blob: 00161a48a45100c611ebaa053f0b1f1486d09f29 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
Alan Cox046c6882009-01-05 14:06:29 +00006 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 */
8
9#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070010#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/mm.h>
12#include <linux/shm.h>
13#include <linux/mman.h>
14#include <linux/pagemap.h>
15#include <linux/swap.h>
16#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/init.h>
19#include <linux/file.h>
20#include <linux/fs.h>
21#include <linux/personality.h>
22#include <linux/security.h>
23#include <linux/hugetlb.h>
24#include <linux/profile.h>
25#include <linux/module.h>
26#include <linux/mount.h>
27#include <linux/mempolicy.h>
28#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070029#include <linux/mmu_notifier.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020030#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/uaccess.h>
33#include <asm/cacheflush.h>
34#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020035#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Jan Beulich42b77722008-07-23 21:27:10 -070037#include "internal.h"
38
Kirill Korotaev3a459752006-09-07 14:17:04 +040039#ifndef arch_mmap_check
40#define arch_mmap_check(addr, len, flags) (0)
41#endif
42
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080043#ifndef arch_rebalance_pgtables
44#define arch_rebalance_pgtables(addr, len) (addr)
45#endif
46
Hugh Dickinse0da3822005-04-19 13:29:15 -070047static void unmap_region(struct mm_struct *mm,
48 struct vm_area_struct *vma, struct vm_area_struct *prev,
49 unsigned long start, unsigned long end);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * WARNING: the debugging will use recursive algorithms so never enable this
53 * unless you know what you are doing.
54 */
55#undef DEBUG_MM_RB
56
57/* description of effects of mapping type and prot in current implementation.
58 * this is due to the limited x86 page protection hardware. The expected
59 * behavior is in parens:
60 *
61 * map_type prot
62 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
63 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
64 * w: (no) no w: (no) no w: (yes) yes w: (no) no
65 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
66 *
67 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
68 * w: (no) no w: (no) no w: (copy) copy w: (no) no
69 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
70 *
71 */
72pgprot_t protection_map[16] = {
73 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
74 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
75};
76
Hugh Dickins804af2c2006-07-26 21:39:49 +010077pgprot_t vm_get_page_prot(unsigned long vm_flags)
78{
Dave Kleikampb845f312008-07-08 00:28:51 +100079 return __pgprot(pgprot_val(protection_map[vm_flags &
80 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
81 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010082}
83EXPORT_SYMBOL(vm_get_page_prot);
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
86int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070087int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070088struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90/*
91 * Check that a process has enough memory to allocate a new virtual
92 * mapping. 0 means there is enough memory for the allocation to
93 * succeed and -ENOMEM implies there is not.
94 *
95 * We currently support three overcommit policies, which are set via the
96 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
97 *
98 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
99 * Additional code 2002 Jul 20 by Robert Love.
100 *
101 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
102 *
103 * Note this is a helper function intended to be used by LSMs which
104 * wish to use this logic.
105 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700106int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107{
108 unsigned long free, allowed;
109
110 vm_acct_memory(pages);
111
112 /*
113 * Sometimes we want to use more memory than we have
114 */
115 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
116 return 0;
117
118 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
119 unsigned long n;
120
Christoph Lameter347ce432006-06-30 01:55:35 -0700121 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 free += nr_swap_pages;
123
124 /*
125 * Any slabs which are created with the
126 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
127 * which are reclaimable, under pressure. The dentry
128 * cache and most inode caches should fall into this
129 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700130 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 /*
133 * Leave the last 3% for root
134 */
135 if (!cap_sys_admin)
136 free -= free / 32;
137
138 if (free > pages)
139 return 0;
140
141 /*
142 * nr_free_pages() is very expensive on large systems,
143 * only call if we're about to fail.
144 */
145 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700146
147 /*
148 * Leave reserved pages. The pages are not for anonymous pages.
149 */
150 if (n <= totalreserve_pages)
151 goto error;
152 else
153 n -= totalreserve_pages;
154
155 /*
156 * Leave the last 3% for root
157 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 if (!cap_sys_admin)
159 n -= n / 32;
160 free += n;
161
162 if (free > pages)
163 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700164
165 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 }
167
168 allowed = (totalram_pages - hugetlb_total_pages())
169 * sysctl_overcommit_ratio / 100;
170 /*
171 * Leave the last 3% for root
172 */
173 if (!cap_sys_admin)
174 allowed -= allowed / 32;
175 allowed += total_swap_pages;
176
177 /* Don't let a single process grow too big:
178 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700179 if (mm)
180 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700182 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700184error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 vm_unacct_memory(pages);
186
187 return -ENOMEM;
188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/*
191 * Requires inode->i_mapping->i_mmap_lock
192 */
193static void __remove_shared_vm_struct(struct vm_area_struct *vma,
194 struct file *file, struct address_space *mapping)
195{
196 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800197 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 if (vma->vm_flags & VM_SHARED)
199 mapping->i_mmap_writable--;
200
201 flush_dcache_mmap_lock(mapping);
202 if (unlikely(vma->vm_flags & VM_NONLINEAR))
203 list_del_init(&vma->shared.vm_set.list);
204 else
205 vma_prio_tree_remove(vma, &mapping->i_mmap);
206 flush_dcache_mmap_unlock(mapping);
207}
208
209/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700210 * Unlink a file-based vm structure from its prio_tree, to hide
211 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700213void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 struct file *file = vma->vm_file;
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 if (file) {
218 struct address_space *mapping = file->f_mapping;
219 spin_lock(&mapping->i_mmap_lock);
220 __remove_shared_vm_struct(vma, file, mapping);
221 spin_unlock(&mapping->i_mmap_lock);
222 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700223}
224
225/*
226 * Close a vm structure and free it, returning the next.
227 */
228static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
229{
230 struct vm_area_struct *next = vma->vm_next;
231
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700232 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (vma->vm_ops && vma->vm_ops->close)
234 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700235 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700236 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700237 if (vma->vm_flags & VM_EXECUTABLE)
238 removed_exe_file_vma(vma->vm_mm);
239 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700240 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700242 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100245SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 unsigned long rlim, retval;
248 unsigned long newbrk, oldbrk;
249 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700250 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 down_write(&mm->mmap_sem);
253
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700254#ifdef CONFIG_COMPAT_BRK
255 min_brk = mm->end_code;
256#else
257 min_brk = mm->start_brk;
258#endif
259 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700261
262 /*
263 * Check against rlimit here. If this check is done later after the test
264 * of oldbrk with newbrk then it can escape the test and let the data
265 * segment grow beyond its set limit the in case where the limit is
266 * not page aligned -Ram Gupta
267 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800268 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100269 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
270 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700271 goto out;
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 newbrk = PAGE_ALIGN(brk);
274 oldbrk = PAGE_ALIGN(mm->brk);
275 if (oldbrk == newbrk)
276 goto set_brk;
277
278 /* Always allow shrinking brk. */
279 if (brk <= mm->brk) {
280 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
281 goto set_brk;
282 goto out;
283 }
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 /* Check against existing mmap mappings. */
286 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
287 goto out;
288
289 /* Ok, looks good - let it rip. */
290 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
291 goto out;
292set_brk:
293 mm->brk = brk;
294out:
295 retval = mm->brk;
296 up_write(&mm->mmap_sem);
297 return retval;
298}
299
300#ifdef DEBUG_MM_RB
301static int browse_rb(struct rb_root *root)
302{
303 int i = 0, j;
304 struct rb_node *nd, *pn = NULL;
305 unsigned long prev = 0, pend = 0;
306
307 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
308 struct vm_area_struct *vma;
309 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
310 if (vma->vm_start < prev)
311 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
312 if (vma->vm_start < pend)
313 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
314 if (vma->vm_start > vma->vm_end)
315 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
316 i++;
317 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800318 prev = vma->vm_start;
319 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 }
321 j = 0;
322 for (nd = pn; nd; nd = rb_prev(nd)) {
323 j++;
324 }
325 if (i != j)
326 printk("backwards %d, forwards %d\n", j, i), i = 0;
327 return i;
328}
329
330void validate_mm(struct mm_struct *mm)
331{
332 int bug = 0;
333 int i = 0;
334 struct vm_area_struct *tmp = mm->mmap;
335 while (tmp) {
336 tmp = tmp->vm_next;
337 i++;
338 }
339 if (i != mm->map_count)
340 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
341 i = browse_rb(&mm->mm_rb);
342 if (i != mm->map_count)
343 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200344 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346#else
347#define validate_mm(mm) do { } while (0)
348#endif
349
350static struct vm_area_struct *
351find_vma_prepare(struct mm_struct *mm, unsigned long addr,
352 struct vm_area_struct **pprev, struct rb_node ***rb_link,
353 struct rb_node ** rb_parent)
354{
355 struct vm_area_struct * vma;
356 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
357
358 __rb_link = &mm->mm_rb.rb_node;
359 rb_prev = __rb_parent = NULL;
360 vma = NULL;
361
362 while (*__rb_link) {
363 struct vm_area_struct *vma_tmp;
364
365 __rb_parent = *__rb_link;
366 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
367
368 if (vma_tmp->vm_end > addr) {
369 vma = vma_tmp;
370 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700371 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 __rb_link = &__rb_parent->rb_left;
373 } else {
374 rb_prev = __rb_parent;
375 __rb_link = &__rb_parent->rb_right;
376 }
377 }
378
379 *pprev = NULL;
380 if (rb_prev)
381 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
382 *rb_link = __rb_link;
383 *rb_parent = __rb_parent;
384 return vma;
385}
386
387static inline void
388__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
389 struct vm_area_struct *prev, struct rb_node *rb_parent)
390{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700391 struct vm_area_struct *next;
392
393 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 if (prev) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700395 next = prev->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 prev->vm_next = vma;
397 } else {
398 mm->mmap = vma;
399 if (rb_parent)
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700400 next = rb_entry(rb_parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 struct vm_area_struct, vm_rb);
402 else
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700403 next = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 }
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700405 vma->vm_next = next;
406 if (next)
407 next->vm_prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408}
409
410void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
411 struct rb_node **rb_link, struct rb_node *rb_parent)
412{
413 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
414 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
415}
416
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700417static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
ZhenwenXu48aae422009-01-06 14:40:21 -0800419 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421 file = vma->vm_file;
422 if (file) {
423 struct address_space *mapping = file->f_mapping;
424
425 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800426 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (vma->vm_flags & VM_SHARED)
428 mapping->i_mmap_writable++;
429
430 flush_dcache_mmap_lock(mapping);
431 if (unlikely(vma->vm_flags & VM_NONLINEAR))
432 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
433 else
434 vma_prio_tree_insert(vma, &mapping->i_mmap);
435 flush_dcache_mmap_unlock(mapping);
436 }
437}
438
439static void
440__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
441 struct vm_area_struct *prev, struct rb_node **rb_link,
442 struct rb_node *rb_parent)
443{
444 __vma_link_list(mm, vma, prev, rb_parent);
445 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
448static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
449 struct vm_area_struct *prev, struct rb_node **rb_link,
450 struct rb_node *rb_parent)
451{
452 struct address_space *mapping = NULL;
453
454 if (vma->vm_file)
455 mapping = vma->vm_file->f_mapping;
456
457 if (mapping) {
458 spin_lock(&mapping->i_mmap_lock);
459 vma->vm_truncate_count = mapping->truncate_count;
460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 __vma_link(mm, vma, prev, rb_link, rb_parent);
463 __vma_link_file(vma);
464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 if (mapping)
466 spin_unlock(&mapping->i_mmap_lock);
467
468 mm->map_count++;
469 validate_mm(mm);
470}
471
472/*
473 * Helper for vma_adjust in the split_vma insert case:
474 * insert vm structure into list and rbtree and anon_vma,
475 * but it has already been inserted into prio_tree earlier.
476 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800477static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
ZhenwenXu48aae422009-01-06 14:40:21 -0800479 struct vm_area_struct *__vma, *prev;
480 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200483 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 __vma_link(mm, vma, prev, rb_link, rb_parent);
485 mm->map_count++;
486}
487
488static inline void
489__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
490 struct vm_area_struct *prev)
491{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700492 struct vm_area_struct *next = vma->vm_next;
493
494 prev->vm_next = next;
495 if (next)
496 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 rb_erase(&vma->vm_rb, &mm->mm_rb);
498 if (mm->mmap_cache == vma)
499 mm->mmap_cache = prev;
500}
501
502/*
503 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
504 * is already present in an i_mmap tree without adjusting the tree.
505 * The following helper function should be used when such adjustments
506 * are necessary. The "insert" vma (if any) is to be inserted
507 * before we drop the necessary locks.
508 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800509int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
511{
512 struct mm_struct *mm = vma->vm_mm;
513 struct vm_area_struct *next = vma->vm_next;
514 struct vm_area_struct *importer = NULL;
515 struct address_space *mapping = NULL;
516 struct prio_tree_root *root = NULL;
Rik van Riel012f1802010-08-09 17:18:40 -0700517 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 long adjust_next = 0;
520 int remove_next = 0;
521
522 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700523 struct vm_area_struct *exporter = NULL;
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (end >= next->vm_end) {
526 /*
527 * vma expands, overlapping all the next, and
528 * perhaps the one after too (mprotect case 6).
529 */
530again: remove_next = 1 + (end > next->vm_end);
531 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700532 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 importer = vma;
534 } else if (end > next->vm_start) {
535 /*
536 * vma expands, overlapping part of the next:
537 * mprotect case 5 shifting the boundary up.
538 */
539 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700540 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 importer = vma;
542 } else if (end < vma->vm_end) {
543 /*
544 * vma shrinks, and !insert tells it's not
545 * split_vma inserting another: so it must be
546 * mprotect case 4 shifting the boundary down.
547 */
548 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700549 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 importer = next;
551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Rik van Riel5beb4932010-03-05 13:42:07 -0800553 /*
554 * Easily overlooked: when mprotect shifts the boundary,
555 * make sure the expanding vma has anon_vma set if the
556 * shrinking vma had, to cover any anon pages imported.
557 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700558 if (exporter && exporter->anon_vma && !importer->anon_vma) {
559 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800560 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700561 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800562 }
563 }
564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (file) {
566 mapping = file->f_mapping;
567 if (!(vma->vm_flags & VM_NONLINEAR))
568 root = &mapping->i_mmap;
569 spin_lock(&mapping->i_mmap_lock);
570 if (importer &&
571 vma->vm_truncate_count != next->vm_truncate_count) {
572 /*
573 * unmap_mapping_range might be in progress:
574 * ensure that the expanding vma is rescanned.
575 */
576 importer->vm_truncate_count = 0;
577 }
578 if (insert) {
579 insert->vm_truncate_count = vma->vm_truncate_count;
580 /*
581 * Put into prio_tree now, so instantiated pages
582 * are visible to arm/parisc __flush_dcache_page
583 * throughout; but we cannot insert into address
584 * space until vma start or end is updated.
585 */
586 __vma_link_file(insert);
587 }
588 }
589
Rik van Riel012f1802010-08-09 17:18:40 -0700590 /*
591 * When changing only vma->vm_end, we don't really need anon_vma
592 * lock. This is a fairly rare case by itself, but the anon_vma
593 * lock may be shared between many sibling processes. Skipping
594 * the lock for brk adjustments makes a difference sometimes.
595 */
596 if (vma->anon_vma && (insert || importer || start != vma->vm_start)) {
597 anon_vma = vma->anon_vma;
598 anon_vma_lock(anon_vma);
599 }
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (root) {
602 flush_dcache_mmap_lock(mapping);
603 vma_prio_tree_remove(vma, root);
604 if (adjust_next)
605 vma_prio_tree_remove(next, root);
606 }
607
608 vma->vm_start = start;
609 vma->vm_end = end;
610 vma->vm_pgoff = pgoff;
611 if (adjust_next) {
612 next->vm_start += adjust_next << PAGE_SHIFT;
613 next->vm_pgoff += adjust_next;
614 }
615
616 if (root) {
617 if (adjust_next)
618 vma_prio_tree_insert(next, root);
619 vma_prio_tree_insert(vma, root);
620 flush_dcache_mmap_unlock(mapping);
621 }
622
623 if (remove_next) {
624 /*
625 * vma_merge has merged next into vma, and needs
626 * us to remove next before dropping the locks.
627 */
628 __vma_unlink(mm, next, vma);
629 if (file)
630 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 } else if (insert) {
632 /*
633 * split_vma has split insert from vma, and needs
634 * us to insert it before dropping the locks
635 * (it may either follow vma or precede it).
636 */
637 __insert_vm_struct(mm, insert);
638 }
639
Rik van Riel012f1802010-08-09 17:18:40 -0700640 if (anon_vma)
641 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 if (mapping)
643 spin_unlock(&mapping->i_mmap_lock);
644
645 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700646 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700648 if (next->vm_flags & VM_EXECUTABLE)
649 removed_exe_file_vma(mm);
650 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800651 if (next->anon_vma)
652 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700654 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 kmem_cache_free(vm_area_cachep, next);
656 /*
657 * In mprotect's case 6 (see comments on vma_merge),
658 * we must remove another next too. It would clutter
659 * up the code too much to do both in one go.
660 */
661 if (remove_next == 2) {
662 next = vma->vm_next;
663 goto again;
664 }
665 }
666
667 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800668
669 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
671
672/*
673 * If the vma has a ->close operation then the driver probably needs to release
674 * per-vma resources, so we don't attempt to merge those.
675 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676static inline int is_mergeable_vma(struct vm_area_struct *vma,
677 struct file *file, unsigned long vm_flags)
678{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700679 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
680 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return 0;
682 if (vma->vm_file != file)
683 return 0;
684 if (vma->vm_ops && vma->vm_ops->close)
685 return 0;
686 return 1;
687}
688
689static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
690 struct anon_vma *anon_vma2)
691{
692 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
693}
694
695/*
696 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
697 * in front of (at a lower virtual address and file offset than) the vma.
698 *
699 * We cannot merge two vmas if they have differently assigned (non-NULL)
700 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
701 *
702 * We don't check here for the merged mmap wrapping around the end of pagecache
703 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
704 * wrap, nor mmaps which cover the final page at index -1UL.
705 */
706static int
707can_vma_merge_before(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) &&
711 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
712 if (vma->vm_pgoff == vm_pgoff)
713 return 1;
714 }
715 return 0;
716}
717
718/*
719 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
720 * beyond (at a higher virtual address and file offset than) the vma.
721 *
722 * We cannot merge two vmas if they have differently assigned (non-NULL)
723 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
724 */
725static int
726can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
727 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
728{
729 if (is_mergeable_vma(vma, file, vm_flags) &&
730 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
731 pgoff_t vm_pglen;
732 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
733 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
734 return 1;
735 }
736 return 0;
737}
738
739/*
740 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
741 * whether that can be merged with its predecessor or its successor.
742 * Or both (it neatly fills a hole).
743 *
744 * In most cases - when called for mmap, brk or mremap - [addr,end) is
745 * certain not to be mapped by the time vma_merge is called; but when
746 * called for mprotect, it is certain to be already mapped (either at
747 * an offset within prev, or at the start of next), and the flags of
748 * this area are about to be changed to vm_flags - and the no-change
749 * case has already been eliminated.
750 *
751 * The following mprotect cases have to be considered, where AAAA is
752 * the area passed down from mprotect_fixup, never extending beyond one
753 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
754 *
755 * AAAA AAAA AAAA AAAA
756 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
757 * cannot merge might become might become might become
758 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
759 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
760 * mremap move: PPPPNNNNNNNN 8
761 * AAAA
762 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
763 * might become case 1 below case 2 below case 3 below
764 *
765 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
766 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
767 */
768struct vm_area_struct *vma_merge(struct mm_struct *mm,
769 struct vm_area_struct *prev, unsigned long addr,
770 unsigned long end, unsigned long vm_flags,
771 struct anon_vma *anon_vma, struct file *file,
772 pgoff_t pgoff, struct mempolicy *policy)
773{
774 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
775 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800776 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 /*
779 * We later require that vma->vm_flags == vm_flags,
780 * so this tests vma->vm_flags & VM_SPECIAL, too.
781 */
782 if (vm_flags & VM_SPECIAL)
783 return NULL;
784
785 if (prev)
786 next = prev->vm_next;
787 else
788 next = mm->mmap;
789 area = next;
790 if (next && next->vm_end == end) /* cases 6, 7, 8 */
791 next = next->vm_next;
792
793 /*
794 * Can it merge with the predecessor?
795 */
796 if (prev && prev->vm_end == addr &&
797 mpol_equal(vma_policy(prev), policy) &&
798 can_vma_merge_after(prev, vm_flags,
799 anon_vma, file, pgoff)) {
800 /*
801 * OK, it can. Can we now merge in the successor as well?
802 */
803 if (next && end == next->vm_start &&
804 mpol_equal(policy, vma_policy(next)) &&
805 can_vma_merge_before(next, vm_flags,
806 anon_vma, file, pgoff+pglen) &&
807 is_mergeable_anon_vma(prev->anon_vma,
808 next->anon_vma)) {
809 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800810 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 next->vm_end, prev->vm_pgoff, NULL);
812 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800813 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800815 if (err)
816 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 return prev;
818 }
819
820 /*
821 * Can this new request be merged in front of next?
822 */
823 if (next && end == next->vm_start &&
824 mpol_equal(policy, vma_policy(next)) &&
825 can_vma_merge_before(next, vm_flags,
826 anon_vma, file, pgoff+pglen)) {
827 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800828 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 addr, prev->vm_pgoff, NULL);
830 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800831 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800833 if (err)
834 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 return area;
836 }
837
838 return NULL;
839}
840
841/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700842 * Rough compatbility check to quickly see if it's even worth looking
843 * at sharing an anon_vma.
844 *
845 * They need to have the same vm_file, and the flags can only differ
846 * in things that mprotect may change.
847 *
848 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
849 * we can merge the two vma's. For example, we refuse to merge a vma if
850 * there is a vm_ops->close() function, because that indicates that the
851 * driver is doing some kind of reference counting. But that doesn't
852 * really matter for the anon_vma sharing case.
853 */
854static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
855{
856 return a->vm_end == b->vm_start &&
857 mpol_equal(vma_policy(a), vma_policy(b)) &&
858 a->vm_file == b->vm_file &&
859 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
860 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
861}
862
863/*
864 * Do some basic sanity checking to see if we can re-use the anon_vma
865 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
866 * the same as 'old', the other will be the new one that is trying
867 * to share the anon_vma.
868 *
869 * NOTE! This runs with mm_sem held for reading, so it is possible that
870 * the anon_vma of 'old' is concurrently in the process of being set up
871 * by another page fault trying to merge _that_. But that's ok: if it
872 * is being set up, that automatically means that it will be a singleton
873 * acceptable for merging, so we can do all of this optimistically. But
874 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
875 *
876 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
877 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
878 * is to return an anon_vma that is "complex" due to having gone through
879 * a fork).
880 *
881 * We also make sure that the two vma's are compatible (adjacent,
882 * and with the same memory policies). That's all stable, even with just
883 * a read lock on the mm_sem.
884 */
885static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
886{
887 if (anon_vma_compatible(a, b)) {
888 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
889
890 if (anon_vma && list_is_singular(&old->anon_vma_chain))
891 return anon_vma;
892 }
893 return NULL;
894}
895
896/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
898 * neighbouring vmas for a suitable anon_vma, before it goes off
899 * to allocate a new anon_vma. It checks because a repetitive
900 * sequence of mprotects and faults may otherwise lead to distinct
901 * anon_vmas being allocated, preventing vma merge in subsequent
902 * mprotect.
903 */
904struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
905{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700906 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 near = vma->vm_next;
910 if (!near)
911 goto try_prev;
912
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700913 anon_vma = reusable_anon_vma(near, vma, near);
914 if (anon_vma)
915 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916try_prev:
917 /*
918 * It is potentially slow to have to call find_vma_prev here.
919 * But it's only on the first write fault on the vma, not
920 * every time, and we could devise a way to avoid it later
921 * (e.g. stash info in next's anon_vma_node when assigning
922 * an anon_vma, or when trying vma_merge). Another time.
923 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200924 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 if (!near)
926 goto none;
927
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700928 anon_vma = reusable_anon_vma(near, near, vma);
929 if (anon_vma)
930 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931none:
932 /*
933 * There's no absolute need to look only at touching neighbours:
934 * we could search further afield for "compatible" anon_vmas.
935 * But it would probably just be a waste of time searching,
936 * or lead to too many vmas hanging off the same anon_vma.
937 * We're trying to allow mprotect remerging later on,
938 * not trying to minimize memory used for anon_vmas.
939 */
940 return NULL;
941}
942
943#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700944void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 struct file *file, long pages)
946{
947 const unsigned long stack_flags
948 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 if (file) {
951 mm->shared_vm += pages;
952 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
953 mm->exec_vm += pages;
954 } else if (flags & stack_flags)
955 mm->stack_vm += pages;
956 if (flags & (VM_RESERVED|VM_IO))
957 mm->reserved_vm += pages;
958}
959#endif /* CONFIG_PROC_FS */
960
961/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700962 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 */
964
ZhenwenXu48aae422009-01-06 14:40:21 -0800965unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 unsigned long len, unsigned long prot,
967 unsigned long flags, unsigned long pgoff)
968{
969 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 struct inode *inode;
971 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700973 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 /*
976 * Does the application expect PROT_READ to imply PROT_EXEC?
977 *
978 * (the exception is when the underlying filesystem is noexec
979 * mounted, in which case we dont add PROT_EXEC.)
980 */
981 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800982 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 prot |= PROT_EXEC;
984
985 if (!len)
986 return -EINVAL;
987
Eric Paris7cd94142007-11-26 18:47:40 -0500988 if (!(flags & MAP_FIXED))
989 addr = round_hint_to_min(addr);
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 /* Careful about overflows.. */
992 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500993 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 return -ENOMEM;
995
996 /* offset overflow? */
997 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
998 return -EOVERFLOW;
999
1000 /* Too many mappings? */
1001 if (mm->map_count > sysctl_max_map_count)
1002 return -ENOMEM;
1003
1004 /* Obtain the address to map to. we verify (or select) it and ensure
1005 * that it represents a valid section of the address space.
1006 */
1007 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1008 if (addr & ~PAGE_MASK)
1009 return addr;
1010
1011 /* Do simple checking here so the lower-level routines won't have
1012 * to. we assume access permissions have been handled by the open
1013 * of the memory object, so we don't do any here.
1014 */
1015 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1016 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1017
Huang Shijiecdf7b342009-09-21 17:03:36 -07001018 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (!can_do_mlock())
1020 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 /* mlock MCL_FUTURE? */
1023 if (vm_flags & VM_LOCKED) {
1024 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001025 locked = len >> PAGE_SHIFT;
1026 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001027 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001028 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1030 return -EAGAIN;
1031 }
1032
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001033 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
1035 if (file) {
1036 switch (flags & MAP_TYPE) {
1037 case MAP_SHARED:
1038 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1039 return -EACCES;
1040
1041 /*
1042 * Make sure we don't allow writing to an append-only
1043 * file..
1044 */
1045 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1046 return -EACCES;
1047
1048 /*
1049 * Make sure there are no mandatory locks on the file.
1050 */
1051 if (locks_verify_locked(inode))
1052 return -EAGAIN;
1053
1054 vm_flags |= VM_SHARED | VM_MAYSHARE;
1055 if (!(file->f_mode & FMODE_WRITE))
1056 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1057
1058 /* fall through */
1059 case MAP_PRIVATE:
1060 if (!(file->f_mode & FMODE_READ))
1061 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001062 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001063 if (vm_flags & VM_EXEC)
1064 return -EPERM;
1065 vm_flags &= ~VM_MAYEXEC;
1066 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001067
1068 if (!file->f_op || !file->f_op->mmap)
1069 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 break;
1071
1072 default:
1073 return -EINVAL;
1074 }
1075 } else {
1076 switch (flags & MAP_TYPE) {
1077 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001078 /*
1079 * Ignore pgoff.
1080 */
1081 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 vm_flags |= VM_SHARED | VM_MAYSHARE;
1083 break;
1084 case MAP_PRIVATE:
1085 /*
1086 * Set pgoff according to addr for anon_vma.
1087 */
1088 pgoff = addr >> PAGE_SHIFT;
1089 break;
1090 default:
1091 return -EINVAL;
1092 }
1093 }
1094
Eric Parised032182007-06-28 15:55:21 -04001095 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (error)
1097 return error;
Eric Parised032182007-06-28 15:55:21 -04001098
Mel Gorman5a6fe122009-02-10 14:02:27 +00001099 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001100}
1101EXPORT_SYMBOL(do_mmap_pgoff);
1102
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001103SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1104 unsigned long, prot, unsigned long, flags,
1105 unsigned long, fd, unsigned long, pgoff)
1106{
1107 struct file *file = NULL;
1108 unsigned long retval = -EBADF;
1109
1110 if (!(flags & MAP_ANONYMOUS)) {
1111 if (unlikely(flags & MAP_HUGETLB))
1112 return -EINVAL;
1113 file = fget(fd);
1114 if (!file)
1115 goto out;
1116 } else if (flags & MAP_HUGETLB) {
1117 struct user_struct *user = NULL;
1118 /*
1119 * VM_NORESERVE is used because the reservations will be
1120 * taken when vm_ops->mmap() is called
1121 * A dummy user value is used because we are not locking
1122 * memory so no accounting is necessary
1123 */
1124 len = ALIGN(len, huge_page_size(&default_hstate));
1125 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1126 &user, HUGETLB_ANONHUGE_INODE);
1127 if (IS_ERR(file))
1128 return PTR_ERR(file);
1129 }
1130
1131 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1132
1133 down_write(&current->mm->mmap_sem);
1134 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1135 up_write(&current->mm->mmap_sem);
1136
1137 if (file)
1138 fput(file);
1139out:
1140 return retval;
1141}
1142
Christoph Hellwiga4679372010-03-10 15:21:15 -08001143#ifdef __ARCH_WANT_SYS_OLD_MMAP
1144struct mmap_arg_struct {
1145 unsigned long addr;
1146 unsigned long len;
1147 unsigned long prot;
1148 unsigned long flags;
1149 unsigned long fd;
1150 unsigned long offset;
1151};
1152
1153SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1154{
1155 struct mmap_arg_struct a;
1156
1157 if (copy_from_user(&a, arg, sizeof(a)))
1158 return -EFAULT;
1159 if (a.offset & ~PAGE_MASK)
1160 return -EINVAL;
1161
1162 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1163 a.offset >> PAGE_SHIFT);
1164}
1165#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1166
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001167/*
1168 * Some shared mappigns will want the pages marked read-only
1169 * to track write events. If so, we'll downgrade vm_page_prot
1170 * to the private version (using protection_map[] without the
1171 * VM_SHARED bit).
1172 */
1173int vma_wants_writenotify(struct vm_area_struct *vma)
1174{
1175 unsigned int vm_flags = vma->vm_flags;
1176
1177 /* If it was private or non-writable, the write bit is already clear */
1178 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1179 return 0;
1180
1181 /* The backer wishes to know when pages are first written to? */
1182 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1183 return 1;
1184
1185 /* The open routine did something to the protections already? */
1186 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001187 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001188 return 0;
1189
1190 /* Specialty mapping? */
1191 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1192 return 0;
1193
1194 /* Can the mapping track the dirty pages? */
1195 return vma->vm_file && vma->vm_file->f_mapping &&
1196 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1197}
1198
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001199/*
1200 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001201 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001202 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001203static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001204{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001205 /*
1206 * hugetlb has its own accounting separate from the core VM
1207 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1208 */
1209 if (file && is_file_hugepages(file))
1210 return 0;
1211
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001212 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1213}
1214
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001215unsigned long mmap_region(struct file *file, unsigned long addr,
1216 unsigned long len, unsigned long flags,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001217 unsigned int vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001218{
1219 struct mm_struct *mm = current->mm;
1220 struct vm_area_struct *vma, *prev;
1221 int correct_wcount = 0;
1222 int error;
1223 struct rb_node **rb_link, *rb_parent;
1224 unsigned long charged = 0;
1225 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 /* Clear old maps */
1228 error = -ENOMEM;
1229munmap_back:
1230 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1231 if (vma && vma->vm_start < addr + len) {
1232 if (do_munmap(mm, addr, len))
1233 return -ENOMEM;
1234 goto munmap_back;
1235 }
1236
1237 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001238 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 return -ENOMEM;
1240
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001241 /*
1242 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001243 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001244 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001245 if ((flags & MAP_NORESERVE)) {
1246 /* We honor MAP_NORESERVE if allowed to overcommit */
1247 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1248 vm_flags |= VM_NORESERVE;
1249
1250 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1251 if (file && is_file_hugepages(file))
1252 vm_flags |= VM_NORESERVE;
1253 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001254
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001255 /*
1256 * Private writable mapping: check memory availability
1257 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001258 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001259 charged = len >> PAGE_SHIFT;
1260 if (security_vm_enough_memory(charged))
1261 return -ENOMEM;
1262 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 }
1264
1265 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001266 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001268 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1269 if (vma)
1270 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
1272 /*
1273 * Determine the object being mapped and call the appropriate
1274 * specific mapper. the address has already been validated, but
1275 * not unmapped, but the maps are removed from the list.
1276 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001277 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (!vma) {
1279 error = -ENOMEM;
1280 goto unacct_error;
1281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
1283 vma->vm_mm = mm;
1284 vma->vm_start = addr;
1285 vma->vm_end = addr + len;
1286 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001287 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001289 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 if (file) {
1292 error = -EINVAL;
1293 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1294 goto free_vma;
1295 if (vm_flags & VM_DENYWRITE) {
1296 error = deny_write_access(file);
1297 if (error)
1298 goto free_vma;
1299 correct_wcount = 1;
1300 }
1301 vma->vm_file = file;
1302 get_file(file);
1303 error = file->f_op->mmap(file, vma);
1304 if (error)
1305 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001306 if (vm_flags & VM_EXECUTABLE)
1307 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001308
1309 /* Can addr have changed??
1310 *
1311 * Answer: Yes, several device drivers can do it in their
1312 * f_op->mmap method. -DaveM
1313 */
1314 addr = vma->vm_start;
1315 pgoff = vma->vm_pgoff;
1316 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 } else if (vm_flags & VM_SHARED) {
1318 error = shmem_zero_setup(vma);
1319 if (error)
1320 goto free_vma;
1321 }
1322
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001323 if (vma_wants_writenotify(vma)) {
1324 pgprot_t pprot = vma->vm_page_prot;
1325
1326 /* Can vma->vm_page_prot have changed??
1327 *
1328 * Answer: Yes, drivers may have changed it in their
1329 * f_op->mmap method.
1330 *
1331 * Ensures that vmas marked as uncached stay that way.
1332 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001333 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001334 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1335 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1336 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001337
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001338 vma_link(mm, vma, prev, rb_link, rb_parent);
1339 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001340
1341 /* Once vma denies write, undo our temporary denial count */
1342 if (correct_wcount)
1343 atomic_inc(&inode->i_writecount);
1344out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001345 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001346
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001348 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001350 if (!mlock_vma_pages_range(vma, addr, addr + len))
1351 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001352 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001353 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 return addr;
1355
1356unmap_and_free_vma:
1357 if (correct_wcount)
1358 atomic_inc(&inode->i_writecount);
1359 vma->vm_file = NULL;
1360 fput(file);
1361
1362 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001363 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1364 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365free_vma:
1366 kmem_cache_free(vm_area_cachep, vma);
1367unacct_error:
1368 if (charged)
1369 vm_unacct_memory(charged);
1370 return error;
1371}
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373/* Get an address range which is currently unmapped.
1374 * For shmat() with addr=0.
1375 *
1376 * Ugly calling convention alert:
1377 * Return value with the low bits set means error value,
1378 * ie
1379 * if (ret & ~PAGE_MASK)
1380 * error = ret;
1381 *
1382 * This function "knows" that -ENOMEM has the bits set.
1383 */
1384#ifndef HAVE_ARCH_UNMAPPED_AREA
1385unsigned long
1386arch_get_unmapped_area(struct file *filp, unsigned long addr,
1387 unsigned long len, unsigned long pgoff, unsigned long flags)
1388{
1389 struct mm_struct *mm = current->mm;
1390 struct vm_area_struct *vma;
1391 unsigned long start_addr;
1392
1393 if (len > TASK_SIZE)
1394 return -ENOMEM;
1395
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001396 if (flags & MAP_FIXED)
1397 return addr;
1398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 if (addr) {
1400 addr = PAGE_ALIGN(addr);
1401 vma = find_vma(mm, addr);
1402 if (TASK_SIZE - len >= addr &&
1403 (!vma || addr + len <= vma->vm_start))
1404 return addr;
1405 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001406 if (len > mm->cached_hole_size) {
1407 start_addr = addr = mm->free_area_cache;
1408 } else {
1409 start_addr = addr = TASK_UNMAPPED_BASE;
1410 mm->cached_hole_size = 0;
1411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
1413full_search:
1414 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1415 /* At this point: (!vma || addr < vma->vm_end). */
1416 if (TASK_SIZE - len < addr) {
1417 /*
1418 * Start a new search - just in case we missed
1419 * some holes.
1420 */
1421 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001422 addr = TASK_UNMAPPED_BASE;
1423 start_addr = addr;
1424 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 goto full_search;
1426 }
1427 return -ENOMEM;
1428 }
1429 if (!vma || addr + len <= vma->vm_start) {
1430 /*
1431 * Remember the place where we stopped the search:
1432 */
1433 mm->free_area_cache = addr + len;
1434 return addr;
1435 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001436 if (addr + mm->cached_hole_size < vma->vm_start)
1437 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 addr = vma->vm_end;
1439 }
1440}
1441#endif
1442
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001443void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
1445 /*
1446 * Is this a new hole at the lowest possible address?
1447 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001448 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1449 mm->free_area_cache = addr;
1450 mm->cached_hole_size = ~0UL;
1451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
1454/*
1455 * This mmap-allocator allocates new areas top-down from below the
1456 * stack's low limit (the base):
1457 */
1458#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1459unsigned long
1460arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1461 const unsigned long len, const unsigned long pgoff,
1462 const unsigned long flags)
1463{
1464 struct vm_area_struct *vma;
1465 struct mm_struct *mm = current->mm;
1466 unsigned long addr = addr0;
1467
1468 /* requested length too big for entire address space */
1469 if (len > TASK_SIZE)
1470 return -ENOMEM;
1471
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001472 if (flags & MAP_FIXED)
1473 return addr;
1474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 /* requesting a specific address */
1476 if (addr) {
1477 addr = PAGE_ALIGN(addr);
1478 vma = find_vma(mm, addr);
1479 if (TASK_SIZE - len >= addr &&
1480 (!vma || addr + len <= vma->vm_start))
1481 return addr;
1482 }
1483
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001484 /* check if free_area_cache is useful for us */
1485 if (len <= mm->cached_hole_size) {
1486 mm->cached_hole_size = 0;
1487 mm->free_area_cache = mm->mmap_base;
1488 }
1489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 /* either no address requested or can't fit in requested address hole */
1491 addr = mm->free_area_cache;
1492
1493 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001494 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 vma = find_vma(mm, addr-len);
1496 if (!vma || addr <= vma->vm_start)
1497 /* remember the address as a hint for next time */
1498 return (mm->free_area_cache = addr-len);
1499 }
1500
Chris Wright73219d12005-06-21 17:14:52 -07001501 if (mm->mmap_base < len)
1502 goto bottomup;
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 addr = mm->mmap_base-len;
1505
1506 do {
1507 /*
1508 * Lookup failure means no vma is above this address,
1509 * else if new region fits below vma->vm_start,
1510 * return with success:
1511 */
1512 vma = find_vma(mm, addr);
1513 if (!vma || addr+len <= vma->vm_start)
1514 /* remember the address as a hint for next time */
1515 return (mm->free_area_cache = addr);
1516
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001517 /* remember the largest hole we saw so far */
1518 if (addr + mm->cached_hole_size < vma->vm_start)
1519 mm->cached_hole_size = vma->vm_start - addr;
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 /* try just below the current vma->vm_start */
1522 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001523 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
Chris Wright73219d12005-06-21 17:14:52 -07001525bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 /*
1527 * A failed mmap() very likely causes application failure,
1528 * so fall back to the bottom-up function here. This scenario
1529 * can happen with large stack limits and large mmap()
1530 * allocations.
1531 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001532 mm->cached_hole_size = ~0UL;
1533 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1535 /*
1536 * Restore the topdown base:
1537 */
1538 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001539 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541 return addr;
1542}
1543#endif
1544
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001545void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
1547 /*
1548 * Is this a new hole at the highest possible address?
1549 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001550 if (addr > mm->free_area_cache)
1551 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
1553 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001554 if (mm->free_area_cache > mm->mmap_base)
1555 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556}
1557
1558unsigned long
1559get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1560 unsigned long pgoff, unsigned long flags)
1561{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001562 unsigned long (*get_area)(struct file *, unsigned long,
1563 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Al Viro9206de92009-12-03 15:23:11 -05001565 unsigned long error = arch_mmap_check(addr, len, flags);
1566 if (error)
1567 return error;
1568
1569 /* Careful about overflows.. */
1570 if (len > TASK_SIZE)
1571 return -ENOMEM;
1572
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001573 get_area = current->mm->get_unmapped_area;
1574 if (file && file->f_op && file->f_op->get_unmapped_area)
1575 get_area = file->f_op->get_unmapped_area;
1576 addr = get_area(file, addr, len, pgoff, flags);
1577 if (IS_ERR_VALUE(addr))
1578 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001580 if (addr > TASK_SIZE - len)
1581 return -ENOMEM;
1582 if (addr & ~PAGE_MASK)
1583 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001584
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001585 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586}
1587
1588EXPORT_SYMBOL(get_unmapped_area);
1589
1590/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001591struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592{
1593 struct vm_area_struct *vma = NULL;
1594
1595 if (mm) {
1596 /* Check the cache first. */
1597 /* (Cache hit rate is typically around 35%.) */
1598 vma = mm->mmap_cache;
1599 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1600 struct rb_node * rb_node;
1601
1602 rb_node = mm->mm_rb.rb_node;
1603 vma = NULL;
1604
1605 while (rb_node) {
1606 struct vm_area_struct * vma_tmp;
1607
1608 vma_tmp = rb_entry(rb_node,
1609 struct vm_area_struct, vm_rb);
1610
1611 if (vma_tmp->vm_end > addr) {
1612 vma = vma_tmp;
1613 if (vma_tmp->vm_start <= addr)
1614 break;
1615 rb_node = rb_node->rb_left;
1616 } else
1617 rb_node = rb_node->rb_right;
1618 }
1619 if (vma)
1620 mm->mmap_cache = vma;
1621 }
1622 }
1623 return vma;
1624}
1625
1626EXPORT_SYMBOL(find_vma);
1627
1628/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1629struct vm_area_struct *
1630find_vma_prev(struct mm_struct *mm, unsigned long addr,
1631 struct vm_area_struct **pprev)
1632{
1633 struct vm_area_struct *vma = NULL, *prev = NULL;
ZhenwenXu48aae422009-01-06 14:40:21 -08001634 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 if (!mm)
1636 goto out;
1637
1638 /* Guard against addr being lower than the first VMA */
1639 vma = mm->mmap;
1640
1641 /* Go through the RB tree quickly. */
1642 rb_node = mm->mm_rb.rb_node;
1643
1644 while (rb_node) {
1645 struct vm_area_struct *vma_tmp;
1646 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1647
1648 if (addr < vma_tmp->vm_end) {
1649 rb_node = rb_node->rb_left;
1650 } else {
1651 prev = vma_tmp;
1652 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1653 break;
1654 rb_node = rb_node->rb_right;
1655 }
1656 }
1657
1658out:
1659 *pprev = prev;
1660 return prev ? prev->vm_next : vma;
1661}
1662
1663/*
1664 * Verify that the stack growth is acceptable and
1665 * update accounting. This is shared with both the
1666 * grow-up and grow-down cases.
1667 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001668static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669{
1670 struct mm_struct *mm = vma->vm_mm;
1671 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001672 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
1674 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001675 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 return -ENOMEM;
1677
1678 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001679 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 return -ENOMEM;
1681
1682 /* mlock limit tests */
1683 if (vma->vm_flags & VM_LOCKED) {
1684 unsigned long locked;
1685 unsigned long limit;
1686 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001687 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1688 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 if (locked > limit && !capable(CAP_IPC_LOCK))
1690 return -ENOMEM;
1691 }
1692
Adam Litke0d59a012007-01-30 14:35:39 -08001693 /* Check to ensure the stack will not grow into a hugetlb-only region */
1694 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1695 vma->vm_end - size;
1696 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1697 return -EFAULT;
1698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 /*
1700 * Overcommit.. This must be the final test, as it will
1701 * update security statistics.
1702 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001703 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 return -ENOMEM;
1705
1706 /* Ok, everything looks good - let it rip */
1707 mm->total_vm += grow;
1708 if (vma->vm_flags & VM_LOCKED)
1709 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001710 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return 0;
1712}
1713
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001714#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001716 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1717 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001719int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720{
1721 int error;
1722
1723 if (!(vma->vm_flags & VM_GROWSUP))
1724 return -EFAULT;
1725
1726 /*
1727 * We must make sure the anon_vma is allocated
1728 * so that the anon_vma locking is not a noop.
1729 */
1730 if (unlikely(anon_vma_prepare(vma)))
1731 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001732 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734 /*
1735 * vma->vm_start/vm_end cannot change under us because the caller
1736 * is required to hold the mmap_sem in read mode. We need the
1737 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001738 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 */
Helge Deller06b32f32006-12-19 19:28:33 +01001740 if (address < PAGE_ALIGN(address+4))
1741 address = PAGE_ALIGN(address+4);
1742 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001743 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001744 return -ENOMEM;
1745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 error = 0;
1747
1748 /* Somebody else might have raced and expanded it already */
1749 if (address > vma->vm_end) {
1750 unsigned long size, grow;
1751
1752 size = address - vma->vm_start;
1753 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1754
1755 error = acct_stack_growth(vma, size, grow);
Eric B Munson3af9e852010-05-18 15:30:49 +01001756 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 vma->vm_end = address;
Eric B Munson3af9e852010-05-18 15:30:49 +01001758 perf_event_mmap(vma);
1759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001761 vma_unlock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 return error;
1763}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001764#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766/*
1767 * vma is the first one with address < vma->vm_start. Have to extend vma.
1768 */
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001769static int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001770 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771{
1772 int error;
1773
1774 /*
1775 * We must make sure the anon_vma is allocated
1776 * so that the anon_vma locking is not a noop.
1777 */
1778 if (unlikely(anon_vma_prepare(vma)))
1779 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001780
1781 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001782 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001783 if (error)
1784 return error;
1785
Rik van Rielbb4a3402010-08-09 17:18:37 -07001786 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
1788 /*
1789 * vma->vm_start/vm_end cannot change under us because the caller
1790 * is required to hold the mmap_sem in read mode. We need the
1791 * anon_vma lock to serialize against concurrent expand_stacks.
1792 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
1794 /* Somebody else might have raced and expanded it already */
1795 if (address < vma->vm_start) {
1796 unsigned long size, grow;
1797
1798 size = vma->vm_end - address;
1799 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1800
1801 error = acct_stack_growth(vma, size, grow);
1802 if (!error) {
1803 vma->vm_start = address;
1804 vma->vm_pgoff -= grow;
Eric B Munson3af9e852010-05-18 15:30:49 +01001805 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 }
1807 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001808 vma_unlock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 return error;
1810}
1811
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001812int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1813{
1814 return expand_downwards(vma, address);
1815}
1816
1817#ifdef CONFIG_STACK_GROWSUP
1818int expand_stack(struct vm_area_struct *vma, unsigned long address)
1819{
1820 return expand_upwards(vma, address);
1821}
1822
1823struct vm_area_struct *
1824find_extend_vma(struct mm_struct *mm, unsigned long addr)
1825{
1826 struct vm_area_struct *vma, *prev;
1827
1828 addr &= PAGE_MASK;
1829 vma = find_vma_prev(mm, addr, &prev);
1830 if (vma && (vma->vm_start <= addr))
1831 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001832 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001833 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001834 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001835 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001836 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001837 return prev;
1838}
1839#else
1840int expand_stack(struct vm_area_struct *vma, unsigned long address)
1841{
1842 return expand_downwards(vma, address);
1843}
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845struct vm_area_struct *
1846find_extend_vma(struct mm_struct * mm, unsigned long addr)
1847{
1848 struct vm_area_struct * vma;
1849 unsigned long start;
1850
1851 addr &= PAGE_MASK;
1852 vma = find_vma(mm,addr);
1853 if (!vma)
1854 return NULL;
1855 if (vma->vm_start <= addr)
1856 return vma;
1857 if (!(vma->vm_flags & VM_GROWSDOWN))
1858 return NULL;
1859 start = vma->vm_start;
1860 if (expand_stack(vma, addr))
1861 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001862 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001863 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001864 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 return vma;
1866}
1867#endif
1868
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001870 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001872 *
1873 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001875static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001877 /* Update high watermark before we lower total_vm */
1878 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001880 long nrpages = vma_pages(vma);
1881
1882 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001883 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001884 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001885 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 validate_mm(mm);
1887}
1888
1889/*
1890 * Get rid of page table information in the indicated region.
1891 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001892 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 */
1894static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001895 struct vm_area_struct *vma, struct vm_area_struct *prev,
1896 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001898 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 struct mmu_gather *tlb;
1900 unsigned long nr_accounted = 0;
1901
1902 lru_add_drain();
1903 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001904 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001905 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 vm_unacct_memory(nr_accounted);
Jan Beulich42b77722008-07-23 21:27:10 -07001907 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001908 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 tlb_finish_mmu(tlb, start, end);
1910}
1911
1912/*
1913 * Create a list of vma's touched by the unmap, removing them from the mm's
1914 * vma list as we go..
1915 */
1916static void
1917detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1918 struct vm_area_struct *prev, unsigned long end)
1919{
1920 struct vm_area_struct **insertion_point;
1921 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001922 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001925 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 do {
1927 rb_erase(&vma->vm_rb, &mm->mm_rb);
1928 mm->map_count--;
1929 tail_vma = vma;
1930 vma = vma->vm_next;
1931 } while (vma && vma->vm_start < end);
1932 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001933 if (vma)
1934 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001936 if (mm->unmap_area == arch_unmap_area)
1937 addr = prev ? prev->vm_end : mm->mmap_base;
1938 else
1939 addr = vma ? vma->vm_start : mm->mmap_base;
1940 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 mm->mmap_cache = NULL; /* Kill the cache. */
1942}
1943
1944/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001945 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1946 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001948static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 unsigned long addr, int new_below)
1950{
1951 struct mempolicy *pol;
1952 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001953 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Andi Kleena5516432008-07-23 21:27:41 -07001955 if (is_vm_hugetlb_page(vma) && (addr &
1956 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 return -EINVAL;
1958
Christoph Lametere94b1762006-12-06 20:33:17 -08001959 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001961 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 /* most fields are the same, copy all, and then fixup */
1964 *new = *vma;
1965
Rik van Riel5beb4932010-03-05 13:42:07 -08001966 INIT_LIST_HEAD(&new->anon_vma_chain);
1967
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 if (new_below)
1969 new->vm_end = addr;
1970 else {
1971 new->vm_start = addr;
1972 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1973 }
1974
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001975 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001977 err = PTR_ERR(pol);
1978 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 }
1980 vma_set_policy(new, pol);
1981
Rik van Riel5beb4932010-03-05 13:42:07 -08001982 if (anon_vma_clone(new, vma))
1983 goto out_free_mpol;
1984
Matt Helsley925d1c42008-04-29 01:01:36 -07001985 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001987 if (vma->vm_flags & VM_EXECUTABLE)
1988 added_exe_file_vma(mm);
1989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
1991 if (new->vm_ops && new->vm_ops->open)
1992 new->vm_ops->open(new);
1993
1994 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08001995 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1997 else
Rik van Riel5beb4932010-03-05 13:42:07 -08001998 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Rik van Riel5beb4932010-03-05 13:42:07 -08002000 /* Success. */
2001 if (!err)
2002 return 0;
2003
2004 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002005 if (new->vm_ops && new->vm_ops->close)
2006 new->vm_ops->close(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002007 if (new->vm_file) {
2008 if (vma->vm_flags & VM_EXECUTABLE)
2009 removed_exe_file_vma(mm);
2010 fput(new->vm_file);
2011 }
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002012 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002013 out_free_mpol:
2014 mpol_put(pol);
2015 out_free_vma:
2016 kmem_cache_free(vm_area_cachep, new);
2017 out_err:
2018 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019}
2020
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002021/*
2022 * Split a vma into two pieces at address 'addr', a new vma is allocated
2023 * either for the first part or the tail.
2024 */
2025int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2026 unsigned long addr, int new_below)
2027{
2028 if (mm->map_count >= sysctl_max_map_count)
2029 return -ENOMEM;
2030
2031 return __split_vma(mm, vma, addr, new_below);
2032}
2033
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034/* Munmap is split into 2 main parts -- this part which finds
2035 * what needs doing, and the areas themselves, which do the
2036 * work. This now handles partial unmappings.
2037 * Jeremy Fitzhardinge <jeremy@goop.org>
2038 */
2039int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2040{
2041 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002042 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2045 return -EINVAL;
2046
2047 if ((len = PAGE_ALIGN(len)) == 0)
2048 return -EINVAL;
2049
2050 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07002051 vma = find_vma_prev(mm, start, &prev);
2052 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07002054 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
2056 /* if it doesn't overlap, we have nothing.. */
2057 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002058 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 return 0;
2060
2061 /*
2062 * If we need to split any vma, do it now to save pain later.
2063 *
2064 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2065 * unmapped vm_area_struct will remain in use: so lower split_vma
2066 * places tmp vma above, and higher split_vma places tmp vma below.
2067 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002068 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002069 int error;
2070
2071 /*
2072 * Make sure that map_count on return from munmap() will
2073 * not exceed its limit; but let map_count go just above
2074 * its limit temporarily, to help free resources as expected.
2075 */
2076 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2077 return -ENOMEM;
2078
2079 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 if (error)
2081 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002082 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 }
2084
2085 /* Does it split the last one? */
2086 last = find_vma(mm, end);
2087 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002088 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 if (error)
2090 return error;
2091 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002092 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
2094 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002095 * unlock any mlock()ed ranges before detaching vmas
2096 */
2097 if (mm->locked_vm) {
2098 struct vm_area_struct *tmp = vma;
2099 while (tmp && tmp->vm_start < end) {
2100 if (tmp->vm_flags & VM_LOCKED) {
2101 mm->locked_vm -= vma_pages(tmp);
2102 munlock_vma_pages_all(tmp);
2103 }
2104 tmp = tmp->vm_next;
2105 }
2106 }
2107
2108 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 * Remove the vma's, and unmap the actual pages
2110 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002111 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2112 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002115 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117 return 0;
2118}
2119
2120EXPORT_SYMBOL(do_munmap);
2121
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002122SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123{
2124 int ret;
2125 struct mm_struct *mm = current->mm;
2126
2127 profile_munmap(addr);
2128
2129 down_write(&mm->mmap_sem);
2130 ret = do_munmap(mm, addr, len);
2131 up_write(&mm->mmap_sem);
2132 return ret;
2133}
2134
2135static inline void verify_mm_writelocked(struct mm_struct *mm)
2136{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002137#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2139 WARN_ON(1);
2140 up_read(&mm->mmap_sem);
2141 }
2142#endif
2143}
2144
2145/*
2146 * this is really a simplified "do_mmap". it only handles
2147 * anonymous maps. eventually we may be able to do some
2148 * brk-specific accounting here.
2149 */
2150unsigned long do_brk(unsigned long addr, unsigned long len)
2151{
2152 struct mm_struct * mm = current->mm;
2153 struct vm_area_struct * vma, * prev;
2154 unsigned long flags;
2155 struct rb_node ** rb_link, * rb_parent;
2156 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002157 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
2159 len = PAGE_ALIGN(len);
2160 if (!len)
2161 return addr;
2162
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002163 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002164 if (error)
2165 return error;
2166
Kirill Korotaev3a459752006-09-07 14:17:04 +04002167 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2168
Al Viro2c6a10162009-12-03 19:40:46 -05002169 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2170 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002171 return error;
2172
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 /*
2174 * mlock MCL_FUTURE?
2175 */
2176 if (mm->def_flags & VM_LOCKED) {
2177 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002178 locked = len >> PAGE_SHIFT;
2179 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002180 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002181 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2183 return -EAGAIN;
2184 }
2185
2186 /*
2187 * mm->mmap_sem is required to protect against another thread
2188 * changing the mappings in case we sleep.
2189 */
2190 verify_mm_writelocked(mm);
2191
2192 /*
2193 * Clear old maps. this also does some error checking for us
2194 */
2195 munmap_back:
2196 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2197 if (vma && vma->vm_start < addr + len) {
2198 if (do_munmap(mm, addr, len))
2199 return -ENOMEM;
2200 goto munmap_back;
2201 }
2202
2203 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002204 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 return -ENOMEM;
2206
2207 if (mm->map_count > sysctl_max_map_count)
2208 return -ENOMEM;
2209
2210 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2211 return -ENOMEM;
2212
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002214 vma = vma_merge(mm, prev, addr, addr + len, flags,
2215 NULL, NULL, pgoff, NULL);
2216 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 goto out;
2218
2219 /*
2220 * create a vma struct for an anonymous mapping
2221 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002222 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 if (!vma) {
2224 vm_unacct_memory(len >> PAGE_SHIFT);
2225 return -ENOMEM;
2226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Rik van Riel5beb4932010-03-05 13:42:07 -08002228 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 vma->vm_mm = mm;
2230 vma->vm_start = addr;
2231 vma->vm_end = addr + len;
2232 vma->vm_pgoff = pgoff;
2233 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002234 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 vma_link(mm, vma, prev, rb_link, rb_parent);
2236out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002237 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 mm->total_vm += len >> PAGE_SHIFT;
2239 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002240 if (!mlock_vma_pages_range(vma, addr, addr + len))
2241 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 }
2243 return addr;
2244}
2245
2246EXPORT_SYMBOL(do_brk);
2247
2248/* Release all mmaps. */
2249void exit_mmap(struct mm_struct *mm)
2250{
2251 struct mmu_gather *tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002252 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002254 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002256 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002257 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002258
Rik van Rielba470de2008-10-18 20:26:50 -07002259 if (mm->locked_vm) {
2260 vma = mm->mmap;
2261 while (vma) {
2262 if (vma->vm_flags & VM_LOCKED)
2263 munlock_vma_pages_all(vma);
2264 vma = vma->vm_next;
2265 }
2266 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002267
2268 arch_exit_mmap(mm);
2269
Rik van Rielba470de2008-10-18 20:26:50 -07002270 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002271 if (!vma) /* Can happen if dup_mmap() received an OOM */
2272 return;
2273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002276 tlb = tlb_gather_mmu(mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002277 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002278 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002279 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 vm_unacct_memory(nr_accounted);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002281
Jan Beulich42b77722008-07-23 21:27:10 -07002282 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07002283 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002286 * Walk the list again, actually closing and freeing it,
2287 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002289 while (vma)
2290 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002291
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002292 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293}
2294
2295/* Insert vm structure into process list sorted by address
2296 * and into the inode's i_mmap tree. If vm_file is non-NULL
2297 * then i_mmap_lock is taken here.
2298 */
2299int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2300{
2301 struct vm_area_struct * __vma, * prev;
2302 struct rb_node ** rb_link, * rb_parent;
2303
2304 /*
2305 * The vm_pgoff of a purely anonymous vma should be irrelevant
2306 * until its first write fault, when page's anon_vma and index
2307 * are set. But now set the vm_pgoff it will almost certainly
2308 * end up with (unless mremap moves it elsewhere before that
2309 * first wfault), so /proc/pid/maps tells a consistent story.
2310 *
2311 * By setting it to reflect the virtual start address of the
2312 * vma, merges and splits can happen in a seamless way, just
2313 * using the existing file pgoff checks and manipulations.
2314 * Similarly in do_mmap_pgoff and in do_brk.
2315 */
2316 if (!vma->vm_file) {
2317 BUG_ON(vma->anon_vma);
2318 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2319 }
2320 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2321 if (__vma && __vma->vm_start < vma->vm_end)
2322 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002323 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002324 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002325 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 vma_link(mm, vma, prev, rb_link, rb_parent);
2327 return 0;
2328}
2329
2330/*
2331 * Copy the vma structure to a new location in the same mm,
2332 * prior to moving page table entries, to effect an mremap move.
2333 */
2334struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2335 unsigned long addr, unsigned long len, pgoff_t pgoff)
2336{
2337 struct vm_area_struct *vma = *vmap;
2338 unsigned long vma_start = vma->vm_start;
2339 struct mm_struct *mm = vma->vm_mm;
2340 struct vm_area_struct *new_vma, *prev;
2341 struct rb_node **rb_link, *rb_parent;
2342 struct mempolicy *pol;
2343
2344 /*
2345 * If anonymous vma has not yet been faulted, update new pgoff
2346 * to match new location, to increase its chance of merging.
2347 */
2348 if (!vma->vm_file && !vma->anon_vma)
2349 pgoff = addr >> PAGE_SHIFT;
2350
2351 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2352 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2353 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2354 if (new_vma) {
2355 /*
2356 * Source vma may have been merged into new_vma
2357 */
2358 if (vma_start >= new_vma->vm_start &&
2359 vma_start < new_vma->vm_end)
2360 *vmap = new_vma;
2361 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002362 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 if (new_vma) {
2364 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002365 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002366 if (IS_ERR(pol))
2367 goto out_free_vma;
2368 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2369 if (anon_vma_clone(new_vma, vma))
2370 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 vma_set_policy(new_vma, pol);
2372 new_vma->vm_start = addr;
2373 new_vma->vm_end = addr + len;
2374 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002375 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002377 if (vma->vm_flags & VM_EXECUTABLE)
2378 added_exe_file_vma(mm);
2379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 if (new_vma->vm_ops && new_vma->vm_ops->open)
2381 new_vma->vm_ops->open(new_vma);
2382 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2383 }
2384 }
2385 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002386
2387 out_free_mempol:
2388 mpol_put(pol);
2389 out_free_vma:
2390 kmem_cache_free(vm_area_cachep, new_vma);
2391 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002393
2394/*
2395 * Return true if the calling process may expand its vm space by the passed
2396 * number of pages
2397 */
2398int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2399{
2400 unsigned long cur = mm->total_vm; /* pages */
2401 unsigned long lim;
2402
Jiri Slaby59e99e52010-03-05 13:41:44 -08002403 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002404
2405 if (cur + npages > lim)
2406 return 0;
2407 return 1;
2408}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002409
2410
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002411static int special_mapping_fault(struct vm_area_struct *vma,
2412 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002413{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002414 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002415 struct page **pages;
2416
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002417 /*
2418 * special mappings have no vm_file, and in that case, the mm
2419 * uses vm_pgoff internally. So we have to subtract it from here.
2420 * We are allowed to do this because we are the mm; do not copy
2421 * this code into drivers!
2422 */
2423 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002424
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002425 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2426 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002427
2428 if (*pages) {
2429 struct page *page = *pages;
2430 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002431 vmf->page = page;
2432 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002433 }
2434
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002435 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002436}
2437
2438/*
2439 * Having a close hook prevents vma merging regardless of flags.
2440 */
2441static void special_mapping_close(struct vm_area_struct *vma)
2442{
2443}
2444
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002445static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002446 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002447 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002448};
2449
2450/*
2451 * Called with mm->mmap_sem held for writing.
2452 * Insert a new vma covering the given region, with the given flags.
2453 * Its pages are supplied by the given array of struct page *.
2454 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2455 * The region past the last page supplied will always produce SIGBUS.
2456 * The array pointer and the pages it points to are assumed to stay alive
2457 * for as long as this mapping might exist.
2458 */
2459int install_special_mapping(struct mm_struct *mm,
2460 unsigned long addr, unsigned long len,
2461 unsigned long vm_flags, struct page **pages)
2462{
2463 struct vm_area_struct *vma;
2464
2465 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2466 if (unlikely(vma == NULL))
2467 return -ENOMEM;
2468
Rik van Riel5beb4932010-03-05 13:42:07 -08002469 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002470 vma->vm_mm = mm;
2471 vma->vm_start = addr;
2472 vma->vm_end = addr + len;
2473
Nick Piggin2f987352008-02-02 03:08:53 +01002474 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002475 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002476
2477 vma->vm_ops = &special_mapping_vmops;
2478 vma->vm_private_data = pages;
2479
2480 if (unlikely(insert_vm_struct(mm, vma))) {
2481 kmem_cache_free(vm_area_cachep, vma);
2482 return -ENOMEM;
2483 }
2484
2485 mm->total_vm += len >> PAGE_SHIFT;
2486
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002487 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002488
Roland McGrathfa5dc222007-02-08 14:20:41 -08002489 return 0;
2490}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002491
2492static DEFINE_MUTEX(mm_all_locks_mutex);
2493
Peter Zijlstra454ed842008-08-11 09:30:25 +02002494static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002495{
Rik van Riel012f1802010-08-09 17:18:40 -07002496 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002497 /*
2498 * The LSB of head.next can't change from under us
2499 * because we hold the mm_all_locks_mutex.
2500 */
Rik van Riel012f1802010-08-09 17:18:40 -07002501 spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002502 /*
2503 * We can safely modify head.next after taking the
Rik van Riel012f1802010-08-09 17:18:40 -07002504 * anon_vma->root->lock. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002505 * the same anon_vma we won't take it again.
2506 *
2507 * No need of atomic instructions here, head.next
2508 * can't change from under us thanks to the
Rik van Riel012f1802010-08-09 17:18:40 -07002509 * anon_vma->root->lock.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002510 */
2511 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002512 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002513 BUG();
2514 }
2515}
2516
Peter Zijlstra454ed842008-08-11 09:30:25 +02002517static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002518{
2519 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2520 /*
2521 * AS_MM_ALL_LOCKS can't change from under us because
2522 * we hold the mm_all_locks_mutex.
2523 *
2524 * Operations on ->flags have to be atomic because
2525 * even if AS_MM_ALL_LOCKS is stable thanks to the
2526 * mm_all_locks_mutex, there may be other cpus
2527 * changing other bitflags in parallel to us.
2528 */
2529 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2530 BUG();
Peter Zijlstra454ed842008-08-11 09:30:25 +02002531 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002532 }
2533}
2534
2535/*
2536 * This operation locks against the VM for all pte/vma/mm related
2537 * operations that could ever happen on a certain mm. This includes
2538 * vmtruncate, try_to_unmap, and all page faults.
2539 *
2540 * The caller must take the mmap_sem in write mode before calling
2541 * mm_take_all_locks(). The caller isn't allowed to release the
2542 * mmap_sem until mm_drop_all_locks() returns.
2543 *
2544 * mmap_sem in write mode is required in order to block all operations
2545 * that could modify pagetables and free pages without need of
2546 * altering the vma layout (for example populate_range() with
2547 * nonlinear vmas). It's also needed in write mode to avoid new
2548 * anon_vmas to be associated with existing vmas.
2549 *
2550 * A single task can't take more than one mm_take_all_locks() in a row
2551 * or it would deadlock.
2552 *
2553 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2554 * mapping->flags avoid to take the same lock twice, if more than one
2555 * vma in this mm is backed by the same anon_vma or address_space.
2556 *
2557 * We can take all the locks in random order because the VM code
2558 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2559 * takes more than one of them in a row. Secondly we're protected
2560 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2561 *
2562 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2563 * that may have to take thousand of locks.
2564 *
2565 * mm_take_all_locks() can fail if it's interrupted by signals.
2566 */
2567int mm_take_all_locks(struct mm_struct *mm)
2568{
2569 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002570 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002571 int ret = -EINTR;
2572
2573 BUG_ON(down_read_trylock(&mm->mmap_sem));
2574
2575 mutex_lock(&mm_all_locks_mutex);
2576
2577 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2578 if (signal_pending(current))
2579 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002580 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002581 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002582 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002583
2584 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2585 if (signal_pending(current))
2586 goto out_unlock;
2587 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002588 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2589 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002590 }
2591
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002592 ret = 0;
2593
2594out_unlock:
2595 if (ret)
2596 mm_drop_all_locks(mm);
2597
2598 return ret;
2599}
2600
2601static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2602{
Rik van Riel012f1802010-08-09 17:18:40 -07002603 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002604 /*
2605 * The LSB of head.next can't change to 0 from under
2606 * us because we hold the mm_all_locks_mutex.
2607 *
2608 * We must however clear the bitflag before unlocking
2609 * the vma so the users using the anon_vma->head will
2610 * never see our bitflag.
2611 *
2612 * No need of atomic instructions here, head.next
2613 * can't change from under us until we release the
Rik van Riel012f1802010-08-09 17:18:40 -07002614 * anon_vma->root->lock.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002615 */
2616 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002617 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002618 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002619 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002620 }
2621}
2622
2623static void vm_unlock_mapping(struct address_space *mapping)
2624{
2625 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2626 /*
2627 * AS_MM_ALL_LOCKS can't change to 0 from under us
2628 * because we hold the mm_all_locks_mutex.
2629 */
2630 spin_unlock(&mapping->i_mmap_lock);
2631 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2632 &mapping->flags))
2633 BUG();
2634 }
2635}
2636
2637/*
2638 * The mmap_sem cannot be released by the caller until
2639 * mm_drop_all_locks() returns.
2640 */
2641void mm_drop_all_locks(struct mm_struct *mm)
2642{
2643 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002644 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002645
2646 BUG_ON(down_read_trylock(&mm->mmap_sem));
2647 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2648
2649 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2650 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002651 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2652 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002653 if (vma->vm_file && vma->vm_file->f_mapping)
2654 vm_unlock_mapping(vma->vm_file->f_mapping);
2655 }
2656
2657 mutex_unlock(&mm_all_locks_mutex);
2658}
David Howells8feae132009-01-08 12:04:47 +00002659
2660/*
2661 * initialise the VMA slab
2662 */
2663void __init mmap_init(void)
2664{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002665 int ret;
2666
2667 ret = percpu_counter_init(&vm_committed_as, 0);
2668 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002669}