blob: a7bf6a31c9f62be11cb8e5819322565b7cf2c266 [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>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040025#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#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/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700454 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
455 * mm's list and rbtree. It has already been inserted into the prio_tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800457static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
ZhenwenXu48aae422009-01-06 14:40:21 -0800459 struct vm_area_struct *__vma, *prev;
460 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200463 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 __vma_link(mm, vma, prev, rb_link, rb_parent);
465 mm->map_count++;
466}
467
468static inline void
469__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
470 struct vm_area_struct *prev)
471{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700472 struct vm_area_struct *next = vma->vm_next;
473
474 prev->vm_next = next;
475 if (next)
476 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 rb_erase(&vma->vm_rb, &mm->mm_rb);
478 if (mm->mmap_cache == vma)
479 mm->mmap_cache = prev;
480}
481
482/*
483 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
484 * is already present in an i_mmap tree without adjusting the tree.
485 * The following helper function should be used when such adjustments
486 * are necessary. The "insert" vma (if any) is to be inserted
487 * before we drop the necessary locks.
488 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800489int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
491{
492 struct mm_struct *mm = vma->vm_mm;
493 struct vm_area_struct *next = vma->vm_next;
494 struct vm_area_struct *importer = NULL;
495 struct address_space *mapping = NULL;
496 struct prio_tree_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700497 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 long adjust_next = 0;
500 int remove_next = 0;
501
502 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700503 struct vm_area_struct *exporter = NULL;
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 if (end >= next->vm_end) {
506 /*
507 * vma expands, overlapping all the next, and
508 * perhaps the one after too (mprotect case 6).
509 */
510again: remove_next = 1 + (end > next->vm_end);
511 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700512 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 importer = vma;
514 } else if (end > next->vm_start) {
515 /*
516 * vma expands, overlapping part of the next:
517 * mprotect case 5 shifting the boundary up.
518 */
519 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700520 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 importer = vma;
522 } else if (end < vma->vm_end) {
523 /*
524 * vma shrinks, and !insert tells it's not
525 * split_vma inserting another: so it must be
526 * mprotect case 4 shifting the boundary down.
527 */
528 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700529 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 importer = next;
531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Rik van Riel5beb4932010-03-05 13:42:07 -0800533 /*
534 * Easily overlooked: when mprotect shifts the boundary,
535 * make sure the expanding vma has anon_vma set if the
536 * shrinking vma had, to cover any anon pages imported.
537 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700538 if (exporter && exporter->anon_vma && !importer->anon_vma) {
539 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800540 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700541 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800542 }
543 }
544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 if (file) {
546 mapping = file->f_mapping;
547 if (!(vma->vm_flags & VM_NONLINEAR))
548 root = &mapping->i_mmap;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700549 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 /*
552 * Put into prio_tree now, so instantiated pages
553 * are visible to arm/parisc __flush_dcache_page
554 * throughout; but we cannot insert into address
555 * space until vma start or end is updated.
556 */
557 __vma_link_file(insert);
558 }
559 }
560
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800561 vma_adjust_trans_huge(vma, start, end, adjust_next);
562
Rik van Riel012f18002010-08-09 17:18:40 -0700563 /*
564 * When changing only vma->vm_end, we don't really need anon_vma
565 * lock. This is a fairly rare case by itself, but the anon_vma
566 * lock may be shared between many sibling processes. Skipping
567 * the lock for brk adjustments makes a difference sometimes.
568 */
Shaohua Li5f70b962011-05-24 17:11:19 -0700569 if (vma->anon_vma && (importer || start != vma->vm_start)) {
Rik van Riel012f18002010-08-09 17:18:40 -0700570 anon_vma = vma->anon_vma;
571 anon_vma_lock(anon_vma);
572 }
573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (root) {
575 flush_dcache_mmap_lock(mapping);
576 vma_prio_tree_remove(vma, root);
577 if (adjust_next)
578 vma_prio_tree_remove(next, root);
579 }
580
581 vma->vm_start = start;
582 vma->vm_end = end;
583 vma->vm_pgoff = pgoff;
584 if (adjust_next) {
585 next->vm_start += adjust_next << PAGE_SHIFT;
586 next->vm_pgoff += adjust_next;
587 }
588
589 if (root) {
590 if (adjust_next)
591 vma_prio_tree_insert(next, root);
592 vma_prio_tree_insert(vma, root);
593 flush_dcache_mmap_unlock(mapping);
594 }
595
596 if (remove_next) {
597 /*
598 * vma_merge has merged next into vma, and needs
599 * us to remove next before dropping the locks.
600 */
601 __vma_unlink(mm, next, vma);
602 if (file)
603 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 } else if (insert) {
605 /*
606 * split_vma has split insert from vma, and needs
607 * us to insert it before dropping the locks
608 * (it may either follow vma or precede it).
609 */
610 __insert_vm_struct(mm, insert);
611 }
612
Rik van Riel012f18002010-08-09 17:18:40 -0700613 if (anon_vma)
614 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700616 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
618 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700619 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700621 if (next->vm_flags & VM_EXECUTABLE)
622 removed_exe_file_vma(mm);
623 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800624 if (next->anon_vma)
625 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700627 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 kmem_cache_free(vm_area_cachep, next);
629 /*
630 * In mprotect's case 6 (see comments on vma_merge),
631 * we must remove another next too. It would clutter
632 * up the code too much to do both in one go.
633 */
634 if (remove_next == 2) {
635 next = vma->vm_next;
636 goto again;
637 }
638 }
639
640 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800641
642 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643}
644
645/*
646 * If the vma has a ->close operation then the driver probably needs to release
647 * per-vma resources, so we don't attempt to merge those.
648 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649static inline int is_mergeable_vma(struct vm_area_struct *vma,
650 struct file *file, unsigned long vm_flags)
651{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700652 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
653 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 return 0;
655 if (vma->vm_file != file)
656 return 0;
657 if (vma->vm_ops && vma->vm_ops->close)
658 return 0;
659 return 1;
660}
661
662static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700663 struct anon_vma *anon_vma2,
664 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Shaohua Li965f55d2011-05-24 17:11:20 -0700666 /*
667 * The list_is_singular() test is to avoid merging VMA cloned from
668 * parents. This can improve scalability caused by anon_vma lock.
669 */
670 if ((!anon_vma1 || !anon_vma2) && (!vma ||
671 list_is_singular(&vma->anon_vma_chain)))
672 return 1;
673 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676/*
677 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
678 * in front of (at a lower virtual address and file offset than) the vma.
679 *
680 * We cannot merge two vmas if they have differently assigned (non-NULL)
681 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
682 *
683 * We don't check here for the merged mmap wrapping around the end of pagecache
684 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
685 * wrap, nor mmaps which cover the final page at index -1UL.
686 */
687static int
688can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
689 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
690{
691 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700692 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 if (vma->vm_pgoff == vm_pgoff)
694 return 1;
695 }
696 return 0;
697}
698
699/*
700 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
701 * beyond (at a higher virtual address and file offset than) the vma.
702 *
703 * We cannot merge two vmas if they have differently assigned (non-NULL)
704 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
705 */
706static int
707can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
708 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
709{
710 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700711 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 pgoff_t vm_pglen;
713 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
714 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
715 return 1;
716 }
717 return 0;
718}
719
720/*
721 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
722 * whether that can be merged with its predecessor or its successor.
723 * Or both (it neatly fills a hole).
724 *
725 * In most cases - when called for mmap, brk or mremap - [addr,end) is
726 * certain not to be mapped by the time vma_merge is called; but when
727 * called for mprotect, it is certain to be already mapped (either at
728 * an offset within prev, or at the start of next), and the flags of
729 * this area are about to be changed to vm_flags - and the no-change
730 * case has already been eliminated.
731 *
732 * The following mprotect cases have to be considered, where AAAA is
733 * the area passed down from mprotect_fixup, never extending beyond one
734 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
735 *
736 * AAAA AAAA AAAA AAAA
737 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
738 * cannot merge might become might become might become
739 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
740 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
741 * mremap move: PPPPNNNNNNNN 8
742 * AAAA
743 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
744 * might become case 1 below case 2 below case 3 below
745 *
746 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
747 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
748 */
749struct vm_area_struct *vma_merge(struct mm_struct *mm,
750 struct vm_area_struct *prev, unsigned long addr,
751 unsigned long end, unsigned long vm_flags,
752 struct anon_vma *anon_vma, struct file *file,
753 pgoff_t pgoff, struct mempolicy *policy)
754{
755 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
756 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800757 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
759 /*
760 * We later require that vma->vm_flags == vm_flags,
761 * so this tests vma->vm_flags & VM_SPECIAL, too.
762 */
763 if (vm_flags & VM_SPECIAL)
764 return NULL;
765
766 if (prev)
767 next = prev->vm_next;
768 else
769 next = mm->mmap;
770 area = next;
771 if (next && next->vm_end == end) /* cases 6, 7, 8 */
772 next = next->vm_next;
773
774 /*
775 * Can it merge with the predecessor?
776 */
777 if (prev && prev->vm_end == addr &&
778 mpol_equal(vma_policy(prev), policy) &&
779 can_vma_merge_after(prev, vm_flags,
780 anon_vma, file, pgoff)) {
781 /*
782 * OK, it can. Can we now merge in the successor as well?
783 */
784 if (next && end == next->vm_start &&
785 mpol_equal(policy, vma_policy(next)) &&
786 can_vma_merge_before(next, vm_flags,
787 anon_vma, file, pgoff+pglen) &&
788 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700789 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800791 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 next->vm_end, prev->vm_pgoff, NULL);
793 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800794 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800796 if (err)
797 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800798 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return prev;
800 }
801
802 /*
803 * Can this new request be merged in front of next?
804 */
805 if (next && end == next->vm_start &&
806 mpol_equal(policy, vma_policy(next)) &&
807 can_vma_merge_before(next, vm_flags,
808 anon_vma, file, pgoff+pglen)) {
809 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800810 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 addr, prev->vm_pgoff, NULL);
812 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800813 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800815 if (err)
816 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800817 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 return area;
819 }
820
821 return NULL;
822}
823
824/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700825 * Rough compatbility check to quickly see if it's even worth looking
826 * at sharing an anon_vma.
827 *
828 * They need to have the same vm_file, and the flags can only differ
829 * in things that mprotect may change.
830 *
831 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
832 * we can merge the two vma's. For example, we refuse to merge a vma if
833 * there is a vm_ops->close() function, because that indicates that the
834 * driver is doing some kind of reference counting. But that doesn't
835 * really matter for the anon_vma sharing case.
836 */
837static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
838{
839 return a->vm_end == b->vm_start &&
840 mpol_equal(vma_policy(a), vma_policy(b)) &&
841 a->vm_file == b->vm_file &&
842 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
843 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
844}
845
846/*
847 * Do some basic sanity checking to see if we can re-use the anon_vma
848 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
849 * the same as 'old', the other will be the new one that is trying
850 * to share the anon_vma.
851 *
852 * NOTE! This runs with mm_sem held for reading, so it is possible that
853 * the anon_vma of 'old' is concurrently in the process of being set up
854 * by another page fault trying to merge _that_. But that's ok: if it
855 * is being set up, that automatically means that it will be a singleton
856 * acceptable for merging, so we can do all of this optimistically. But
857 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
858 *
859 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
860 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
861 * is to return an anon_vma that is "complex" due to having gone through
862 * a fork).
863 *
864 * We also make sure that the two vma's are compatible (adjacent,
865 * and with the same memory policies). That's all stable, even with just
866 * a read lock on the mm_sem.
867 */
868static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
869{
870 if (anon_vma_compatible(a, b)) {
871 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
872
873 if (anon_vma && list_is_singular(&old->anon_vma_chain))
874 return anon_vma;
875 }
876 return NULL;
877}
878
879/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
881 * neighbouring vmas for a suitable anon_vma, before it goes off
882 * to allocate a new anon_vma. It checks because a repetitive
883 * sequence of mprotects and faults may otherwise lead to distinct
884 * anon_vmas being allocated, preventing vma merge in subsequent
885 * mprotect.
886 */
887struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
888{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700889 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 near = vma->vm_next;
893 if (!near)
894 goto try_prev;
895
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700896 anon_vma = reusable_anon_vma(near, vma, near);
897 if (anon_vma)
898 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -0700900 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 if (!near)
902 goto none;
903
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700904 anon_vma = reusable_anon_vma(near, near, vma);
905 if (anon_vma)
906 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907none:
908 /*
909 * There's no absolute need to look only at touching neighbours:
910 * we could search further afield for "compatible" anon_vmas.
911 * But it would probably just be a waste of time searching,
912 * or lead to too many vmas hanging off the same anon_vma.
913 * We're trying to allow mprotect remerging later on,
914 * not trying to minimize memory used for anon_vmas.
915 */
916 return NULL;
917}
918
919#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700920void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 struct file *file, long pages)
922{
923 const unsigned long stack_flags
924 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
925
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 if (file) {
927 mm->shared_vm += pages;
928 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
929 mm->exec_vm += pages;
930 } else if (flags & stack_flags)
931 mm->stack_vm += pages;
932 if (flags & (VM_RESERVED|VM_IO))
933 mm->reserved_vm += pages;
934}
935#endif /* CONFIG_PROC_FS */
936
937/*
Al Viro40401532012-02-13 03:58:52 +0000938 * If a hint addr is less than mmap_min_addr change hint to be as
939 * low as possible but still greater than mmap_min_addr
940 */
941static inline unsigned long round_hint_to_min(unsigned long hint)
942{
943 hint &= PAGE_MASK;
944 if (((void *)hint != NULL) &&
945 (hint < mmap_min_addr))
946 return PAGE_ALIGN(mmap_min_addr);
947 return hint;
948}
949
950/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700951 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 */
953
ZhenwenXu48aae422009-01-06 14:40:21 -0800954unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 unsigned long len, unsigned long prot,
956 unsigned long flags, unsigned long pgoff)
957{
958 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 struct inode *inode;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900960 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700962 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 /*
965 * Does the application expect PROT_READ to imply PROT_EXEC?
966 *
967 * (the exception is when the underlying filesystem is noexec
968 * mounted, in which case we dont add PROT_EXEC.)
969 */
970 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800971 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 prot |= PROT_EXEC;
973
974 if (!len)
975 return -EINVAL;
976
Eric Paris7cd94142007-11-26 18:47:40 -0500977 if (!(flags & MAP_FIXED))
978 addr = round_hint_to_min(addr);
979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 /* Careful about overflows.. */
981 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500982 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 return -ENOMEM;
984
985 /* offset overflow? */
986 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
987 return -EOVERFLOW;
988
989 /* Too many mappings? */
990 if (mm->map_count > sysctl_max_map_count)
991 return -ENOMEM;
992
993 /* Obtain the address to map to. we verify (or select) it and ensure
994 * that it represents a valid section of the address space.
995 */
996 addr = get_unmapped_area(file, addr, len, pgoff, flags);
997 if (addr & ~PAGE_MASK)
998 return addr;
999
1000 /* Do simple checking here so the lower-level routines won't have
1001 * to. we assume access permissions have been handled by the open
1002 * of the memory object, so we don't do any here.
1003 */
1004 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1005 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1006
Huang Shijiecdf7b342009-09-21 17:03:36 -07001007 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (!can_do_mlock())
1009 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 /* mlock MCL_FUTURE? */
1012 if (vm_flags & VM_LOCKED) {
1013 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001014 locked = len >> PAGE_SHIFT;
1015 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001016 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001017 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1019 return -EAGAIN;
1020 }
1021
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001022 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
1024 if (file) {
1025 switch (flags & MAP_TYPE) {
1026 case MAP_SHARED:
1027 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1028 return -EACCES;
1029
1030 /*
1031 * Make sure we don't allow writing to an append-only
1032 * file..
1033 */
1034 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1035 return -EACCES;
1036
1037 /*
1038 * Make sure there are no mandatory locks on the file.
1039 */
1040 if (locks_verify_locked(inode))
1041 return -EAGAIN;
1042
1043 vm_flags |= VM_SHARED | VM_MAYSHARE;
1044 if (!(file->f_mode & FMODE_WRITE))
1045 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1046
1047 /* fall through */
1048 case MAP_PRIVATE:
1049 if (!(file->f_mode & FMODE_READ))
1050 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001051 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001052 if (vm_flags & VM_EXEC)
1053 return -EPERM;
1054 vm_flags &= ~VM_MAYEXEC;
1055 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001056
1057 if (!file->f_op || !file->f_op->mmap)
1058 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 break;
1060
1061 default:
1062 return -EINVAL;
1063 }
1064 } else {
1065 switch (flags & MAP_TYPE) {
1066 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001067 /*
1068 * Ignore pgoff.
1069 */
1070 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 vm_flags |= VM_SHARED | VM_MAYSHARE;
1072 break;
1073 case MAP_PRIVATE:
1074 /*
1075 * Set pgoff according to addr for anon_vma.
1076 */
1077 pgoff = addr >> PAGE_SHIFT;
1078 break;
1079 default:
1080 return -EINVAL;
1081 }
1082 }
1083
Eric Parised032182007-06-28 15:55:21 -04001084 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 if (error)
1086 return error;
Eric Parised032182007-06-28 15:55:21 -04001087
Mel Gorman5a6fe122009-02-10 14:02:27 +00001088 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001089}
1090EXPORT_SYMBOL(do_mmap_pgoff);
1091
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001092SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1093 unsigned long, prot, unsigned long, flags,
1094 unsigned long, fd, unsigned long, pgoff)
1095{
1096 struct file *file = NULL;
1097 unsigned long retval = -EBADF;
1098
1099 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001100 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001101 if (unlikely(flags & MAP_HUGETLB))
1102 return -EINVAL;
1103 file = fget(fd);
1104 if (!file)
1105 goto out;
1106 } else if (flags & MAP_HUGETLB) {
1107 struct user_struct *user = NULL;
1108 /*
1109 * VM_NORESERVE is used because the reservations will be
1110 * taken when vm_ops->mmap() is called
1111 * A dummy user value is used because we are not locking
1112 * memory so no accounting is necessary
1113 */
Steven Truelove40716e22012-03-21 16:34:14 -07001114 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1115 VM_NORESERVE, &user,
1116 HUGETLB_ANONHUGE_INODE);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001117 if (IS_ERR(file))
1118 return PTR_ERR(file);
1119 }
1120
1121 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1122
1123 down_write(&current->mm->mmap_sem);
1124 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1125 up_write(&current->mm->mmap_sem);
1126
1127 if (file)
1128 fput(file);
1129out:
1130 return retval;
1131}
1132
Christoph Hellwiga4679372010-03-10 15:21:15 -08001133#ifdef __ARCH_WANT_SYS_OLD_MMAP
1134struct mmap_arg_struct {
1135 unsigned long addr;
1136 unsigned long len;
1137 unsigned long prot;
1138 unsigned long flags;
1139 unsigned long fd;
1140 unsigned long offset;
1141};
1142
1143SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1144{
1145 struct mmap_arg_struct a;
1146
1147 if (copy_from_user(&a, arg, sizeof(a)))
1148 return -EFAULT;
1149 if (a.offset & ~PAGE_MASK)
1150 return -EINVAL;
1151
1152 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1153 a.offset >> PAGE_SHIFT);
1154}
1155#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1156
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001157/*
1158 * Some shared mappigns will want the pages marked read-only
1159 * to track write events. If so, we'll downgrade vm_page_prot
1160 * to the private version (using protection_map[] without the
1161 * VM_SHARED bit).
1162 */
1163int vma_wants_writenotify(struct vm_area_struct *vma)
1164{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001165 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001166
1167 /* If it was private or non-writable, the write bit is already clear */
1168 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1169 return 0;
1170
1171 /* The backer wishes to know when pages are first written to? */
1172 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1173 return 1;
1174
1175 /* The open routine did something to the protections already? */
1176 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001177 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001178 return 0;
1179
1180 /* Specialty mapping? */
1181 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1182 return 0;
1183
1184 /* Can the mapping track the dirty pages? */
1185 return vma->vm_file && vma->vm_file->f_mapping &&
1186 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1187}
1188
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001189/*
1190 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001191 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001192 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001193static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001194{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001195 /*
1196 * hugetlb has its own accounting separate from the core VM
1197 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1198 */
1199 if (file && is_file_hugepages(file))
1200 return 0;
1201
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001202 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1203}
1204
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001205unsigned long mmap_region(struct file *file, unsigned long addr,
1206 unsigned long len, unsigned long flags,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001207 vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001208{
1209 struct mm_struct *mm = current->mm;
1210 struct vm_area_struct *vma, *prev;
1211 int correct_wcount = 0;
1212 int error;
1213 struct rb_node **rb_link, *rb_parent;
1214 unsigned long charged = 0;
1215 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 /* Clear old maps */
1218 error = -ENOMEM;
1219munmap_back:
1220 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1221 if (vma && vma->vm_start < addr + len) {
1222 if (do_munmap(mm, addr, len))
1223 return -ENOMEM;
1224 goto munmap_back;
1225 }
1226
1227 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001228 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 return -ENOMEM;
1230
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001231 /*
1232 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001233 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001234 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001235 if ((flags & MAP_NORESERVE)) {
1236 /* We honor MAP_NORESERVE if allowed to overcommit */
1237 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1238 vm_flags |= VM_NORESERVE;
1239
1240 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1241 if (file && is_file_hugepages(file))
1242 vm_flags |= VM_NORESERVE;
1243 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001244
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001245 /*
1246 * Private writable mapping: check memory availability
1247 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001248 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001249 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001250 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001251 return -ENOMEM;
1252 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 }
1254
1255 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001256 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001258 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1259 if (vma)
1260 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
1262 /*
1263 * Determine the object being mapped and call the appropriate
1264 * specific mapper. the address has already been validated, but
1265 * not unmapped, but the maps are removed from the list.
1266 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001267 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 if (!vma) {
1269 error = -ENOMEM;
1270 goto unacct_error;
1271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
1273 vma->vm_mm = mm;
1274 vma->vm_start = addr;
1275 vma->vm_end = addr + len;
1276 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001277 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001279 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
Hugh Dickinsce8fea72012-03-06 12:28:52 -08001281 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1285 goto free_vma;
1286 if (vm_flags & VM_DENYWRITE) {
1287 error = deny_write_access(file);
1288 if (error)
1289 goto free_vma;
1290 correct_wcount = 1;
1291 }
1292 vma->vm_file = file;
1293 get_file(file);
1294 error = file->f_op->mmap(file, vma);
1295 if (error)
1296 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001297 if (vm_flags & VM_EXECUTABLE)
1298 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001299
1300 /* Can addr have changed??
1301 *
1302 * Answer: Yes, several device drivers can do it in their
1303 * f_op->mmap method. -DaveM
1304 */
1305 addr = vma->vm_start;
1306 pgoff = vma->vm_pgoff;
1307 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 } else if (vm_flags & VM_SHARED) {
Al Viro835ee792012-03-05 06:39:47 +00001309 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1310 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 error = shmem_zero_setup(vma);
1312 if (error)
1313 goto free_vma;
1314 }
1315
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001316 if (vma_wants_writenotify(vma)) {
1317 pgprot_t pprot = vma->vm_page_prot;
1318
1319 /* Can vma->vm_page_prot have changed??
1320 *
1321 * Answer: Yes, drivers may have changed it in their
1322 * f_op->mmap method.
1323 *
1324 * Ensures that vmas marked as uncached stay that way.
1325 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001326 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001327 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1328 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1329 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001330
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001331 vma_link(mm, vma, prev, rb_link, rb_parent);
1332 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001333
1334 /* Once vma denies write, undo our temporary denial count */
1335 if (correct_wcount)
1336 atomic_inc(&inode->i_writecount);
1337out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001338 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001341 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001343 if (!mlock_vma_pages_range(vma, addr, addr + len))
1344 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001345 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001346 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 return addr;
1348
1349unmap_and_free_vma:
1350 if (correct_wcount)
1351 atomic_inc(&inode->i_writecount);
1352 vma->vm_file = NULL;
1353 fput(file);
1354
1355 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001356 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1357 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358free_vma:
1359 kmem_cache_free(vm_area_cachep, vma);
1360unacct_error:
1361 if (charged)
1362 vm_unacct_memory(charged);
1363 return error;
1364}
1365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366/* Get an address range which is currently unmapped.
1367 * For shmat() with addr=0.
1368 *
1369 * Ugly calling convention alert:
1370 * Return value with the low bits set means error value,
1371 * ie
1372 * if (ret & ~PAGE_MASK)
1373 * error = ret;
1374 *
1375 * This function "knows" that -ENOMEM has the bits set.
1376 */
1377#ifndef HAVE_ARCH_UNMAPPED_AREA
1378unsigned long
1379arch_get_unmapped_area(struct file *filp, unsigned long addr,
1380 unsigned long len, unsigned long pgoff, unsigned long flags)
1381{
1382 struct mm_struct *mm = current->mm;
1383 struct vm_area_struct *vma;
1384 unsigned long start_addr;
1385
1386 if (len > TASK_SIZE)
1387 return -ENOMEM;
1388
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001389 if (flags & MAP_FIXED)
1390 return addr;
1391
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 if (addr) {
1393 addr = PAGE_ALIGN(addr);
1394 vma = find_vma(mm, addr);
1395 if (TASK_SIZE - len >= addr &&
1396 (!vma || addr + len <= vma->vm_start))
1397 return addr;
1398 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001399 if (len > mm->cached_hole_size) {
1400 start_addr = addr = mm->free_area_cache;
1401 } else {
1402 start_addr = addr = TASK_UNMAPPED_BASE;
1403 mm->cached_hole_size = 0;
1404 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406full_search:
1407 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1408 /* At this point: (!vma || addr < vma->vm_end). */
1409 if (TASK_SIZE - len < addr) {
1410 /*
1411 * Start a new search - just in case we missed
1412 * some holes.
1413 */
1414 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001415 addr = TASK_UNMAPPED_BASE;
1416 start_addr = addr;
1417 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 goto full_search;
1419 }
1420 return -ENOMEM;
1421 }
1422 if (!vma || addr + len <= vma->vm_start) {
1423 /*
1424 * Remember the place where we stopped the search:
1425 */
1426 mm->free_area_cache = addr + len;
1427 return addr;
1428 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001429 if (addr + mm->cached_hole_size < vma->vm_start)
1430 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 addr = vma->vm_end;
1432 }
1433}
1434#endif
1435
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001436void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
1438 /*
1439 * Is this a new hole at the lowest possible address?
1440 */
Xiao Guangrongf44d2192012-03-21 16:33:56 -07001441 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001442 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
1445/*
1446 * This mmap-allocator allocates new areas top-down from below the
1447 * stack's low limit (the base):
1448 */
1449#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1450unsigned long
1451arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1452 const unsigned long len, const unsigned long pgoff,
1453 const unsigned long flags)
1454{
1455 struct vm_area_struct *vma;
1456 struct mm_struct *mm = current->mm;
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001457 unsigned long addr = addr0, start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 /* requested length too big for entire address space */
1460 if (len > TASK_SIZE)
1461 return -ENOMEM;
1462
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001463 if (flags & MAP_FIXED)
1464 return addr;
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 /* requesting a specific address */
1467 if (addr) {
1468 addr = PAGE_ALIGN(addr);
1469 vma = find_vma(mm, addr);
1470 if (TASK_SIZE - len >= addr &&
1471 (!vma || addr + len <= vma->vm_start))
1472 return addr;
1473 }
1474
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001475 /* check if free_area_cache is useful for us */
1476 if (len <= mm->cached_hole_size) {
1477 mm->cached_hole_size = 0;
1478 mm->free_area_cache = mm->mmap_base;
1479 }
1480
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001481try_again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 /* either no address requested or can't fit in requested address hole */
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001483 start_addr = addr = mm->free_area_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001485 if (addr < len)
1486 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001488 addr -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 do {
1490 /*
1491 * Lookup failure means no vma is above this address,
1492 * else if new region fits below vma->vm_start,
1493 * return with success:
1494 */
1495 vma = find_vma(mm, addr);
1496 if (!vma || addr+len <= vma->vm_start)
1497 /* remember the address as a hint for next time */
1498 return (mm->free_area_cache = addr);
1499
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001500 /* remember the largest hole we saw so far */
1501 if (addr + mm->cached_hole_size < vma->vm_start)
1502 mm->cached_hole_size = vma->vm_start - addr;
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 /* try just below the current vma->vm_start */
1505 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001506 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001508fail:
1509 /*
1510 * if hint left us with no space for the requested
1511 * mapping then try again:
1512 *
1513 * Note: this is different with the case of bottomup
1514 * which does the fully line-search, but we use find_vma
1515 * here that causes some holes skipped.
1516 */
1517 if (start_addr != mm->mmap_base) {
1518 mm->free_area_cache = mm->mmap_base;
1519 mm->cached_hole_size = 0;
1520 goto try_again;
1521 }
1522
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 /*
1524 * A failed mmap() very likely causes application failure,
1525 * so fall back to the bottom-up function here. This scenario
1526 * can happen with large stack limits and large mmap()
1527 * allocations.
1528 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001529 mm->cached_hole_size = ~0UL;
1530 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1532 /*
1533 * Restore the topdown base:
1534 */
1535 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001536 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
1538 return addr;
1539}
1540#endif
1541
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001542void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543{
1544 /*
1545 * Is this a new hole at the highest possible address?
1546 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001547 if (addr > mm->free_area_cache)
1548 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001551 if (mm->free_area_cache > mm->mmap_base)
1552 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553}
1554
1555unsigned long
1556get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1557 unsigned long pgoff, unsigned long flags)
1558{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001559 unsigned long (*get_area)(struct file *, unsigned long,
1560 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
Al Viro9206de92009-12-03 15:23:11 -05001562 unsigned long error = arch_mmap_check(addr, len, flags);
1563 if (error)
1564 return error;
1565
1566 /* Careful about overflows.. */
1567 if (len > TASK_SIZE)
1568 return -ENOMEM;
1569
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001570 get_area = current->mm->get_unmapped_area;
1571 if (file && file->f_op && file->f_op->get_unmapped_area)
1572 get_area = file->f_op->get_unmapped_area;
1573 addr = get_area(file, addr, len, pgoff, flags);
1574 if (IS_ERR_VALUE(addr))
1575 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001577 if (addr > TASK_SIZE - len)
1578 return -ENOMEM;
1579 if (addr & ~PAGE_MASK)
1580 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001581
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001582 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583}
1584
1585EXPORT_SYMBOL(get_unmapped_area);
1586
1587/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001588struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589{
1590 struct vm_area_struct *vma = NULL;
1591
1592 if (mm) {
1593 /* Check the cache first. */
1594 /* (Cache hit rate is typically around 35%.) */
1595 vma = mm->mmap_cache;
1596 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1597 struct rb_node * rb_node;
1598
1599 rb_node = mm->mm_rb.rb_node;
1600 vma = NULL;
1601
1602 while (rb_node) {
1603 struct vm_area_struct * vma_tmp;
1604
1605 vma_tmp = rb_entry(rb_node,
1606 struct vm_area_struct, vm_rb);
1607
1608 if (vma_tmp->vm_end > addr) {
1609 vma = vma_tmp;
1610 if (vma_tmp->vm_start <= addr)
1611 break;
1612 rb_node = rb_node->rb_left;
1613 } else
1614 rb_node = rb_node->rb_right;
1615 }
1616 if (vma)
1617 mm->mmap_cache = vma;
1618 }
1619 }
1620 return vma;
1621}
1622
1623EXPORT_SYMBOL(find_vma);
1624
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001625/*
1626 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001627 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628struct vm_area_struct *
1629find_vma_prev(struct mm_struct *mm, unsigned long addr,
1630 struct vm_area_struct **pprev)
1631{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001632 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001634 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05001635 if (vma) {
1636 *pprev = vma->vm_prev;
1637 } else {
1638 struct rb_node *rb_node = mm->mm_rb.rb_node;
1639 *pprev = NULL;
1640 while (rb_node) {
1641 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1642 rb_node = rb_node->rb_right;
1643 }
1644 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001645 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646}
1647
1648/*
1649 * Verify that the stack growth is acceptable and
1650 * update accounting. This is shared with both the
1651 * grow-up and grow-down cases.
1652 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001653static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 struct mm_struct *mm = vma->vm_mm;
1656 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001657 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001660 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return -ENOMEM;
1662
1663 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001664 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 return -ENOMEM;
1666
1667 /* mlock limit tests */
1668 if (vma->vm_flags & VM_LOCKED) {
1669 unsigned long locked;
1670 unsigned long limit;
1671 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001672 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1673 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 if (locked > limit && !capable(CAP_IPC_LOCK))
1675 return -ENOMEM;
1676 }
1677
Adam Litke0d59a012007-01-30 14:35:39 -08001678 /* Check to ensure the stack will not grow into a hugetlb-only region */
1679 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1680 vma->vm_end - size;
1681 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1682 return -EFAULT;
1683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 /*
1685 * Overcommit.. This must be the final test, as it will
1686 * update security statistics.
1687 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001688 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 return -ENOMEM;
1690
1691 /* Ok, everything looks good - let it rip */
1692 mm->total_vm += grow;
1693 if (vma->vm_flags & VM_LOCKED)
1694 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001695 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 return 0;
1697}
1698
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001699#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001701 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1702 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001704int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
1706 int error;
1707
1708 if (!(vma->vm_flags & VM_GROWSUP))
1709 return -EFAULT;
1710
1711 /*
1712 * We must make sure the anon_vma is allocated
1713 * so that the anon_vma locking is not a noop.
1714 */
1715 if (unlikely(anon_vma_prepare(vma)))
1716 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001717 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 /*
1720 * vma->vm_start/vm_end cannot change under us because the caller
1721 * is required to hold the mmap_sem in read mode. We need the
1722 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001723 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 */
Helge Deller06b32f32006-12-19 19:28:33 +01001725 if (address < PAGE_ALIGN(address+4))
1726 address = PAGE_ALIGN(address+4);
1727 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001728 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001729 return -ENOMEM;
1730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 error = 0;
1732
1733 /* Somebody else might have raced and expanded it already */
1734 if (address > vma->vm_end) {
1735 unsigned long size, grow;
1736
1737 size = address - vma->vm_start;
1738 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1739
Hugh Dickins42c36f62011-05-09 17:44:42 -07001740 error = -ENOMEM;
1741 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1742 error = acct_stack_growth(vma, size, grow);
1743 if (!error) {
1744 vma->vm_end = address;
1745 perf_event_mmap(vma);
1746 }
Eric B Munson3af9e852010-05-18 15:30:49 +01001747 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001749 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001750 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 return error;
1752}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001753#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755/*
1756 * vma is the first one with address < vma->vm_start. Have to extend vma.
1757 */
Michal Hockod05f3162011-05-24 17:11:44 -07001758int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001759 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
1761 int error;
1762
1763 /*
1764 * We must make sure the anon_vma is allocated
1765 * so that the anon_vma locking is not a noop.
1766 */
1767 if (unlikely(anon_vma_prepare(vma)))
1768 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001769
1770 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001771 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001772 if (error)
1773 return error;
1774
Rik van Rielbb4a3402010-08-09 17:18:37 -07001775 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776
1777 /*
1778 * vma->vm_start/vm_end cannot change under us because the caller
1779 * is required to hold the mmap_sem in read mode. We need the
1780 * anon_vma lock to serialize against concurrent expand_stacks.
1781 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
1783 /* Somebody else might have raced and expanded it already */
1784 if (address < vma->vm_start) {
1785 unsigned long size, grow;
1786
1787 size = vma->vm_end - address;
1788 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1789
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001790 error = -ENOMEM;
1791 if (grow <= vma->vm_pgoff) {
1792 error = acct_stack_growth(vma, size, grow);
1793 if (!error) {
1794 vma->vm_start = address;
1795 vma->vm_pgoff -= grow;
1796 perf_event_mmap(vma);
1797 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 }
1799 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001800 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001801 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 return error;
1803}
1804
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001805#ifdef CONFIG_STACK_GROWSUP
1806int expand_stack(struct vm_area_struct *vma, unsigned long address)
1807{
1808 return expand_upwards(vma, address);
1809}
1810
1811struct vm_area_struct *
1812find_extend_vma(struct mm_struct *mm, unsigned long addr)
1813{
1814 struct vm_area_struct *vma, *prev;
1815
1816 addr &= PAGE_MASK;
1817 vma = find_vma_prev(mm, addr, &prev);
1818 if (vma && (vma->vm_start <= addr))
1819 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001820 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001821 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001822 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001823 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001824 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001825 return prev;
1826}
1827#else
1828int expand_stack(struct vm_area_struct *vma, unsigned long address)
1829{
1830 return expand_downwards(vma, address);
1831}
1832
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833struct vm_area_struct *
1834find_extend_vma(struct mm_struct * mm, unsigned long addr)
1835{
1836 struct vm_area_struct * vma;
1837 unsigned long start;
1838
1839 addr &= PAGE_MASK;
1840 vma = find_vma(mm,addr);
1841 if (!vma)
1842 return NULL;
1843 if (vma->vm_start <= addr)
1844 return vma;
1845 if (!(vma->vm_flags & VM_GROWSDOWN))
1846 return NULL;
1847 start = vma->vm_start;
1848 if (expand_stack(vma, addr))
1849 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001850 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001851 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 return vma;
1854}
1855#endif
1856
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001858 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001860 *
1861 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001863static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001865 /* Update high watermark before we lower total_vm */
1866 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001868 long nrpages = vma_pages(vma);
1869
1870 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001871 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001872 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001873 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 validate_mm(mm);
1875}
1876
1877/*
1878 * Get rid of page table information in the indicated region.
1879 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001880 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 */
1882static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001883 struct vm_area_struct *vma, struct vm_area_struct *prev,
1884 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001886 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001887 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 unsigned long nr_accounted = 0;
1889
1890 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001891 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001892 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001893 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 vm_unacct_memory(nr_accounted);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001895 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1896 next ? next->vm_start : 0);
1897 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
1899
1900/*
1901 * Create a list of vma's touched by the unmap, removing them from the mm's
1902 * vma list as we go..
1903 */
1904static void
1905detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1906 struct vm_area_struct *prev, unsigned long end)
1907{
1908 struct vm_area_struct **insertion_point;
1909 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001910 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001913 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 do {
1915 rb_erase(&vma->vm_rb, &mm->mm_rb);
1916 mm->map_count--;
1917 tail_vma = vma;
1918 vma = vma->vm_next;
1919 } while (vma && vma->vm_start < end);
1920 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001921 if (vma)
1922 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001924 if (mm->unmap_area == arch_unmap_area)
1925 addr = prev ? prev->vm_end : mm->mmap_base;
1926 else
1927 addr = vma ? vma->vm_start : mm->mmap_base;
1928 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 mm->mmap_cache = NULL; /* Kill the cache. */
1930}
1931
1932/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001933 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1934 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001936static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 unsigned long addr, int new_below)
1938{
1939 struct mempolicy *pol;
1940 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001941 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Andi Kleena5516432008-07-23 21:27:41 -07001943 if (is_vm_hugetlb_page(vma) && (addr &
1944 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 return -EINVAL;
1946
Christoph Lametere94b1762006-12-06 20:33:17 -08001947 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001949 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951 /* most fields are the same, copy all, and then fixup */
1952 *new = *vma;
1953
Rik van Riel5beb4932010-03-05 13:42:07 -08001954 INIT_LIST_HEAD(&new->anon_vma_chain);
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 if (new_below)
1957 new->vm_end = addr;
1958 else {
1959 new->vm_start = addr;
1960 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1961 }
1962
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001963 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001965 err = PTR_ERR(pol);
1966 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 }
1968 vma_set_policy(new, pol);
1969
Rik van Riel5beb4932010-03-05 13:42:07 -08001970 if (anon_vma_clone(new, vma))
1971 goto out_free_mpol;
1972
Matt Helsley925d1c42008-04-29 01:01:36 -07001973 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001975 if (vma->vm_flags & VM_EXECUTABLE)
1976 added_exe_file_vma(mm);
1977 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979 if (new->vm_ops && new->vm_ops->open)
1980 new->vm_ops->open(new);
1981
1982 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08001983 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1985 else
Rik van Riel5beb4932010-03-05 13:42:07 -08001986 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Rik van Riel5beb4932010-03-05 13:42:07 -08001988 /* Success. */
1989 if (!err)
1990 return 0;
1991
1992 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04001993 if (new->vm_ops && new->vm_ops->close)
1994 new->vm_ops->close(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08001995 if (new->vm_file) {
1996 if (vma->vm_flags & VM_EXECUTABLE)
1997 removed_exe_file_vma(mm);
1998 fput(new->vm_file);
1999 }
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002000 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002001 out_free_mpol:
2002 mpol_put(pol);
2003 out_free_vma:
2004 kmem_cache_free(vm_area_cachep, new);
2005 out_err:
2006 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007}
2008
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002009/*
2010 * Split a vma into two pieces at address 'addr', a new vma is allocated
2011 * either for the first part or the tail.
2012 */
2013int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2014 unsigned long addr, int new_below)
2015{
2016 if (mm->map_count >= sysctl_max_map_count)
2017 return -ENOMEM;
2018
2019 return __split_vma(mm, vma, addr, new_below);
2020}
2021
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022/* Munmap is split into 2 main parts -- this part which finds
2023 * what needs doing, and the areas themselves, which do the
2024 * work. This now handles partial unmappings.
2025 * Jeremy Fitzhardinge <jeremy@goop.org>
2026 */
2027int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2028{
2029 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002030 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
2032 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2033 return -EINVAL;
2034
2035 if ((len = PAGE_ALIGN(len)) == 0)
2036 return -EINVAL;
2037
2038 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002039 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002040 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002042 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002043 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 /* if it doesn't overlap, we have nothing.. */
2046 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002047 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return 0;
2049
2050 /*
2051 * If we need to split any vma, do it now to save pain later.
2052 *
2053 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2054 * unmapped vm_area_struct will remain in use: so lower split_vma
2055 * places tmp vma above, and higher split_vma places tmp vma below.
2056 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002057 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002058 int error;
2059
2060 /*
2061 * Make sure that map_count on return from munmap() will
2062 * not exceed its limit; but let map_count go just above
2063 * its limit temporarily, to help free resources as expected.
2064 */
2065 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2066 return -ENOMEM;
2067
2068 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (error)
2070 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002071 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073
2074 /* Does it split the last one? */
2075 last = find_vma(mm, end);
2076 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002077 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (error)
2079 return error;
2080 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002081 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
2083 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002084 * unlock any mlock()ed ranges before detaching vmas
2085 */
2086 if (mm->locked_vm) {
2087 struct vm_area_struct *tmp = vma;
2088 while (tmp && tmp->vm_start < end) {
2089 if (tmp->vm_flags & VM_LOCKED) {
2090 mm->locked_vm -= vma_pages(tmp);
2091 munlock_vma_pages_all(tmp);
2092 }
2093 tmp = tmp->vm_next;
2094 }
2095 }
2096
2097 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 * Remove the vma's, and unmap the actual pages
2099 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002100 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2101 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
2103 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002104 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106 return 0;
2107}
2108
2109EXPORT_SYMBOL(do_munmap);
2110
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002111SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112{
2113 int ret;
2114 struct mm_struct *mm = current->mm;
2115
2116 profile_munmap(addr);
2117
2118 down_write(&mm->mmap_sem);
2119 ret = do_munmap(mm, addr, len);
2120 up_write(&mm->mmap_sem);
2121 return ret;
2122}
2123
2124static inline void verify_mm_writelocked(struct mm_struct *mm)
2125{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002126#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2128 WARN_ON(1);
2129 up_read(&mm->mmap_sem);
2130 }
2131#endif
2132}
2133
2134/*
2135 * this is really a simplified "do_mmap". it only handles
2136 * anonymous maps. eventually we may be able to do some
2137 * brk-specific accounting here.
2138 */
2139unsigned long do_brk(unsigned long addr, unsigned long len)
2140{
2141 struct mm_struct * mm = current->mm;
2142 struct vm_area_struct * vma, * prev;
2143 unsigned long flags;
2144 struct rb_node ** rb_link, * rb_parent;
2145 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002146 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
2148 len = PAGE_ALIGN(len);
2149 if (!len)
2150 return addr;
2151
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002152 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002153 if (error)
2154 return error;
2155
Kirill Korotaev3a459752006-09-07 14:17:04 +04002156 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2157
Al Viro2c6a1012009-12-03 19:40:46 -05002158 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2159 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002160 return error;
2161
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 /*
2163 * mlock MCL_FUTURE?
2164 */
2165 if (mm->def_flags & VM_LOCKED) {
2166 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002167 locked = len >> PAGE_SHIFT;
2168 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002169 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002170 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2172 return -EAGAIN;
2173 }
2174
2175 /*
2176 * mm->mmap_sem is required to protect against another thread
2177 * changing the mappings in case we sleep.
2178 */
2179 verify_mm_writelocked(mm);
2180
2181 /*
2182 * Clear old maps. this also does some error checking for us
2183 */
2184 munmap_back:
2185 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2186 if (vma && vma->vm_start < addr + len) {
2187 if (do_munmap(mm, addr, len))
2188 return -ENOMEM;
2189 goto munmap_back;
2190 }
2191
2192 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002193 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 return -ENOMEM;
2195
2196 if (mm->map_count > sysctl_max_map_count)
2197 return -ENOMEM;
2198
Al Viro191c5422012-02-13 03:58:52 +00002199 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 return -ENOMEM;
2201
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002203 vma = vma_merge(mm, prev, addr, addr + len, flags,
2204 NULL, NULL, pgoff, NULL);
2205 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 goto out;
2207
2208 /*
2209 * create a vma struct for an anonymous mapping
2210 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002211 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 if (!vma) {
2213 vm_unacct_memory(len >> PAGE_SHIFT);
2214 return -ENOMEM;
2215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216
Rik van Riel5beb4932010-03-05 13:42:07 -08002217 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 vma->vm_mm = mm;
2219 vma->vm_start = addr;
2220 vma->vm_end = addr + len;
2221 vma->vm_pgoff = pgoff;
2222 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002223 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 vma_link(mm, vma, prev, rb_link, rb_parent);
2225out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002226 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 mm->total_vm += len >> PAGE_SHIFT;
2228 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002229 if (!mlock_vma_pages_range(vma, addr, addr + len))
2230 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 }
2232 return addr;
2233}
2234
2235EXPORT_SYMBOL(do_brk);
2236
2237/* Release all mmaps. */
2238void exit_mmap(struct mm_struct *mm)
2239{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002240 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002241 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 unsigned long nr_accounted = 0;
2243
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002244 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002245 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002246
Rik van Rielba470de2008-10-18 20:26:50 -07002247 if (mm->locked_vm) {
2248 vma = mm->mmap;
2249 while (vma) {
2250 if (vma->vm_flags & VM_LOCKED)
2251 munlock_vma_pages_all(vma);
2252 vma = vma->vm_next;
2253 }
2254 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002255
2256 arch_exit_mmap(mm);
2257
Rik van Rielba470de2008-10-18 20:26:50 -07002258 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002259 if (!vma) /* Can happen if dup_mmap() received an OOM */
2260 return;
2261
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 flush_cache_mm(mm);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002264 tlb_gather_mmu(&tlb, mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002265 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002266 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Al Viro6e8bb012012-03-05 13:41:15 -05002267 unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 vm_unacct_memory(nr_accounted);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002269
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002270 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Al Viro853f5e22012-03-05 14:03:47 -05002271 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002274 * Walk the list again, actually closing and freeing it,
2275 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002277 while (vma)
2278 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002279
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002280 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281}
2282
2283/* Insert vm structure into process list sorted by address
2284 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002285 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 */
2287int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2288{
2289 struct vm_area_struct * __vma, * prev;
2290 struct rb_node ** rb_link, * rb_parent;
2291
2292 /*
2293 * The vm_pgoff of a purely anonymous vma should be irrelevant
2294 * until its first write fault, when page's anon_vma and index
2295 * are set. But now set the vm_pgoff it will almost certainly
2296 * end up with (unless mremap moves it elsewhere before that
2297 * first wfault), so /proc/pid/maps tells a consistent story.
2298 *
2299 * By setting it to reflect the virtual start address of the
2300 * vma, merges and splits can happen in a seamless way, just
2301 * using the existing file pgoff checks and manipulations.
2302 * Similarly in do_mmap_pgoff and in do_brk.
2303 */
2304 if (!vma->vm_file) {
2305 BUG_ON(vma->anon_vma);
2306 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2307 }
2308 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2309 if (__vma && __vma->vm_start < vma->vm_end)
2310 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002311 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002312 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002313 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 vma_link(mm, vma, prev, rb_link, rb_parent);
2315 return 0;
2316}
2317
2318/*
2319 * Copy the vma structure to a new location in the same mm,
2320 * prior to moving page table entries, to effect an mremap move.
2321 */
2322struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2323 unsigned long addr, unsigned long len, pgoff_t pgoff)
2324{
2325 struct vm_area_struct *vma = *vmap;
2326 unsigned long vma_start = vma->vm_start;
2327 struct mm_struct *mm = vma->vm_mm;
2328 struct vm_area_struct *new_vma, *prev;
2329 struct rb_node **rb_link, *rb_parent;
2330 struct mempolicy *pol;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002331 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
2333 /*
2334 * If anonymous vma has not yet been faulted, update new pgoff
2335 * to match new location, to increase its chance of merging.
2336 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002337 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002339 faulted_in_anon_vma = false;
2340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
2342 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2343 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2344 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2345 if (new_vma) {
2346 /*
2347 * Source vma may have been merged into new_vma
2348 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002349 if (unlikely(vma_start >= new_vma->vm_start &&
2350 vma_start < new_vma->vm_end)) {
2351 /*
2352 * The only way we can get a vma_merge with
2353 * self during an mremap is if the vma hasn't
2354 * been faulted in yet and we were allowed to
2355 * reset the dst vma->vm_pgoff to the
2356 * destination address of the mremap to allow
2357 * the merge to happen. mremap must change the
2358 * vm_pgoff linearity between src and dst vmas
2359 * (in turn preventing a vma_merge) to be
2360 * safe. It is only safe to keep the vm_pgoff
2361 * linear if there are no pages mapped yet.
2362 */
2363 VM_BUG_ON(faulted_in_anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 *vmap = new_vma;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002365 } else
2366 anon_vma_moveto_tail(new_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002368 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 if (new_vma) {
2370 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002371 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002372 if (IS_ERR(pol))
2373 goto out_free_vma;
2374 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2375 if (anon_vma_clone(new_vma, vma))
2376 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 vma_set_policy(new_vma, pol);
2378 new_vma->vm_start = addr;
2379 new_vma->vm_end = addr + len;
2380 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002381 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002383 if (vma->vm_flags & VM_EXECUTABLE)
2384 added_exe_file_vma(mm);
2385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 if (new_vma->vm_ops && new_vma->vm_ops->open)
2387 new_vma->vm_ops->open(new_vma);
2388 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2389 }
2390 }
2391 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002392
2393 out_free_mempol:
2394 mpol_put(pol);
2395 out_free_vma:
2396 kmem_cache_free(vm_area_cachep, new_vma);
2397 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002399
2400/*
2401 * Return true if the calling process may expand its vm space by the passed
2402 * number of pages
2403 */
2404int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2405{
2406 unsigned long cur = mm->total_vm; /* pages */
2407 unsigned long lim;
2408
Jiri Slaby59e99e52010-03-05 13:41:44 -08002409 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002410
2411 if (cur + npages > lim)
2412 return 0;
2413 return 1;
2414}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002415
2416
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002417static int special_mapping_fault(struct vm_area_struct *vma,
2418 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002419{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002420 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002421 struct page **pages;
2422
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002423 /*
2424 * special mappings have no vm_file, and in that case, the mm
2425 * uses vm_pgoff internally. So we have to subtract it from here.
2426 * We are allowed to do this because we are the mm; do not copy
2427 * this code into drivers!
2428 */
2429 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002430
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002431 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2432 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002433
2434 if (*pages) {
2435 struct page *page = *pages;
2436 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002437 vmf->page = page;
2438 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002439 }
2440
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002441 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002442}
2443
2444/*
2445 * Having a close hook prevents vma merging regardless of flags.
2446 */
2447static void special_mapping_close(struct vm_area_struct *vma)
2448{
2449}
2450
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002451static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002452 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002453 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002454};
2455
2456/*
2457 * Called with mm->mmap_sem held for writing.
2458 * Insert a new vma covering the given region, with the given flags.
2459 * Its pages are supplied by the given array of struct page *.
2460 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2461 * The region past the last page supplied will always produce SIGBUS.
2462 * The array pointer and the pages it points to are assumed to stay alive
2463 * for as long as this mapping might exist.
2464 */
2465int install_special_mapping(struct mm_struct *mm,
2466 unsigned long addr, unsigned long len,
2467 unsigned long vm_flags, struct page **pages)
2468{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002469 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002470 struct vm_area_struct *vma;
2471
2472 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2473 if (unlikely(vma == NULL))
2474 return -ENOMEM;
2475
Rik van Riel5beb4932010-03-05 13:42:07 -08002476 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002477 vma->vm_mm = mm;
2478 vma->vm_start = addr;
2479 vma->vm_end = addr + len;
2480
Nick Piggin2f987352008-02-02 03:08:53 +01002481 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002482 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002483
2484 vma->vm_ops = &special_mapping_vmops;
2485 vma->vm_private_data = pages;
2486
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002487 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2488 if (ret)
2489 goto out;
2490
2491 ret = insert_vm_struct(mm, vma);
2492 if (ret)
2493 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002494
2495 mm->total_vm += len >> PAGE_SHIFT;
2496
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002497 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002498
Roland McGrathfa5dc222007-02-08 14:20:41 -08002499 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002500
2501out:
2502 kmem_cache_free(vm_area_cachep, vma);
2503 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002504}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002505
2506static DEFINE_MUTEX(mm_all_locks_mutex);
2507
Peter Zijlstra454ed842008-08-11 09:30:25 +02002508static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002509{
Rik van Riel012f18002010-08-09 17:18:40 -07002510 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002511 /*
2512 * The LSB of head.next can't change from under us
2513 * because we hold the mm_all_locks_mutex.
2514 */
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002515 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002516 /*
2517 * We can safely modify head.next after taking the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002518 * anon_vma->root->mutex. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002519 * the same anon_vma we won't take it again.
2520 *
2521 * No need of atomic instructions here, head.next
2522 * can't change from under us thanks to the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002523 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002524 */
2525 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002526 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002527 BUG();
2528 }
2529}
2530
Peter Zijlstra454ed842008-08-11 09:30:25 +02002531static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002532{
2533 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2534 /*
2535 * AS_MM_ALL_LOCKS can't change from under us because
2536 * we hold the mm_all_locks_mutex.
2537 *
2538 * Operations on ->flags have to be atomic because
2539 * even if AS_MM_ALL_LOCKS is stable thanks to the
2540 * mm_all_locks_mutex, there may be other cpus
2541 * changing other bitflags in parallel to us.
2542 */
2543 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2544 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002545 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002546 }
2547}
2548
2549/*
2550 * This operation locks against the VM for all pte/vma/mm related
2551 * operations that could ever happen on a certain mm. This includes
2552 * vmtruncate, try_to_unmap, and all page faults.
2553 *
2554 * The caller must take the mmap_sem in write mode before calling
2555 * mm_take_all_locks(). The caller isn't allowed to release the
2556 * mmap_sem until mm_drop_all_locks() returns.
2557 *
2558 * mmap_sem in write mode is required in order to block all operations
2559 * that could modify pagetables and free pages without need of
2560 * altering the vma layout (for example populate_range() with
2561 * nonlinear vmas). It's also needed in write mode to avoid new
2562 * anon_vmas to be associated with existing vmas.
2563 *
2564 * A single task can't take more than one mm_take_all_locks() in a row
2565 * or it would deadlock.
2566 *
2567 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2568 * mapping->flags avoid to take the same lock twice, if more than one
2569 * vma in this mm is backed by the same anon_vma or address_space.
2570 *
2571 * We can take all the locks in random order because the VM code
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002572 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002573 * takes more than one of them in a row. Secondly we're protected
2574 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2575 *
2576 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2577 * that may have to take thousand of locks.
2578 *
2579 * mm_take_all_locks() can fail if it's interrupted by signals.
2580 */
2581int mm_take_all_locks(struct mm_struct *mm)
2582{
2583 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002584 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002585
2586 BUG_ON(down_read_trylock(&mm->mmap_sem));
2587
2588 mutex_lock(&mm_all_locks_mutex);
2589
2590 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2591 if (signal_pending(current))
2592 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002593 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002594 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002595 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002596
2597 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2598 if (signal_pending(current))
2599 goto out_unlock;
2600 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002601 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2602 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002603 }
2604
Kautuk Consul584cff52011-10-31 17:08:59 -07002605 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002606
2607out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07002608 mm_drop_all_locks(mm);
2609 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002610}
2611
2612static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2613{
Rik van Riel012f18002010-08-09 17:18:40 -07002614 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002615 /*
2616 * The LSB of head.next can't change to 0 from under
2617 * us because we hold the mm_all_locks_mutex.
2618 *
2619 * We must however clear the bitflag before unlocking
2620 * the vma so the users using the anon_vma->head will
2621 * never see our bitflag.
2622 *
2623 * No need of atomic instructions here, head.next
2624 * can't change from under us until we release the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002625 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002626 */
2627 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002628 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002629 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002630 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002631 }
2632}
2633
2634static void vm_unlock_mapping(struct address_space *mapping)
2635{
2636 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2637 /*
2638 * AS_MM_ALL_LOCKS can't change to 0 from under us
2639 * because we hold the mm_all_locks_mutex.
2640 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002641 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002642 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2643 &mapping->flags))
2644 BUG();
2645 }
2646}
2647
2648/*
2649 * The mmap_sem cannot be released by the caller until
2650 * mm_drop_all_locks() returns.
2651 */
2652void mm_drop_all_locks(struct mm_struct *mm)
2653{
2654 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002655 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002656
2657 BUG_ON(down_read_trylock(&mm->mmap_sem));
2658 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2659
2660 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2661 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002662 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2663 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002664 if (vma->vm_file && vma->vm_file->f_mapping)
2665 vm_unlock_mapping(vma->vm_file->f_mapping);
2666 }
2667
2668 mutex_unlock(&mm_all_locks_mutex);
2669}
David Howells8feae132009-01-08 12:04:47 +00002670
2671/*
2672 * initialise the VMA slab
2673 */
2674void __init mmap_init(void)
2675{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002676 int ret;
2677
2678 ret = percpu_counter_init(&vm_committed_as, 0);
2679 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002680}