags & VM_EXECUTABLE)
2385 added_exe_file_vma(mm);
2386 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 if (new_vma->vm_ops && new_vma->vm_ops->open)
2388 new_vma->vm_ops->open(new_vma);
2389 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2390 }
2391 }
2392 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002393
2394 out_free_mempol:
2395 mpol_put(pol);
2396 out_free_vma:
2397 kmem_cache_free(vm_area_cachep, new_vma);
2398 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002400
2401/*
2402 * Return true if the calling process may expand its vm space by the passed
2403 * number of pages
2404 */
2405int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2406{
2407 unsigned long cur = mm->total_vm; /* pages */
2408 unsigned long lim;
2409
Jiri Slaby59e99e52010-03-05 13:41:44 -08002410 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002411
2412 if (cur + npages > lim)
2413 return 0;
2414 return 1;
2415}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002416
2417
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002418static int special_mapping_fault(struct vm_area_struct *vma,
2419 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002420{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002421 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002422 struct page **pages;
2423
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002424 /*
2425 * special mappings have no vm_file, and in that case, the mm
2426 * uses vm_pgoff internally. So we have to subtract it from here.
2427 * We are allowed to do this because we are the mm; do not copy
2428 * this code into drivers!
2429 */
2430 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002431
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002432 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2433 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002434
2435 if (*pages) {
2436 struct page *page = *pages;
2437 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002438 vmf->page = page;
2439 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002440 }
2441
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002442 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002443}
2444
2445/*
2446 * Having a close hook prevents vma merging regardless of flags.
2447 */
2448static void special_mapping_close(struct vm_area_struct *vma)
2449{
2450}
2451
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002452static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002453 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002454 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002455};
2456
2457/*
2458 * Called with mm->mmap_sem held for writing.
2459 * Insert a new vma covering the given region, with the given flags.
2460 * Its pages are supplied by the given array of struct page *.
2461 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2462 * The region past the last page supplied will always produce SIGBUS.
2463 * The array pointer and the pages it points to are assumed to stay alive
2464 * for as long as this mapping might exist.
2465 */
2466int install_special_mapping(struct mm_struct *mm,
2467 unsigned long addr, unsigned long len,
2468 unsigned long vm_flags, struct page **pages)
2469{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002470 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002471 struct vm_area_struct *vma;
2472
2473 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2474 if (unlikely(vma == NULL))
2475 return -ENOMEM;
2476
Rik van Riel5beb4932010-03-05 13:42:07 -08002477 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002478 vma->vm_mm = mm;
2479 vma->vm_start = addr;
2480 vma->vm_end = addr + len;
2481
Nick Piggin2f987352008-02-02 03:08:53 +01002482 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002483 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002484
2485 vma->vm_ops = &special_mapping_vmops;
2486 vma->vm_private_data = pages;
2487
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002488 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2489 if (ret)
2490 goto out;
2491
2492 ret = insert_vm_struct(mm, vma);
2493 if (ret)
2494 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002495
2496 mm->total_vm += len >> PAGE_SHIFT;
2497
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002498 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002499
Roland McGrathfa5dc222007-02-08 14:20:41 -08002500 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002501
2502out:
2503 kmem_cache_free(vm_area_cachep, vma);
2504 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002505}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002506
2507static DEFINE_MUTEX(mm_all_locks_mutex);
2508
Peter Zijlstra454ed842008-08-11 09:30:25 +02002509static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002510{
Rik van Riel012f1802010-08-09 17:18:40 -07002511 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002512 /*
2513 * The LSB of head.next can't change from under us
2514 * because we hold the mm_all_locks_mutex.
2515 */
Rik van Riel012f1802010-08-09 17:18:40 -07002516 spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002517 /*
2518 * We can safely modify head.next after taking the
Rik van Riel012f1802010-08-09 17:18:40 -07002519 * anon_vma->root->lock. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002520 * the same anon_vma we won't take it again.
2521 *
2522 * No need of atomic instructions here, head.next
2523 * can't change from under us thanks to the
Rik van Riel012f1802010-08-09 17:18:40 -07002524 * anon_vma->root->lock.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002525 */
2526 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002527 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002528 BUG();
2529 }
2530}
2531
Peter Zijlstra454ed842008-08-11 09:30:25 +02002532static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002533{
2534 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2535 /*
2536 * AS_MM_ALL_LOCKS can't change from under us because
2537 * we hold the mm_all_locks_mutex.
2538 *
2539 * Operations on ->flags have to be atomic because
2540 * even if AS_MM_ALL_LOCKS is stable thanks to the
2541 * mm_all_locks_mutex, there may be other cpus
2542 * changing other bitflags in parallel to us.
2543 */
2544 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2545 BUG();
Peter Zijlstra454ed842008-08-11 09:30:25 +02002546 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002547 }
2548}
2549
2550/*
2551 * This operation locks against the VM for all pte/vma/mm related
2552 * operations that could ever happen on a certain mm. This includes
2553 * vmtruncate, try_to_unmap, and all page faults.
2554 *
2555 * The caller must take the mmap_sem in write mode before calling
2556 * mm_take_all_locks(). The caller isn't allowed to release the
2557 * mmap_sem until mm_drop_all_locks() returns.
2558 *
2559 * mmap_sem in write mode is required in order to block all operations
2560 * that could modify pagetables and free pages without need of
2561 * altering the vma layout (for example populate_range() with
2562 * nonlinear vmas). It's also needed in write mode to avoid new
2563 * anon_vmas to be associated with existing vmas.
2564 *
2565 * A single task can't take more than one mm_take_all_locks() in a row
2566 * or it would deadlock.
2567 *
2568 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2569 * mapping->flags avoid to take the same lock twice, if more than one
2570 * vma in this mm is backed by the same anon_vma or address_space.
2571 *
2572 * We can take all the locks in random order because the VM code
2573 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2574 * takes more than one of them in a row. Secondly we're protected
2575 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2576 *
2577 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2578 * that may have to take thousand of locks.
2579 *
2580 * mm_take_all_locks() can fail if it's interrupted by signals.
2581 */
2582int mm_take_all_locks(struct mm_struct *mm)
2583{
2584 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002585 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002586 int ret = -EINTR;
2587
2588 BUG_ON(down_read_trylock(&mm->mmap_sem));
2589
2590 mutex_lock(&mm_all_locks_mutex);
2591
2592 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2593 if (signal_pending(current))
2594 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002595 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002596 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002597 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002598
2599 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2600 if (signal_pending(current))
2601 goto out_unlock;
2602 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002603 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2604 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002605 }
2606
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002607 ret = 0;
2608
2609out_unlock:
2610 if (ret)
2611 mm_drop_all_locks(mm);
2612
2613 return ret;
2614}
2615
2616static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2617{
Rik van Riel012f1802010-08-09 17:18:40 -07002618 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002619 /*
2620 * The LSB of head.next can't change to 0 from under
2621 * us because we hold the mm_all_locks_mutex.
2622 *
2623 * We must however clear the bitflag before unlocking
2624 * the vma so the users using the anon_vma->head will
2625 * never see our bitflag.
2626 *
2627 * No need of atomic instructions here, head.next
2628 * can't change from under us until we release the
Rik van Riel012f1802010-08-09 17:18:40 -07002629 * anon_vma->root->lock.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002630 */
2631 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002632 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002633 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002634 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002635 }
2636}
2637
2638static void vm_unlock_mapping(struct address_space *mapping)
2639{
2640 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2641 /*
2642 * AS_MM_ALL_LOCKS can't change to 0 from under us
2643 * because we hold the mm_all_locks_mutex.
2644 */
2645 spin_unlock(&mapping->i_mmap_lock);
2646 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2647 &mapping->flags))
2648 BUG();
2649 }
2650}
2651
2652/*
2653 * The mmap_sem cannot be released by the caller until
2654 * mm_drop_all_locks() returns.
2655 */
2656void mm_drop_all_locks(struct mm_struct *mm)
2657{
2658 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002659 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002660
2661 BUG_ON(down_read_trylock(&mm->mmap_sem));
2662 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2663
2664 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2665 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002666 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2667 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002668 if (vma->vm_file && vma->vm_file->f_mapping)
2669 vm_unlock_mapping(vma->vm_file->f_mapping);
2670 }
2671
2672 mutex_unlock(&mm_all_locks_mutex);
2673}
David Howells8feae132009-01-08 12:04:47 +00002674
2675/*
2676 * initialise the VMA slab
2677 */
2678void __init mmap_init(void)
2679{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002680 int ret;
2681
2682 ret = percpu_counter_init(&vm_committed_as, 0);
2683 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002684}