blob: 3c0061f744f50cad8fd79eab118d7a54561fa654 [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) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700125 free = global_page_state(NR_FREE_PAGES);
126 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Dmitry Finkc15bef32011-07-25 17:12:19 -0700128 /*
129 * shmem pages shouldn't be counted as free in this
130 * case, they can't be purged, only swapped out, and
131 * that won't affect the overall amount of available
132 * memory in the system.
133 */
134 free -= global_page_state(NR_SHMEM);
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 free += nr_swap_pages;
137
138 /*
139 * Any slabs which are created with the
140 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
141 * which are reclaimable, under pressure. The dentry
142 * cache and most inode caches should fall into this
143 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700144 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700147 * Leave reserved pages. The pages are not for anonymous pages.
148 */
149 if (free <= totalreserve_pages)
150 goto error;
151 else
152 free -= totalreserve_pages;
153
154 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 * Leave the last 3% for root
156 */
157 if (!cap_sys_admin)
158 free -= free / 32;
159
160 if (free > pages)
161 return 0;
162
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700163 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 }
165
166 allowed = (totalram_pages - hugetlb_total_pages())
167 * sysctl_overcommit_ratio / 100;
168 /*
169 * Leave the last 3% for root
170 */
171 if (!cap_sys_admin)
172 allowed -= allowed / 32;
173 allowed += total_swap_pages;
174
175 /* Don't let a single process grow too big:
176 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700177 if (mm)
178 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700180 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700182error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 vm_unacct_memory(pages);
184
185 return -ENOMEM;
186}
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700189 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 */
191static void __remove_shared_vm_struct(struct vm_area_struct *vma,
192 struct file *file, struct address_space *mapping)
193{
194 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800195 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 if (vma->vm_flags & VM_SHARED)
197 mapping->i_mmap_writable--;
198
199 flush_dcache_mmap_lock(mapping);
200 if (unlikely(vma->vm_flags & VM_NONLINEAR))
201 list_del_init(&vma->shared.vm_set.list);
202 else
203 vma_prio_tree_remove(vma, &mapping->i_mmap);
204 flush_dcache_mmap_unlock(mapping);
205}
206
207/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700208 * Unlink a file-based vm structure from its prio_tree, to hide
209 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700211void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
213 struct file *file = vma->vm_file;
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (file) {
216 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700217 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700219 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700221}
222
223/*
224 * Close a vm structure and free it, returning the next.
225 */
226static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
227{
228 struct vm_area_struct *next = vma->vm_next;
229
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700230 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 if (vma->vm_ops && vma->vm_ops->close)
232 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700233 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700234 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700235 if (vma->vm_flags & VM_EXECUTABLE)
236 removed_exe_file_vma(vma->vm_mm);
237 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700238 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700240 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100243SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 unsigned long rlim, retval;
246 unsigned long newbrk, oldbrk;
247 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700248 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 down_write(&mm->mmap_sem);
251
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700252#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800253 /*
254 * CONFIG_COMPAT_BRK can still be overridden by setting
255 * randomize_va_space to 2, which will still cause mm->start_brk
256 * to be arbitrarily shifted
257 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700258 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800259 min_brk = mm->start_brk;
260 else
261 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700262#else
263 min_brk = mm->start_brk;
264#endif
265 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700267
268 /*
269 * Check against rlimit here. If this check is done later after the test
270 * of oldbrk with newbrk then it can escape the test and let the data
271 * segment grow beyond its set limit the in case where the limit is
272 * not page aligned -Ram Gupta
273 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800274 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100275 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
276 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700277 goto out;
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 newbrk = PAGE_ALIGN(brk);
280 oldbrk = PAGE_ALIGN(mm->brk);
281 if (oldbrk == newbrk)
282 goto set_brk;
283
284 /* Always allow shrinking brk. */
285 if (brk <= mm->brk) {
286 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
287 goto set_brk;
288 goto out;
289 }
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 /* Check against existing mmap mappings. */
292 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
293 goto out;
294
295 /* Ok, looks good - let it rip. */
296 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
297 goto out;
298set_brk:
299 mm->brk = brk;
300out:
301 retval = mm->brk;
302 up_write(&mm->mmap_sem);
303 return retval;
304}
305
306#ifdef DEBUG_MM_RB
307static int browse_rb(struct rb_root *root)
308{
309 int i = 0, j;
310 struct rb_node *nd, *pn = NULL;
311 unsigned long prev = 0, pend = 0;
312
313 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
314 struct vm_area_struct *vma;
315 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
316 if (vma->vm_start < prev)
317 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
318 if (vma->vm_start < pend)
319 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
320 if (vma->vm_start > vma->vm_end)
321 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
322 i++;
323 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800324 prev = vma->vm_start;
325 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 }
327 j = 0;
328 for (nd = pn; nd; nd = rb_prev(nd)) {
329 j++;
330 }
331 if (i != j)
332 printk("backwards %d, forwards %d\n", j, i), i = 0;
333 return i;
334}
335
336void validate_mm(struct mm_struct *mm)
337{
338 int bug = 0;
339 int i = 0;
340 struct vm_area_struct *tmp = mm->mmap;
341 while (tmp) {
342 tmp = tmp->vm_next;
343 i++;
344 }
345 if (i != mm->map_count)
346 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
347 i = browse_rb(&mm->mm_rb);
348 if (i != mm->map_count)
349 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200350 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351}
352#else
353#define validate_mm(mm) do { } while (0)
354#endif
355
356static struct vm_area_struct *
357find_vma_prepare(struct mm_struct *mm, unsigned long addr,
358 struct vm_area_struct **pprev, struct rb_node ***rb_link,
359 struct rb_node ** rb_parent)
360{
361 struct vm_area_struct * vma;
362 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
363
364 __rb_link = &mm->mm_rb.rb_node;
365 rb_prev = __rb_parent = NULL;
366 vma = NULL;
367
368 while (*__rb_link) {
369 struct vm_area_struct *vma_tmp;
370
371 __rb_parent = *__rb_link;
372 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
373
374 if (vma_tmp->vm_end > addr) {
375 vma = vma_tmp;
376 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700377 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 __rb_link = &__rb_parent->rb_left;
379 } else {
380 rb_prev = __rb_parent;
381 __rb_link = &__rb_parent->rb_right;
382 }
383 }
384
385 *pprev = NULL;
386 if (rb_prev)
387 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
388 *rb_link = __rb_link;
389 *rb_parent = __rb_parent;
390 return vma;
391}
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
394 struct rb_node **rb_link, struct rb_node *rb_parent)
395{
396 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
397 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
398}
399
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700400static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
ZhenwenXu48aae422009-01-06 14:40:21 -0800402 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
404 file = vma->vm_file;
405 if (file) {
406 struct address_space *mapping = file->f_mapping;
407
408 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800409 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 if (vma->vm_flags & VM_SHARED)
411 mapping->i_mmap_writable++;
412
413 flush_dcache_mmap_lock(mapping);
414 if (unlikely(vma->vm_flags & VM_NONLINEAR))
415 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
416 else
417 vma_prio_tree_insert(vma, &mapping->i_mmap);
418 flush_dcache_mmap_unlock(mapping);
419 }
420}
421
422static void
423__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
424 struct vm_area_struct *prev, struct rb_node **rb_link,
425 struct rb_node *rb_parent)
426{
427 __vma_link_list(mm, vma, prev, rb_parent);
428 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429}
430
431static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
432 struct vm_area_struct *prev, struct rb_node **rb_link,
433 struct rb_node *rb_parent)
434{
435 struct address_space *mapping = NULL;
436
437 if (vma->vm_file)
438 mapping = vma->vm_file->f_mapping;
439
Peter Zijlstra97a89412011-05-24 17:12:04 -0700440 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700441 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
443 __vma_link(mm, vma, prev, rb_link, rb_parent);
444 __vma_link_file(vma);
445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700447 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449 mm->map_count++;
450 validate_mm(mm);
451}
452
453/*
454 * Helper for vma_adjust in the split_vma insert case:
455 * insert vm structure into list and rbtree and anon_vma,
456 * but it has already been inserted into prio_tree earlier.
457 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800458static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
ZhenwenXu48aae422009-01-06 14:40:21 -0800460 struct vm_area_struct *__vma, *prev;
461 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200464 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 __vma_link(mm, vma, prev, rb_link, rb_parent);
466 mm->map_count++;
467}
468
469static inline void
470__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
471 struct vm_area_struct *prev)
472{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700473 struct vm_area_struct *next = vma->vm_next;
474
475 prev->vm_next = next;
476 if (next)
477 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 rb_erase(&vma->vm_rb, &mm->mm_rb);
479 if (mm->mmap_cache == vma)
480 mm->mmap_cache = prev;
481}
482
483/*
484 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
485 * is already present in an i_mmap tree without adjusting the tree.
486 * The following helper function should be used when such adjustments
487 * are necessary. The "insert" vma (if any) is to be inserted
488 * before we drop the necessary locks.
489 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800490int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
492{
493 struct mm_struct *mm = vma->vm_mm;
494 struct vm_area_struct *next = vma->vm_next;
495 struct vm_area_struct *importer = NULL;
496 struct address_space *mapping = NULL;
497 struct prio_tree_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700498 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 long adjust_next = 0;
501 int remove_next = 0;
502
503 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700504 struct vm_area_struct *exporter = NULL;
505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 if (end >= next->vm_end) {
507 /*
508 * vma expands, overlapping all the next, and
509 * perhaps the one after too (mprotect case 6).
510 */
511again: remove_next = 1 + (end > next->vm_end);
512 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700513 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 importer = vma;
515 } else if (end > next->vm_start) {
516 /*
517 * vma expands, overlapping part of the next:
518 * mprotect case 5 shifting the boundary up.
519 */
520 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700521 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 importer = vma;
523 } else if (end < vma->vm_end) {
524 /*
525 * vma shrinks, and !insert tells it's not
526 * split_vma inserting another: so it must be
527 * mprotect case 4 shifting the boundary down.
528 */
529 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700530 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 importer = next;
532 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Rik van Riel5beb4932010-03-05 13:42:07 -0800534 /*
535 * Easily overlooked: when mprotect shifts the boundary,
536 * make sure the expanding vma has anon_vma set if the
537 * shrinking vma had, to cover any anon pages imported.
538 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700539 if (exporter && exporter->anon_vma && !importer->anon_vma) {
540 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800541 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700542 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800543 }
544 }
545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 if (file) {
547 mapping = file->f_mapping;
548 if (!(vma->vm_flags & VM_NONLINEAR))
549 root = &mapping->i_mmap;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700550 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 /*
553 * Put into prio_tree now, so instantiated pages
554 * are visible to arm/parisc __flush_dcache_page
555 * throughout; but we cannot insert into address
556 * space until vma start or end is updated.
557 */
558 __vma_link_file(insert);
559 }
560 }
561
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800562 vma_adjust_trans_huge(vma, start, end, adjust_next);
563
Rik van Riel012f18002010-08-09 17:18:40 -0700564 /*
565 * When changing only vma->vm_end, we don't really need anon_vma
566 * lock. This is a fairly rare case by itself, but the anon_vma
567 * lock may be shared between many sibling processes. Skipping
568 * the lock for brk adjustments makes a difference sometimes.
569 */
Shaohua Li5f70b962011-05-24 17:11:19 -0700570 if (vma->anon_vma && (importer || start != vma->vm_start)) {
Rik van Riel012f18002010-08-09 17:18:40 -0700571 anon_vma = vma->anon_vma;
572 anon_vma_lock(anon_vma);
573 }
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (root) {
576 flush_dcache_mmap_lock(mapping);
577 vma_prio_tree_remove(vma, root);
578 if (adjust_next)
579 vma_prio_tree_remove(next, root);
580 }
581
582 vma->vm_start = start;
583 vma->vm_end = end;
584 vma->vm_pgoff = pgoff;
585 if (adjust_next) {
586 next->vm_start += adjust_next << PAGE_SHIFT;
587 next->vm_pgoff += adjust_next;
588 }
589
590 if (root) {
591 if (adjust_next)
592 vma_prio_tree_insert(next, root);
593 vma_prio_tree_insert(vma, root);
594 flush_dcache_mmap_unlock(mapping);
595 }
596
597 if (remove_next) {
598 /*
599 * vma_merge has merged next into vma, and needs
600 * us to remove next before dropping the locks.
601 */
602 __vma_unlink(mm, next, vma);
603 if (file)
604 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 } else if (insert) {
606 /*
607 * split_vma has split insert from vma, and needs
608 * us to insert it before dropping the locks
609 * (it may either follow vma or precede it).
610 */
611 __insert_vm_struct(mm, insert);
612 }
613
Rik van Riel012f18002010-08-09 17:18:40 -0700614 if (anon_vma)
615 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700617 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700620 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700622 if (next->vm_flags & VM_EXECUTABLE)
623 removed_exe_file_vma(mm);
624 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800625 if (next->anon_vma)
626 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700628 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 kmem_cache_free(vm_area_cachep, next);
630 /*
631 * In mprotect's case 6 (see comments on vma_merge),
632 * we must remove another next too. It would clutter
633 * up the code too much to do both in one go.
634 */
635 if (remove_next == 2) {
636 next = vma->vm_next;
637 goto again;
638 }
639 }
640
641 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800642
643 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
646/*
647 * If the vma has a ->close operation then the driver probably needs to release
648 * per-vma resources, so we don't attempt to merge those.
649 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650static inline int is_mergeable_vma(struct vm_area_struct *vma,
651 struct file *file, unsigned long vm_flags)
652{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700653 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
654 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 return 0;
656 if (vma->vm_file != file)
657 return 0;
658 if (vma->vm_ops && vma->vm_ops->close)
659 return 0;
660 return 1;
661}
662
663static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700664 struct anon_vma *anon_vma2,
665 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
Shaohua Li965f55d2011-05-24 17:11:20 -0700667 /*
668 * The list_is_singular() test is to avoid merging VMA cloned from
669 * parents. This can improve scalability caused by anon_vma lock.
670 */
671 if ((!anon_vma1 || !anon_vma2) && (!vma ||
672 list_is_singular(&vma->anon_vma_chain)))
673 return 1;
674 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
677/*
678 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
679 * in front of (at a lower virtual address and file offset than) the vma.
680 *
681 * We cannot merge two vmas if they have differently assigned (non-NULL)
682 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
683 *
684 * We don't check here for the merged mmap wrapping around the end of pagecache
685 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
686 * wrap, nor mmaps which cover the final page at index -1UL.
687 */
688static int
689can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
690 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
691{
692 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700693 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 if (vma->vm_pgoff == vm_pgoff)
695 return 1;
696 }
697 return 0;
698}
699
700/*
701 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
702 * beyond (at a higher virtual address and file offset than) the vma.
703 *
704 * We cannot merge two vmas if they have differently assigned (non-NULL)
705 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
706 */
707static int
708can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
709 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
710{
711 if (is_mergeable_vma(vma, file, vm_flags) &&
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 pgoff_t vm_pglen;
714 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
715 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
716 return 1;
717 }
718 return 0;
719}
720
721/*
722 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
723 * whether that can be merged with its predecessor or its successor.
724 * Or both (it neatly fills a hole).
725 *
726 * In most cases - when called for mmap, brk or mremap - [addr,end) is
727 * certain not to be mapped by the time vma_merge is called; but when
728 * called for mprotect, it is certain to be already mapped (either at
729 * an offset within prev, or at the start of next), and the flags of
730 * this area are about to be changed to vm_flags - and the no-change
731 * case has already been eliminated.
732 *
733 * The following mprotect cases have to be considered, where AAAA is
734 * the area passed down from mprotect_fixup, never extending beyond one
735 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
736 *
737 * AAAA AAAA AAAA AAAA
738 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
739 * cannot merge might become might become might become
740 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
741 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
742 * mremap move: PPPPNNNNNNNN 8
743 * AAAA
744 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
745 * might become case 1 below case 2 below case 3 below
746 *
747 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
748 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
749 */
750struct vm_area_struct *vma_merge(struct mm_struct *mm,
751 struct vm_area_struct *prev, unsigned long addr,
752 unsigned long end, unsigned long vm_flags,
753 struct anon_vma *anon_vma, struct file *file,
754 pgoff_t pgoff, struct mempolicy *policy)
755{
756 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
757 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800758 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
760 /*
761 * We later require that vma->vm_flags == vm_flags,
762 * so this tests vma->vm_flags & VM_SPECIAL, too.
763 */
764 if (vm_flags & VM_SPECIAL)
765 return NULL;
766
767 if (prev)
768 next = prev->vm_next;
769 else
770 next = mm->mmap;
771 area = next;
772 if (next && next->vm_end == end) /* cases 6, 7, 8 */
773 next = next->vm_next;
774
775 /*
776 * Can it merge with the predecessor?
777 */
778 if (prev && prev->vm_end == addr &&
779 mpol_equal(vma_policy(prev), policy) &&
780 can_vma_merge_after(prev, vm_flags,
781 anon_vma, file, pgoff)) {
782 /*
783 * OK, it can. Can we now merge in the successor as well?
784 */
785 if (next && end == next->vm_start &&
786 mpol_equal(policy, vma_policy(next)) &&
787 can_vma_merge_before(next, vm_flags,
788 anon_vma, file, pgoff+pglen) &&
789 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700790 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800792 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 next->vm_end, prev->vm_pgoff, NULL);
794 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800795 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800797 if (err)
798 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800799 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 return prev;
801 }
802
803 /*
804 * Can this new request be merged in front of next?
805 */
806 if (next && end == next->vm_start &&
807 mpol_equal(policy, vma_policy(next)) &&
808 can_vma_merge_before(next, vm_flags,
809 anon_vma, file, pgoff+pglen)) {
810 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800811 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 addr, prev->vm_pgoff, NULL);
813 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800814 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 next->vm_pgoff - pglen, 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(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return area;
820 }
821
822 return NULL;
823}
824
825/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700826 * Rough compatbility check to quickly see if it's even worth looking
827 * at sharing an anon_vma.
828 *
829 * They need to have the same vm_file, and the flags can only differ
830 * in things that mprotect may change.
831 *
832 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
833 * we can merge the two vma's. For example, we refuse to merge a vma if
834 * there is a vm_ops->close() function, because that indicates that the
835 * driver is doing some kind of reference counting. But that doesn't
836 * really matter for the anon_vma sharing case.
837 */
838static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
839{
840 return a->vm_end == b->vm_start &&
841 mpol_equal(vma_policy(a), vma_policy(b)) &&
842 a->vm_file == b->vm_file &&
843 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
844 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
845}
846
847/*
848 * Do some basic sanity checking to see if we can re-use the anon_vma
849 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
850 * the same as 'old', the other will be the new one that is trying
851 * to share the anon_vma.
852 *
853 * NOTE! This runs with mm_sem held for reading, so it is possible that
854 * the anon_vma of 'old' is concurrently in the process of being set up
855 * by another page fault trying to merge _that_. But that's ok: if it
856 * is being set up, that automatically means that it will be a singleton
857 * acceptable for merging, so we can do all of this optimistically. But
858 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
859 *
860 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
861 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
862 * is to return an anon_vma that is "complex" due to having gone through
863 * a fork).
864 *
865 * We also make sure that the two vma's are compatible (adjacent,
866 * and with the same memory policies). That's all stable, even with just
867 * a read lock on the mm_sem.
868 */
869static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
870{
871 if (anon_vma_compatible(a, b)) {
872 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
873
874 if (anon_vma && list_is_singular(&old->anon_vma_chain))
875 return anon_vma;
876 }
877 return NULL;
878}
879
880/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
882 * neighbouring vmas for a suitable anon_vma, before it goes off
883 * to allocate a new anon_vma. It checks because a repetitive
884 * sequence of mprotects and faults may otherwise lead to distinct
885 * anon_vmas being allocated, preventing vma merge in subsequent
886 * mprotect.
887 */
888struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
889{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700890 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
893 near = vma->vm_next;
894 if (!near)
895 goto try_prev;
896
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700897 anon_vma = reusable_anon_vma(near, vma, near);
898 if (anon_vma)
899 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -0700901 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if (!near)
903 goto none;
904
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700905 anon_vma = reusable_anon_vma(near, near, vma);
906 if (anon_vma)
907 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908none:
909 /*
910 * There's no absolute need to look only at touching neighbours:
911 * we could search further afield for "compatible" anon_vmas.
912 * But it would probably just be a waste of time searching,
913 * or lead to too many vmas hanging off the same anon_vma.
914 * We're trying to allow mprotect remerging later on,
915 * not trying to minimize memory used for anon_vmas.
916 */
917 return NULL;
918}
919
920#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700921void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 struct file *file, long pages)
923{
924 const unsigned long stack_flags
925 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
926
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (file) {
928 mm->shared_vm += pages;
929 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
930 mm->exec_vm += pages;
931 } else if (flags & stack_flags)
932 mm->stack_vm += pages;
933 if (flags & (VM_RESERVED|VM_IO))
934 mm->reserved_vm += pages;
935}
936#endif /* CONFIG_PROC_FS */
937
938/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700939 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 */
941
ZhenwenXu48aae422009-01-06 14:40:21 -0800942unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 unsigned long len, unsigned long prot,
944 unsigned long flags, unsigned long pgoff)
945{
946 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 struct inode *inode;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900948 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700950 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 /*
953 * Does the application expect PROT_READ to imply PROT_EXEC?
954 *
955 * (the exception is when the underlying filesystem is noexec
956 * mounted, in which case we dont add PROT_EXEC.)
957 */
958 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800959 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 prot |= PROT_EXEC;
961
962 if (!len)
963 return -EINVAL;
964
Eric Paris7cd94142007-11-26 18:47:40 -0500965 if (!(flags & MAP_FIXED))
966 addr = round_hint_to_min(addr);
967
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 /* Careful about overflows.. */
969 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500970 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 return -ENOMEM;
972
973 /* offset overflow? */
974 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
975 return -EOVERFLOW;
976
977 /* Too many mappings? */
978 if (mm->map_count > sysctl_max_map_count)
979 return -ENOMEM;
980
981 /* Obtain the address to map to. we verify (or select) it and ensure
982 * that it represents a valid section of the address space.
983 */
984 addr = get_unmapped_area(file, addr, len, pgoff, flags);
985 if (addr & ~PAGE_MASK)
986 return addr;
987
988 /* Do simple checking here so the lower-level routines won't have
989 * to. we assume access permissions have been handled by the open
990 * of the memory object, so we don't do any here.
991 */
992 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
993 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
994
Huang Shijiecdf7b342009-09-21 17:03:36 -0700995 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 if (!can_do_mlock())
997 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -0700998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 /* mlock MCL_FUTURE? */
1000 if (vm_flags & VM_LOCKED) {
1001 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001002 locked = len >> PAGE_SHIFT;
1003 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001004 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001005 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1007 return -EAGAIN;
1008 }
1009
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001010 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
1012 if (file) {
1013 switch (flags & MAP_TYPE) {
1014 case MAP_SHARED:
1015 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1016 return -EACCES;
1017
1018 /*
1019 * Make sure we don't allow writing to an append-only
1020 * file..
1021 */
1022 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1023 return -EACCES;
1024
1025 /*
1026 * Make sure there are no mandatory locks on the file.
1027 */
1028 if (locks_verify_locked(inode))
1029 return -EAGAIN;
1030
1031 vm_flags |= VM_SHARED | VM_MAYSHARE;
1032 if (!(file->f_mode & FMODE_WRITE))
1033 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1034
1035 /* fall through */
1036 case MAP_PRIVATE:
1037 if (!(file->f_mode & FMODE_READ))
1038 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001039 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001040 if (vm_flags & VM_EXEC)
1041 return -EPERM;
1042 vm_flags &= ~VM_MAYEXEC;
1043 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001044
1045 if (!file->f_op || !file->f_op->mmap)
1046 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 break;
1048
1049 default:
1050 return -EINVAL;
1051 }
1052 } else {
1053 switch (flags & MAP_TYPE) {
1054 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001055 /*
1056 * Ignore pgoff.
1057 */
1058 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 vm_flags |= VM_SHARED | VM_MAYSHARE;
1060 break;
1061 case MAP_PRIVATE:
1062 /*
1063 * Set pgoff according to addr for anon_vma.
1064 */
1065 pgoff = addr >> PAGE_SHIFT;
1066 break;
1067 default:
1068 return -EINVAL;
1069 }
1070 }
1071
Eric Parised032182007-06-28 15:55:21 -04001072 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (error)
1074 return error;
Eric Parised032182007-06-28 15:55:21 -04001075
Mel Gorman5a6fe122009-02-10 14:02:27 +00001076 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001077}
1078EXPORT_SYMBOL(do_mmap_pgoff);
1079
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001080SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1081 unsigned long, prot, unsigned long, flags,
1082 unsigned long, fd, unsigned long, pgoff)
1083{
1084 struct file *file = NULL;
1085 unsigned long retval = -EBADF;
1086
1087 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001088 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001089 if (unlikely(flags & MAP_HUGETLB))
1090 return -EINVAL;
1091 file = fget(fd);
1092 if (!file)
1093 goto out;
1094 } else if (flags & MAP_HUGETLB) {
1095 struct user_struct *user = NULL;
1096 /*
1097 * VM_NORESERVE is used because the reservations will be
1098 * taken when vm_ops->mmap() is called
1099 * A dummy user value is used because we are not locking
1100 * memory so no accounting is necessary
1101 */
1102 len = ALIGN(len, huge_page_size(&default_hstate));
1103 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1104 &user, HUGETLB_ANONHUGE_INODE);
1105 if (IS_ERR(file))
1106 return PTR_ERR(file);
1107 }
1108
1109 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1110
1111 down_write(&current->mm->mmap_sem);
1112 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1113 up_write(&current->mm->mmap_sem);
1114
1115 if (file)
1116 fput(file);
1117out:
1118 return retval;
1119}
1120
Christoph Hellwiga4679372010-03-10 15:21:15 -08001121#ifdef __ARCH_WANT_SYS_OLD_MMAP
1122struct mmap_arg_struct {
1123 unsigned long addr;
1124 unsigned long len;
1125 unsigned long prot;
1126 unsigned long flags;
1127 unsigned long fd;
1128 unsigned long offset;
1129};
1130
1131SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1132{
1133 struct mmap_arg_struct a;
1134
1135 if (copy_from_user(&a, arg, sizeof(a)))
1136 return -EFAULT;
1137 if (a.offset & ~PAGE_MASK)
1138 return -EINVAL;
1139
1140 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1141 a.offset >> PAGE_SHIFT);
1142}
1143#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1144
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001145/*
1146 * Some shared mappigns will want the pages marked read-only
1147 * to track write events. If so, we'll downgrade vm_page_prot
1148 * to the private version (using protection_map[] without the
1149 * VM_SHARED bit).
1150 */
1151int vma_wants_writenotify(struct vm_area_struct *vma)
1152{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001153 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001154
1155 /* If it was private or non-writable, the write bit is already clear */
1156 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1157 return 0;
1158
1159 /* The backer wishes to know when pages are first written to? */
1160 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1161 return 1;
1162
1163 /* The open routine did something to the protections already? */
1164 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001165 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001166 return 0;
1167
1168 /* Specialty mapping? */
1169 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1170 return 0;
1171
1172 /* Can the mapping track the dirty pages? */
1173 return vma->vm_file && vma->vm_file->f_mapping &&
1174 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1175}
1176
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001177/*
1178 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001179 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001180 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001181static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001182{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001183 /*
1184 * hugetlb has its own accounting separate from the core VM
1185 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1186 */
1187 if (file && is_file_hugepages(file))
1188 return 0;
1189
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001190 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1191}
1192
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001193unsigned long mmap_region(struct file *file, unsigned long addr,
1194 unsigned long len, unsigned long flags,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001195 vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001196{
1197 struct mm_struct *mm = current->mm;
1198 struct vm_area_struct *vma, *prev;
1199 int correct_wcount = 0;
1200 int error;
1201 struct rb_node **rb_link, *rb_parent;
1202 unsigned long charged = 0;
1203 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 /* Clear old maps */
1206 error = -ENOMEM;
1207munmap_back:
1208 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1209 if (vma && vma->vm_start < addr + len) {
1210 if (do_munmap(mm, addr, len))
1211 return -ENOMEM;
1212 goto munmap_back;
1213 }
1214
1215 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001216 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 return -ENOMEM;
1218
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001219 /*
1220 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001221 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001222 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001223 if ((flags & MAP_NORESERVE)) {
1224 /* We honor MAP_NORESERVE if allowed to overcommit */
1225 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1226 vm_flags |= VM_NORESERVE;
1227
1228 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1229 if (file && is_file_hugepages(file))
1230 vm_flags |= VM_NORESERVE;
1231 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001232
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001233 /*
1234 * Private writable mapping: check memory availability
1235 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001236 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001237 charged = len >> PAGE_SHIFT;
1238 if (security_vm_enough_memory(charged))
1239 return -ENOMEM;
1240 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 }
1242
1243 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001244 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001246 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1247 if (vma)
1248 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250 /*
1251 * Determine the object being mapped and call the appropriate
1252 * specific mapper. the address has already been validated, but
1253 * not unmapped, but the maps are removed from the list.
1254 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001255 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 if (!vma) {
1257 error = -ENOMEM;
1258 goto unacct_error;
1259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261 vma->vm_mm = mm;
1262 vma->vm_start = addr;
1263 vma->vm_end = addr + len;
1264 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001265 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001267 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
1269 if (file) {
1270 error = -EINVAL;
1271 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1272 goto free_vma;
1273 if (vm_flags & VM_DENYWRITE) {
1274 error = deny_write_access(file);
1275 if (error)
1276 goto free_vma;
1277 correct_wcount = 1;
1278 }
1279 vma->vm_file = file;
1280 get_file(file);
1281 error = file->f_op->mmap(file, vma);
1282 if (error)
1283 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001284 if (vm_flags & VM_EXECUTABLE)
1285 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001286
1287 /* Can addr have changed??
1288 *
1289 * Answer: Yes, several device drivers can do it in their
1290 * f_op->mmap method. -DaveM
1291 */
1292 addr = vma->vm_start;
1293 pgoff = vma->vm_pgoff;
1294 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 } else if (vm_flags & VM_SHARED) {
1296 error = shmem_zero_setup(vma);
1297 if (error)
1298 goto free_vma;
1299 }
1300
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001301 if (vma_wants_writenotify(vma)) {
1302 pgprot_t pprot = vma->vm_page_prot;
1303
1304 /* Can vma->vm_page_prot have changed??
1305 *
1306 * Answer: Yes, drivers may have changed it in their
1307 * f_op->mmap method.
1308 *
1309 * Ensures that vmas marked as uncached stay that way.
1310 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001311 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001312 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1313 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1314 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001315
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001316 vma_link(mm, vma, prev, rb_link, rb_parent);
1317 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001318
1319 /* Once vma denies write, undo our temporary denial count */
1320 if (correct_wcount)
1321 atomic_inc(&inode->i_writecount);
1322out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001323 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001326 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001328 if (!mlock_vma_pages_range(vma, addr, addr + len))
1329 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001330 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001331 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 return addr;
1333
1334unmap_and_free_vma:
1335 if (correct_wcount)
1336 atomic_inc(&inode->i_writecount);
1337 vma->vm_file = NULL;
1338 fput(file);
1339
1340 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001341 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1342 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343free_vma:
1344 kmem_cache_free(vm_area_cachep, vma);
1345unacct_error:
1346 if (charged)
1347 vm_unacct_memory(charged);
1348 return error;
1349}
1350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351/* Get an address range which is currently unmapped.
1352 * For shmat() with addr=0.
1353 *
1354 * Ugly calling convention alert:
1355 * Return value with the low bits set means error value,
1356 * ie
1357 * if (ret & ~PAGE_MASK)
1358 * error = ret;
1359 *
1360 * This function "knows" that -ENOMEM has the bits set.
1361 */
1362#ifndef HAVE_ARCH_UNMAPPED_AREA
1363unsigned long
1364arch_get_unmapped_area(struct file *filp, unsigned long addr,
1365 unsigned long len, unsigned long pgoff, unsigned long flags)
1366{
1367 struct mm_struct *mm = current->mm;
1368 struct vm_area_struct *vma;
1369 unsigned long start_addr;
1370
1371 if (len > TASK_SIZE)
1372 return -ENOMEM;
1373
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001374 if (flags & MAP_FIXED)
1375 return addr;
1376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 if (addr) {
1378 addr = PAGE_ALIGN(addr);
1379 vma = find_vma(mm, addr);
1380 if (TASK_SIZE - len >= addr &&
1381 (!vma || addr + len <= vma->vm_start))
1382 return addr;
1383 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001384 if (len > mm->cached_hole_size) {
1385 start_addr = addr = mm->free_area_cache;
1386 } else {
1387 start_addr = addr = TASK_UNMAPPED_BASE;
1388 mm->cached_hole_size = 0;
1389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391full_search:
1392 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1393 /* At this point: (!vma || addr < vma->vm_end). */
1394 if (TASK_SIZE - len < addr) {
1395 /*
1396 * Start a new search - just in case we missed
1397 * some holes.
1398 */
1399 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001400 addr = TASK_UNMAPPED_BASE;
1401 start_addr = addr;
1402 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 goto full_search;
1404 }
1405 return -ENOMEM;
1406 }
1407 if (!vma || addr + len <= vma->vm_start) {
1408 /*
1409 * Remember the place where we stopped the search:
1410 */
1411 mm->free_area_cache = addr + len;
1412 return addr;
1413 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001414 if (addr + mm->cached_hole_size < vma->vm_start)
1415 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 addr = vma->vm_end;
1417 }
1418}
1419#endif
1420
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001421void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422{
1423 /*
1424 * Is this a new hole at the lowest possible address?
1425 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001426 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1427 mm->free_area_cache = addr;
1428 mm->cached_hole_size = ~0UL;
1429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430}
1431
1432/*
1433 * This mmap-allocator allocates new areas top-down from below the
1434 * stack's low limit (the base):
1435 */
1436#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1437unsigned long
1438arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1439 const unsigned long len, const unsigned long pgoff,
1440 const unsigned long flags)
1441{
1442 struct vm_area_struct *vma;
1443 struct mm_struct *mm = current->mm;
1444 unsigned long addr = addr0;
1445
1446 /* requested length too big for entire address space */
1447 if (len > TASK_SIZE)
1448 return -ENOMEM;
1449
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001450 if (flags & MAP_FIXED)
1451 return addr;
1452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 /* requesting a specific address */
1454 if (addr) {
1455 addr = PAGE_ALIGN(addr);
1456 vma = find_vma(mm, addr);
1457 if (TASK_SIZE - len >= addr &&
1458 (!vma || addr + len <= vma->vm_start))
1459 return addr;
1460 }
1461
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001462 /* check if free_area_cache is useful for us */
1463 if (len <= mm->cached_hole_size) {
1464 mm->cached_hole_size = 0;
1465 mm->free_area_cache = mm->mmap_base;
1466 }
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 /* either no address requested or can't fit in requested address hole */
1469 addr = mm->free_area_cache;
1470
1471 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001472 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 vma = find_vma(mm, addr-len);
1474 if (!vma || addr <= vma->vm_start)
1475 /* remember the address as a hint for next time */
1476 return (mm->free_area_cache = addr-len);
1477 }
1478
Chris Wright73219d12005-06-21 17:14:52 -07001479 if (mm->mmap_base < len)
1480 goto bottomup;
1481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 addr = mm->mmap_base-len;
1483
1484 do {
1485 /*
1486 * Lookup failure means no vma is above this address,
1487 * else if new region fits below vma->vm_start,
1488 * return with success:
1489 */
1490 vma = find_vma(mm, addr);
1491 if (!vma || addr+len <= vma->vm_start)
1492 /* remember the address as a hint for next time */
1493 return (mm->free_area_cache = addr);
1494
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001495 /* remember the largest hole we saw so far */
1496 if (addr + mm->cached_hole_size < vma->vm_start)
1497 mm->cached_hole_size = vma->vm_start - addr;
1498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 /* try just below the current vma->vm_start */
1500 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001501 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
Chris Wright73219d12005-06-21 17:14:52 -07001503bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 /*
1505 * A failed mmap() very likely causes application failure,
1506 * so fall back to the bottom-up function here. This scenario
1507 * can happen with large stack limits and large mmap()
1508 * allocations.
1509 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001510 mm->cached_hole_size = ~0UL;
1511 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1513 /*
1514 * Restore the topdown base:
1515 */
1516 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001517 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
1519 return addr;
1520}
1521#endif
1522
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001523void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 /*
1526 * Is this a new hole at the highest possible address?
1527 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001528 if (addr > mm->free_area_cache)
1529 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001532 if (mm->free_area_cache > mm->mmap_base)
1533 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534}
1535
1536unsigned long
1537get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1538 unsigned long pgoff, unsigned long flags)
1539{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001540 unsigned long (*get_area)(struct file *, unsigned long,
1541 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
Al Viro9206de92009-12-03 15:23:11 -05001543 unsigned long error = arch_mmap_check(addr, len, flags);
1544 if (error)
1545 return error;
1546
1547 /* Careful about overflows.. */
1548 if (len > TASK_SIZE)
1549 return -ENOMEM;
1550
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001551 get_area = current->mm->get_unmapped_area;
1552 if (file && file->f_op && file->f_op->get_unmapped_area)
1553 get_area = file->f_op->get_unmapped_area;
1554 addr = get_area(file, addr, len, pgoff, flags);
1555 if (IS_ERR_VALUE(addr))
1556 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001558 if (addr > TASK_SIZE - len)
1559 return -ENOMEM;
1560 if (addr & ~PAGE_MASK)
1561 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001562
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001563 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564}
1565
1566EXPORT_SYMBOL(get_unmapped_area);
1567
1568/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001569struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570{
1571 struct vm_area_struct *vma = NULL;
1572
1573 if (mm) {
1574 /* Check the cache first. */
1575 /* (Cache hit rate is typically around 35%.) */
1576 vma = mm->mmap_cache;
1577 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1578 struct rb_node * rb_node;
1579
1580 rb_node = mm->mm_rb.rb_node;
1581 vma = NULL;
1582
1583 while (rb_node) {
1584 struct vm_area_struct * vma_tmp;
1585
1586 vma_tmp = rb_entry(rb_node,
1587 struct vm_area_struct, vm_rb);
1588
1589 if (vma_tmp->vm_end > addr) {
1590 vma = vma_tmp;
1591 if (vma_tmp->vm_start <= addr)
1592 break;
1593 rb_node = rb_node->rb_left;
1594 } else
1595 rb_node = rb_node->rb_right;
1596 }
1597 if (vma)
1598 mm->mmap_cache = vma;
1599 }
1600 }
1601 return vma;
1602}
1603
1604EXPORT_SYMBOL(find_vma);
1605
1606/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1607struct vm_area_struct *
1608find_vma_prev(struct mm_struct *mm, unsigned long addr,
1609 struct vm_area_struct **pprev)
1610{
1611 struct vm_area_struct *vma = NULL, *prev = NULL;
ZhenwenXu48aae422009-01-06 14:40:21 -08001612 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 if (!mm)
1614 goto out;
1615
1616 /* Guard against addr being lower than the first VMA */
1617 vma = mm->mmap;
1618
1619 /* Go through the RB tree quickly. */
1620 rb_node = mm->mm_rb.rb_node;
1621
1622 while (rb_node) {
1623 struct vm_area_struct *vma_tmp;
1624 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1625
1626 if (addr < vma_tmp->vm_end) {
1627 rb_node = rb_node->rb_left;
1628 } else {
1629 prev = vma_tmp;
1630 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1631 break;
1632 rb_node = rb_node->rb_right;
1633 }
1634 }
1635
1636out:
1637 *pprev = prev;
1638 return prev ? prev->vm_next : vma;
1639}
1640
1641/*
1642 * Verify that the stack growth is acceptable and
1643 * update accounting. This is shared with both the
1644 * grow-up and grow-down cases.
1645 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001646static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 struct mm_struct *mm = vma->vm_mm;
1649 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001650 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001653 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 return -ENOMEM;
1655
1656 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001657 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 return -ENOMEM;
1659
1660 /* mlock limit tests */
1661 if (vma->vm_flags & VM_LOCKED) {
1662 unsigned long locked;
1663 unsigned long limit;
1664 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001665 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1666 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 if (locked > limit && !capable(CAP_IPC_LOCK))
1668 return -ENOMEM;
1669 }
1670
Adam Litke0d59a012007-01-30 14:35:39 -08001671 /* Check to ensure the stack will not grow into a hugetlb-only region */
1672 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1673 vma->vm_end - size;
1674 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1675 return -EFAULT;
1676
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 /*
1678 * Overcommit.. This must be the final test, as it will
1679 * update security statistics.
1680 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001681 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 return -ENOMEM;
1683
1684 /* Ok, everything looks good - let it rip */
1685 mm->total_vm += grow;
1686 if (vma->vm_flags & VM_LOCKED)
1687 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001688 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 return 0;
1690}
1691
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001692#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001694 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1695 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001697int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698{
1699 int error;
1700
1701 if (!(vma->vm_flags & VM_GROWSUP))
1702 return -EFAULT;
1703
1704 /*
1705 * We must make sure the anon_vma is allocated
1706 * so that the anon_vma locking is not a noop.
1707 */
1708 if (unlikely(anon_vma_prepare(vma)))
1709 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001710 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712 /*
1713 * vma->vm_start/vm_end cannot change under us because the caller
1714 * is required to hold the mmap_sem in read mode. We need the
1715 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001716 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 */
Helge Deller06b32f32006-12-19 19:28:33 +01001718 if (address < PAGE_ALIGN(address+4))
1719 address = PAGE_ALIGN(address+4);
1720 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001721 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001722 return -ENOMEM;
1723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 error = 0;
1725
1726 /* Somebody else might have raced and expanded it already */
1727 if (address > vma->vm_end) {
1728 unsigned long size, grow;
1729
1730 size = address - vma->vm_start;
1731 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1732
Hugh Dickins42c36f62011-05-09 17:44:42 -07001733 error = -ENOMEM;
1734 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1735 error = acct_stack_growth(vma, size, grow);
1736 if (!error) {
1737 vma->vm_end = address;
1738 perf_event_mmap(vma);
1739 }
Eric B Munson3af9e852010-05-18 15:30:49 +01001740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001742 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001743 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 return error;
1745}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001746#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748/*
1749 * vma is the first one with address < vma->vm_start. Have to extend vma.
1750 */
Michal Hockod05f3162011-05-24 17:11:44 -07001751int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001752 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
1754 int error;
1755
1756 /*
1757 * We must make sure the anon_vma is allocated
1758 * so that the anon_vma locking is not a noop.
1759 */
1760 if (unlikely(anon_vma_prepare(vma)))
1761 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001762
1763 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001764 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001765 if (error)
1766 return error;
1767
Rik van Rielbb4a3402010-08-09 17:18:37 -07001768 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 /*
1771 * vma->vm_start/vm_end cannot change under us because the caller
1772 * is required to hold the mmap_sem in read mode. We need the
1773 * anon_vma lock to serialize against concurrent expand_stacks.
1774 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
1776 /* Somebody else might have raced and expanded it already */
1777 if (address < vma->vm_start) {
1778 unsigned long size, grow;
1779
1780 size = vma->vm_end - address;
1781 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1782
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001783 error = -ENOMEM;
1784 if (grow <= vma->vm_pgoff) {
1785 error = acct_stack_growth(vma, size, grow);
1786 if (!error) {
1787 vma->vm_start = address;
1788 vma->vm_pgoff -= grow;
1789 perf_event_mmap(vma);
1790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 }
1792 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001793 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001794 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 return error;
1796}
1797
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001798#ifdef CONFIG_STACK_GROWSUP
1799int expand_stack(struct vm_area_struct *vma, unsigned long address)
1800{
1801 return expand_upwards(vma, address);
1802}
1803
1804struct vm_area_struct *
1805find_extend_vma(struct mm_struct *mm, unsigned long addr)
1806{
1807 struct vm_area_struct *vma, *prev;
1808
1809 addr &= PAGE_MASK;
1810 vma = find_vma_prev(mm, addr, &prev);
1811 if (vma && (vma->vm_start <= addr))
1812 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001813 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001814 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001815 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001816 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001817 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001818 return prev;
1819}
1820#else
1821int expand_stack(struct vm_area_struct *vma, unsigned long address)
1822{
1823 return expand_downwards(vma, address);
1824}
1825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826struct vm_area_struct *
1827find_extend_vma(struct mm_struct * mm, unsigned long addr)
1828{
1829 struct vm_area_struct * vma;
1830 unsigned long start;
1831
1832 addr &= PAGE_MASK;
1833 vma = find_vma(mm,addr);
1834 if (!vma)
1835 return NULL;
1836 if (vma->vm_start <= addr)
1837 return vma;
1838 if (!(vma->vm_flags & VM_GROWSDOWN))
1839 return NULL;
1840 start = vma->vm_start;
1841 if (expand_stack(vma, addr))
1842 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001843 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001844 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001845 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 return vma;
1847}
1848#endif
1849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001851 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001853 *
1854 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001856static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001858 /* Update high watermark before we lower total_vm */
1859 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001861 long nrpages = vma_pages(vma);
1862
1863 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001864 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001865 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001866 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 validate_mm(mm);
1868}
1869
1870/*
1871 * Get rid of page table information in the indicated region.
1872 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001873 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 */
1875static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001876 struct vm_area_struct *vma, struct vm_area_struct *prev,
1877 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001879 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001880 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 unsigned long nr_accounted = 0;
1882
1883 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001884 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001885 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001886 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 vm_unacct_memory(nr_accounted);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001888 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1889 next ? next->vm_start : 0);
1890 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891}
1892
1893/*
1894 * Create a list of vma's touched by the unmap, removing them from the mm's
1895 * vma list as we go..
1896 */
1897static void
1898detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1899 struct vm_area_struct *prev, unsigned long end)
1900{
1901 struct vm_area_struct **insertion_point;
1902 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001903 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
1905 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001906 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 do {
1908 rb_erase(&vma->vm_rb, &mm->mm_rb);
1909 mm->map_count--;
1910 tail_vma = vma;
1911 vma = vma->vm_next;
1912 } while (vma && vma->vm_start < end);
1913 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001914 if (vma)
1915 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001917 if (mm->unmap_area == arch_unmap_area)
1918 addr = prev ? prev->vm_end : mm->mmap_base;
1919 else
1920 addr = vma ? vma->vm_start : mm->mmap_base;
1921 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 mm->mmap_cache = NULL; /* Kill the cache. */
1923}
1924
1925/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001926 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1927 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001929static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 unsigned long addr, int new_below)
1931{
1932 struct mempolicy *pol;
1933 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001934 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Andi Kleena5516432008-07-23 21:27:41 -07001936 if (is_vm_hugetlb_page(vma) && (addr &
1937 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 return -EINVAL;
1939
Christoph Lametere94b1762006-12-06 20:33:17 -08001940 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001942 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944 /* most fields are the same, copy all, and then fixup */
1945 *new = *vma;
1946
Rik van Riel5beb4932010-03-05 13:42:07 -08001947 INIT_LIST_HEAD(&new->anon_vma_chain);
1948
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 if (new_below)
1950 new->vm_end = addr;
1951 else {
1952 new->vm_start = addr;
1953 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1954 }
1955
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001956 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001958 err = PTR_ERR(pol);
1959 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 }
1961 vma_set_policy(new, pol);
1962
Rik van Riel5beb4932010-03-05 13:42:07 -08001963 if (anon_vma_clone(new, vma))
1964 goto out_free_mpol;
1965
Matt Helsley925d1c42008-04-29 01:01:36 -07001966 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001968 if (vma->vm_flags & VM_EXECUTABLE)
1969 added_exe_file_vma(mm);
1970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
1972 if (new->vm_ops && new->vm_ops->open)
1973 new->vm_ops->open(new);
1974
1975 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08001976 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1978 else
Rik van Riel5beb4932010-03-05 13:42:07 -08001979 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Rik van Riel5beb4932010-03-05 13:42:07 -08001981 /* Success. */
1982 if (!err)
1983 return 0;
1984
1985 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04001986 if (new->vm_ops && new->vm_ops->close)
1987 new->vm_ops->close(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08001988 if (new->vm_file) {
1989 if (vma->vm_flags & VM_EXECUTABLE)
1990 removed_exe_file_vma(mm);
1991 fput(new->vm_file);
1992 }
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07001993 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08001994 out_free_mpol:
1995 mpol_put(pol);
1996 out_free_vma:
1997 kmem_cache_free(vm_area_cachep, new);
1998 out_err:
1999 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000}
2001
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002002/*
2003 * Split a vma into two pieces at address 'addr', a new vma is allocated
2004 * either for the first part or the tail.
2005 */
2006int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2007 unsigned long addr, int new_below)
2008{
2009 if (mm->map_count >= sysctl_max_map_count)
2010 return -ENOMEM;
2011
2012 return __split_vma(mm, vma, addr, new_below);
2013}
2014
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015/* Munmap is split into 2 main parts -- this part which finds
2016 * what needs doing, and the areas themselves, which do the
2017 * work. This now handles partial unmappings.
2018 * Jeremy Fitzhardinge <jeremy@goop.org>
2019 */
2020int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2021{
2022 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002023 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024
2025 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2026 return -EINVAL;
2027
2028 if ((len = PAGE_ALIGN(len)) == 0)
2029 return -EINVAL;
2030
2031 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002032 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002033 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002035 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002036 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
2038 /* if it doesn't overlap, we have nothing.. */
2039 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002040 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 return 0;
2042
2043 /*
2044 * If we need to split any vma, do it now to save pain later.
2045 *
2046 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2047 * unmapped vm_area_struct will remain in use: so lower split_vma
2048 * places tmp vma above, and higher split_vma places tmp vma below.
2049 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002050 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002051 int error;
2052
2053 /*
2054 * Make sure that map_count on return from munmap() will
2055 * not exceed its limit; but let map_count go just above
2056 * its limit temporarily, to help free resources as expected.
2057 */
2058 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2059 return -ENOMEM;
2060
2061 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 if (error)
2063 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002064 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
2066
2067 /* Does it split the last one? */
2068 last = find_vma(mm, end);
2069 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002070 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 if (error)
2072 return error;
2073 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002074 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
2076 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002077 * unlock any mlock()ed ranges before detaching vmas
2078 */
2079 if (mm->locked_vm) {
2080 struct vm_area_struct *tmp = vma;
2081 while (tmp && tmp->vm_start < end) {
2082 if (tmp->vm_flags & VM_LOCKED) {
2083 mm->locked_vm -= vma_pages(tmp);
2084 munlock_vma_pages_all(tmp);
2085 }
2086 tmp = tmp->vm_next;
2087 }
2088 }
2089
2090 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 * Remove the vma's, and unmap the actual pages
2092 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002093 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2094 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
2096 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002097 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
2099 return 0;
2100}
2101
2102EXPORT_SYMBOL(do_munmap);
2103
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002104SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105{
2106 int ret;
2107 struct mm_struct *mm = current->mm;
2108
2109 profile_munmap(addr);
2110
2111 down_write(&mm->mmap_sem);
2112 ret = do_munmap(mm, addr, len);
2113 up_write(&mm->mmap_sem);
2114 return ret;
2115}
2116
2117static inline void verify_mm_writelocked(struct mm_struct *mm)
2118{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002119#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2121 WARN_ON(1);
2122 up_read(&mm->mmap_sem);
2123 }
2124#endif
2125}
2126
2127/*
2128 * this is really a simplified "do_mmap". it only handles
2129 * anonymous maps. eventually we may be able to do some
2130 * brk-specific accounting here.
2131 */
2132unsigned long do_brk(unsigned long addr, unsigned long len)
2133{
2134 struct mm_struct * mm = current->mm;
2135 struct vm_area_struct * vma, * prev;
2136 unsigned long flags;
2137 struct rb_node ** rb_link, * rb_parent;
2138 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002139 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141 len = PAGE_ALIGN(len);
2142 if (!len)
2143 return addr;
2144
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002145 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002146 if (error)
2147 return error;
2148
Kirill Korotaev3a459752006-09-07 14:17:04 +04002149 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2150
Al Viro2c6a1012009-12-03 19:40:46 -05002151 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2152 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002153 return error;
2154
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 /*
2156 * mlock MCL_FUTURE?
2157 */
2158 if (mm->def_flags & VM_LOCKED) {
2159 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002160 locked = len >> PAGE_SHIFT;
2161 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002162 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002163 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2165 return -EAGAIN;
2166 }
2167
2168 /*
2169 * mm->mmap_sem is required to protect against another thread
2170 * changing the mappings in case we sleep.
2171 */
2172 verify_mm_writelocked(mm);
2173
2174 /*
2175 * Clear old maps. this also does some error checking for us
2176 */
2177 munmap_back:
2178 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2179 if (vma && vma->vm_start < addr + len) {
2180 if (do_munmap(mm, addr, len))
2181 return -ENOMEM;
2182 goto munmap_back;
2183 }
2184
2185 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002186 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 return -ENOMEM;
2188
2189 if (mm->map_count > sysctl_max_map_count)
2190 return -ENOMEM;
2191
2192 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2193 return -ENOMEM;
2194
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002196 vma = vma_merge(mm, prev, addr, addr + len, flags,
2197 NULL, NULL, pgoff, NULL);
2198 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 goto out;
2200
2201 /*
2202 * create a vma struct for an anonymous mapping
2203 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002204 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 if (!vma) {
2206 vm_unacct_memory(len >> PAGE_SHIFT);
2207 return -ENOMEM;
2208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
Rik van Riel5beb4932010-03-05 13:42:07 -08002210 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 vma->vm_mm = mm;
2212 vma->vm_start = addr;
2213 vma->vm_end = addr + len;
2214 vma->vm_pgoff = pgoff;
2215 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002216 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 vma_link(mm, vma, prev, rb_link, rb_parent);
2218out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002219 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 mm->total_vm += len >> PAGE_SHIFT;
2221 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002222 if (!mlock_vma_pages_range(vma, addr, addr + len))
2223 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 }
2225 return addr;
2226}
2227
2228EXPORT_SYMBOL(do_brk);
2229
2230/* Release all mmaps. */
2231void exit_mmap(struct mm_struct *mm)
2232{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002233 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002234 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002236 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002238 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002239 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002240
Rik van Rielba470de2008-10-18 20:26:50 -07002241 if (mm->locked_vm) {
2242 vma = mm->mmap;
2243 while (vma) {
2244 if (vma->vm_flags & VM_LOCKED)
2245 munlock_vma_pages_all(vma);
2246 vma = vma->vm_next;
2247 }
2248 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002249
2250 arch_exit_mmap(mm);
2251
Rik van Rielba470de2008-10-18 20:26:50 -07002252 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002253 if (!vma) /* Can happen if dup_mmap() received an OOM */
2254 return;
2255
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 flush_cache_mm(mm);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002258 tlb_gather_mmu(&tlb, mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002259 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002260 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002261 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 vm_unacct_memory(nr_accounted);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002263
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002264 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2265 tlb_finish_mmu(&tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002268 * Walk the list again, actually closing and freeing it,
2269 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002271 while (vma)
2272 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002273
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002274 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275}
2276
2277/* Insert vm structure into process list sorted by address
2278 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002279 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 */
2281int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2282{
2283 struct vm_area_struct * __vma, * prev;
2284 struct rb_node ** rb_link, * rb_parent;
2285
2286 /*
2287 * The vm_pgoff of a purely anonymous vma should be irrelevant
2288 * until its first write fault, when page's anon_vma and index
2289 * are set. But now set the vm_pgoff it will almost certainly
2290 * end up with (unless mremap moves it elsewhere before that
2291 * first wfault), so /proc/pid/maps tells a consistent story.
2292 *
2293 * By setting it to reflect the virtual start address of the
2294 * vma, merges and splits can happen in a seamless way, just
2295 * using the existing file pgoff checks and manipulations.
2296 * Similarly in do_mmap_pgoff and in do_brk.
2297 */
2298 if (!vma->vm_file) {
2299 BUG_ON(vma->anon_vma);
2300 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2301 }
2302 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2303 if (__vma && __vma->vm_start < vma->vm_end)
2304 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002305 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002306 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002307 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 vma_link(mm, vma, prev, rb_link, rb_parent);
2309 return 0;
2310}
2311
2312/*
2313 * Copy the vma structure to a new location in the same mm,
2314 * prior to moving page table entries, to effect an mremap move.
2315 */
2316struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2317 unsigned long addr, unsigned long len, pgoff_t pgoff)
2318{
2319 struct vm_area_struct *vma = *vmap;
2320 unsigned long vma_start = vma->vm_start;
2321 struct mm_struct *mm = vma->vm_mm;
2322 struct vm_area_struct *new_vma, *prev;
2323 struct rb_node **rb_link, *rb_parent;
2324 struct mempolicy *pol;
2325
2326 /*
2327 * If anonymous vma has not yet been faulted, update new pgoff
2328 * to match new location, to increase its chance of merging.
2329 */
2330 if (!vma->vm_file && !vma->anon_vma)
2331 pgoff = addr >> PAGE_SHIFT;
2332
2333 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2334 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2335 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2336 if (new_vma) {
2337 /*
2338 * Source vma may have been merged into new_vma
2339 */
2340 if (vma_start >= new_vma->vm_start &&
2341 vma_start < new_vma->vm_end)
2342 *vmap = new_vma;
2343 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002344 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 if (new_vma) {
2346 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002347 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002348 if (IS_ERR(pol))
2349 goto out_free_vma;
2350 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2351 if (anon_vma_clone(new_vma, vma))
2352 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 vma_set_policy(new_vma, pol);
2354 new_vma->vm_start = addr;
2355 new_vma->vm_end = addr + len;
2356 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002357 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002359 if (vma->vm_flags & VM_EXECUTABLE)
2360 added_exe_file_vma(mm);
2361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 if (new_vma->vm_ops && new_vma->vm_ops->open)
2363 new_vma->vm_ops->open(new_vma);
2364 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2365 }
2366 }
2367 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002368
2369 out_free_mempol:
2370 mpol_put(pol);
2371 out_free_vma:
2372 kmem_cache_free(vm_area_cachep, new_vma);
2373 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002375
2376/*
2377 * Return true if the calling process may expand its vm space by the passed
2378 * number of pages
2379 */
2380int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2381{
2382 unsigned long cur = mm->total_vm; /* pages */
2383 unsigned long lim;
2384
Jiri Slaby59e99e52010-03-05 13:41:44 -08002385 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002386
2387 if (cur + npages > lim)
2388 return 0;
2389 return 1;
2390}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002391
2392
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002393static int special_mapping_fault(struct vm_area_struct *vma,
2394 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002395{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002396 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002397 struct page **pages;
2398
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002399 /*
2400 * special mappings have no vm_file, and in that case, the mm
2401 * uses vm_pgoff internally. So we have to subtract it from here.
2402 * We are allowed to do this because we are the mm; do not copy
2403 * this code into drivers!
2404 */
2405 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002406
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002407 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2408 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002409
2410 if (*pages) {
2411 struct page *page = *pages;
2412 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002413 vmf->page = page;
2414 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002415 }
2416
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002417 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002418}
2419
2420/*
2421 * Having a close hook prevents vma merging regardless of flags.
2422 */
2423static void special_mapping_close(struct vm_area_struct *vma)
2424{
2425}
2426
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002427static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002428 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002429 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002430};
2431
2432/*
2433 * Called with mm->mmap_sem held for writing.
2434 * Insert a new vma covering the given region, with the given flags.
2435 * Its pages are supplied by the given array of struct page *.
2436 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2437 * The region past the last page supplied will always produce SIGBUS.
2438 * The array pointer and the pages it points to are assumed to stay alive
2439 * for as long as this mapping might exist.
2440 */
2441int install_special_mapping(struct mm_struct *mm,
2442 unsigned long addr, unsigned long len,
2443 unsigned long vm_flags, struct page **pages)
2444{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002445 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002446 struct vm_area_struct *vma;
2447
2448 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2449 if (unlikely(vma == NULL))
2450 return -ENOMEM;
2451
Rik van Riel5beb4932010-03-05 13:42:07 -08002452 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002453 vma->vm_mm = mm;
2454 vma->vm_start = addr;
2455 vma->vm_end = addr + len;
2456
Nick Piggin2f987352008-02-02 03:08:53 +01002457 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002458 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002459
2460 vma->vm_ops = &special_mapping_vmops;
2461 vma->vm_private_data = pages;
2462
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002463 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2464 if (ret)
2465 goto out;
2466
2467 ret = insert_vm_struct(mm, vma);
2468 if (ret)
2469 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002470
2471 mm->total_vm += len >> PAGE_SHIFT;
2472
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002473 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002474
Roland McGrathfa5dc222007-02-08 14:20:41 -08002475 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002476
2477out:
2478 kmem_cache_free(vm_area_cachep, vma);
2479 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002480}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002481
2482static DEFINE_MUTEX(mm_all_locks_mutex);
2483
Peter Zijlstra454ed842008-08-11 09:30:25 +02002484static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002485{
Rik van Riel012f18002010-08-09 17:18:40 -07002486 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002487 /*
2488 * The LSB of head.next can't change from under us
2489 * because we hold the mm_all_locks_mutex.
2490 */
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002491 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002492 /*
2493 * We can safely modify head.next after taking the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002494 * anon_vma->root->mutex. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002495 * the same anon_vma we won't take it again.
2496 *
2497 * No need of atomic instructions here, head.next
2498 * can't change from under us thanks to the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002499 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002500 */
2501 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002502 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002503 BUG();
2504 }
2505}
2506
Peter Zijlstra454ed842008-08-11 09:30:25 +02002507static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002508{
2509 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2510 /*
2511 * AS_MM_ALL_LOCKS can't change from under us because
2512 * we hold the mm_all_locks_mutex.
2513 *
2514 * Operations on ->flags have to be atomic because
2515 * even if AS_MM_ALL_LOCKS is stable thanks to the
2516 * mm_all_locks_mutex, there may be other cpus
2517 * changing other bitflags in parallel to us.
2518 */
2519 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2520 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002521 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002522 }
2523}
2524
2525/*
2526 * This operation locks against the VM for all pte/vma/mm related
2527 * operations that could ever happen on a certain mm. This includes
2528 * vmtruncate, try_to_unmap, and all page faults.
2529 *
2530 * The caller must take the mmap_sem in write mode before calling
2531 * mm_take_all_locks(). The caller isn't allowed to release the
2532 * mmap_sem until mm_drop_all_locks() returns.
2533 *
2534 * mmap_sem in write mode is required in order to block all operations
2535 * that could modify pagetables and free pages without need of
2536 * altering the vma layout (for example populate_range() with
2537 * nonlinear vmas). It's also needed in write mode to avoid new
2538 * anon_vmas to be associated with existing vmas.
2539 *
2540 * A single task can't take more than one mm_take_all_locks() in a row
2541 * or it would deadlock.
2542 *
2543 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2544 * mapping->flags avoid to take the same lock twice, if more than one
2545 * vma in this mm is backed by the same anon_vma or address_space.
2546 *
2547 * We can take all the locks in random order because the VM code
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002548 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002549 * takes more than one of them in a row. Secondly we're protected
2550 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2551 *
2552 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2553 * that may have to take thousand of locks.
2554 *
2555 * mm_take_all_locks() can fail if it's interrupted by signals.
2556 */
2557int mm_take_all_locks(struct mm_struct *mm)
2558{
2559 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002560 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002561
2562 BUG_ON(down_read_trylock(&mm->mmap_sem));
2563
2564 mutex_lock(&mm_all_locks_mutex);
2565
2566 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2567 if (signal_pending(current))
2568 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002569 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002570 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002571 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002572
2573 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2574 if (signal_pending(current))
2575 goto out_unlock;
2576 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002577 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2578 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002579 }
2580
Kautuk Consul584cff52011-10-31 17:08:59 -07002581 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002582
2583out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07002584 mm_drop_all_locks(mm);
2585 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002586}
2587
2588static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2589{
Rik van Riel012f18002010-08-09 17:18:40 -07002590 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002591 /*
2592 * The LSB of head.next can't change to 0 from under
2593 * us because we hold the mm_all_locks_mutex.
2594 *
2595 * We must however clear the bitflag before unlocking
2596 * the vma so the users using the anon_vma->head will
2597 * never see our bitflag.
2598 *
2599 * No need of atomic instructions here, head.next
2600 * can't change from under us until we release the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002601 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002602 */
2603 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002604 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002605 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002606 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002607 }
2608}
2609
2610static void vm_unlock_mapping(struct address_space *mapping)
2611{
2612 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2613 /*
2614 * AS_MM_ALL_LOCKS can't change to 0 from under us
2615 * because we hold the mm_all_locks_mutex.
2616 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002617 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002618 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2619 &mapping->flags))
2620 BUG();
2621 }
2622}
2623
2624/*
2625 * The mmap_sem cannot be released by the caller until
2626 * mm_drop_all_locks() returns.
2627 */
2628void mm_drop_all_locks(struct mm_struct *mm)
2629{
2630 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002631 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002632
2633 BUG_ON(down_read_trylock(&mm->mmap_sem));
2634 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2635
2636 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2637 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002638 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2639 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002640 if (vma->vm_file && vma->vm_file->f_mapping)
2641 vm_unlock_mapping(vma->vm_file->f_mapping);
2642 }
2643
2644 mutex_unlock(&mm_all_locks_mutex);
2645}
David Howells8feae132009-01-08 12:04:47 +00002646
2647/*
2648 * initialise the VMA slab
2649 */
2650void __init mmap_init(void)
2651{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002652 int ret;
2653
2654 ret = percpu_counter_init(&vm_committed_as, 0);
2655 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002656}