blob: ee2298936fe677870ad4979b98bd8d6ada7c3439 [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 */
268 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
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{
391 if (prev) {
392 vma->vm_next = prev->vm_next;
393 prev->vm_next = vma;
394 } else {
395 mm->mmap = vma;
396 if (rb_parent)
397 vma->vm_next = rb_entry(rb_parent,
398 struct vm_area_struct, vm_rb);
399 else
400 vma->vm_next = NULL;
401 }
402}
403
404void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
405 struct rb_node **rb_link, struct rb_node *rb_parent)
406{
407 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
408 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
409}
410
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700411static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
ZhenwenXu48aae422009-01-06 14:40:21 -0800413 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
415 file = vma->vm_file;
416 if (file) {
417 struct address_space *mapping = file->f_mapping;
418
419 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800420 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 if (vma->vm_flags & VM_SHARED)
422 mapping->i_mmap_writable++;
423
424 flush_dcache_mmap_lock(mapping);
425 if (unlikely(vma->vm_flags & VM_NONLINEAR))
426 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
427 else
428 vma_prio_tree_insert(vma, &mapping->i_mmap);
429 flush_dcache_mmap_unlock(mapping);
430 }
431}
432
433static void
434__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
435 struct vm_area_struct *prev, struct rb_node **rb_link,
436 struct rb_node *rb_parent)
437{
438 __vma_link_list(mm, vma, prev, rb_parent);
439 __vma_link_rb(mm, vma, rb_link, rb_parent);
440 __anon_vma_link(vma);
441}
442
443static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
444 struct vm_area_struct *prev, struct rb_node **rb_link,
445 struct rb_node *rb_parent)
446{
447 struct address_space *mapping = NULL;
448
449 if (vma->vm_file)
450 mapping = vma->vm_file->f_mapping;
451
452 if (mapping) {
453 spin_lock(&mapping->i_mmap_lock);
454 vma->vm_truncate_count = mapping->truncate_count;
455 }
456 anon_vma_lock(vma);
457
458 __vma_link(mm, vma, prev, rb_link, rb_parent);
459 __vma_link_file(vma);
460
461 anon_vma_unlock(vma);
462 if (mapping)
463 spin_unlock(&mapping->i_mmap_lock);
464
465 mm->map_count++;
466 validate_mm(mm);
467}
468
469/*
470 * Helper for vma_adjust in the split_vma insert case:
471 * insert vm structure into list and rbtree and anon_vma,
472 * but it has already been inserted into prio_tree earlier.
473 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800474static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
ZhenwenXu48aae422009-01-06 14:40:21 -0800476 struct vm_area_struct *__vma, *prev;
477 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200480 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 __vma_link(mm, vma, prev, rb_link, rb_parent);
482 mm->map_count++;
483}
484
485static inline void
486__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
487 struct vm_area_struct *prev)
488{
489 prev->vm_next = vma->vm_next;
490 rb_erase(&vma->vm_rb, &mm->mm_rb);
491 if (mm->mmap_cache == vma)
492 mm->mmap_cache = prev;
493}
494
495/*
496 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
497 * is already present in an i_mmap tree without adjusting the tree.
498 * The following helper function should be used when such adjustments
499 * are necessary. The "insert" vma (if any) is to be inserted
500 * before we drop the necessary locks.
501 */
502void vma_adjust(struct vm_area_struct *vma, unsigned long start,
503 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
504{
505 struct mm_struct *mm = vma->vm_mm;
506 struct vm_area_struct *next = vma->vm_next;
507 struct vm_area_struct *importer = NULL;
508 struct address_space *mapping = NULL;
509 struct prio_tree_root *root = NULL;
510 struct file *file = vma->vm_file;
511 struct anon_vma *anon_vma = NULL;
512 long adjust_next = 0;
513 int remove_next = 0;
514
515 if (next && !insert) {
516 if (end >= next->vm_end) {
517 /*
518 * vma expands, overlapping all the next, and
519 * perhaps the one after too (mprotect case 6).
520 */
521again: remove_next = 1 + (end > next->vm_end);
522 end = next->vm_end;
523 anon_vma = next->anon_vma;
524 importer = vma;
525 } else if (end > next->vm_start) {
526 /*
527 * vma expands, overlapping part of the next:
528 * mprotect case 5 shifting the boundary up.
529 */
530 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
531 anon_vma = next->anon_vma;
532 importer = vma;
533 } else if (end < vma->vm_end) {
534 /*
535 * vma shrinks, and !insert tells it's not
536 * split_vma inserting another: so it must be
537 * mprotect case 4 shifting the boundary down.
538 */
539 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
540 anon_vma = next->anon_vma;
541 importer = next;
542 }
543 }
544
545 if (file) {
546 mapping = file->f_mapping;
547 if (!(vma->vm_flags & VM_NONLINEAR))
548 root = &mapping->i_mmap;
549 spin_lock(&mapping->i_mmap_lock);
550 if (importer &&
551 vma->vm_truncate_count != next->vm_truncate_count) {
552 /*
553 * unmap_mapping_range might be in progress:
554 * ensure that the expanding vma is rescanned.
555 */
556 importer->vm_truncate_count = 0;
557 }
558 if (insert) {
559 insert->vm_truncate_count = vma->vm_truncate_count;
560 /*
561 * Put into prio_tree now, so instantiated pages
562 * are visible to arm/parisc __flush_dcache_page
563 * throughout; but we cannot insert into address
564 * space until vma start or end is updated.
565 */
566 __vma_link_file(insert);
567 }
568 }
569
570 /*
571 * When changing only vma->vm_end, we don't really need
Lee Schermerhorn252c5f92009-09-21 17:03:40 -0700572 * anon_vma lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 */
Lee Schermerhorn252c5f92009-09-21 17:03:40 -0700574 if (vma->anon_vma && (insert || importer || start != vma->vm_start))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 anon_vma = vma->anon_vma;
576 if (anon_vma) {
577 spin_lock(&anon_vma->lock);
578 /*
579 * Easily overlooked: when mprotect shifts the boundary,
580 * make sure the expanding vma has anon_vma set if the
581 * shrinking vma had, to cover any anon pages imported.
582 */
583 if (importer && !importer->anon_vma) {
584 importer->anon_vma = anon_vma;
585 __anon_vma_link(importer);
586 }
587 }
588
589 if (root) {
590 flush_dcache_mmap_lock(mapping);
591 vma_prio_tree_remove(vma, root);
592 if (adjust_next)
593 vma_prio_tree_remove(next, root);
594 }
595
596 vma->vm_start = start;
597 vma->vm_end = end;
598 vma->vm_pgoff = pgoff;
599 if (adjust_next) {
600 next->vm_start += adjust_next << PAGE_SHIFT;
601 next->vm_pgoff += adjust_next;
602 }
603
604 if (root) {
605 if (adjust_next)
606 vma_prio_tree_insert(next, root);
607 vma_prio_tree_insert(vma, root);
608 flush_dcache_mmap_unlock(mapping);
609 }
610
611 if (remove_next) {
612 /*
613 * vma_merge has merged next into vma, and needs
614 * us to remove next before dropping the locks.
615 */
616 __vma_unlink(mm, next, vma);
617 if (file)
618 __remove_shared_vm_struct(next, file, mapping);
619 if (next->anon_vma)
620 __anon_vma_merge(vma, next);
621 } else if (insert) {
622 /*
623 * split_vma has split insert from vma, and needs
624 * us to insert it before dropping the locks
625 * (it may either follow vma or precede it).
626 */
627 __insert_vm_struct(mm, insert);
628 }
629
630 if (anon_vma)
631 spin_unlock(&anon_vma->lock);
632 if (mapping)
633 spin_unlock(&mapping->i_mmap_lock);
634
635 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700636 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700638 if (next->vm_flags & VM_EXECUTABLE)
639 removed_exe_file_vma(mm);
640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700642 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 kmem_cache_free(vm_area_cachep, next);
644 /*
645 * In mprotect's case 6 (see comments on vma_merge),
646 * we must remove another next too. It would clutter
647 * up the code too much to do both in one go.
648 */
649 if (remove_next == 2) {
650 next = vma->vm_next;
651 goto again;
652 }
653 }
654
655 validate_mm(mm);
656}
657
658/*
659 * If the vma has a ->close operation then the driver probably needs to release
660 * per-vma resources, so we don't attempt to merge those.
661 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662static inline int is_mergeable_vma(struct vm_area_struct *vma,
663 struct file *file, unsigned long vm_flags)
664{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700665 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
666 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 return 0;
668 if (vma->vm_file != file)
669 return 0;
670 if (vma->vm_ops && vma->vm_ops->close)
671 return 0;
672 return 1;
673}
674
675static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
676 struct anon_vma *anon_vma2)
677{
678 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
679}
680
681/*
682 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
683 * in front of (at a lower virtual address and file offset than) the vma.
684 *
685 * We cannot merge two vmas if they have differently assigned (non-NULL)
686 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
687 *
688 * We don't check here for the merged mmap wrapping around the end of pagecache
689 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
690 * wrap, nor mmaps which cover the final page at index -1UL.
691 */
692static int
693can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
694 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
695{
696 if (is_mergeable_vma(vma, file, vm_flags) &&
697 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
698 if (vma->vm_pgoff == vm_pgoff)
699 return 1;
700 }
701 return 0;
702}
703
704/*
705 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
706 * beyond (at a higher virtual address and file offset than) the vma.
707 *
708 * We cannot merge two vmas if they have differently assigned (non-NULL)
709 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
710 */
711static int
712can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
713 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
714{
715 if (is_mergeable_vma(vma, file, vm_flags) &&
716 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
717 pgoff_t vm_pglen;
718 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
719 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
720 return 1;
721 }
722 return 0;
723}
724
725/*
726 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
727 * whether that can be merged with its predecessor or its successor.
728 * Or both (it neatly fills a hole).
729 *
730 * In most cases - when called for mmap, brk or mremap - [addr,end) is
731 * certain not to be mapped by the time vma_merge is called; but when
732 * called for mprotect, it is certain to be already mapped (either at
733 * an offset within prev, or at the start of next), and the flags of
734 * this area are about to be changed to vm_flags - and the no-change
735 * case has already been eliminated.
736 *
737 * The following mprotect cases have to be considered, where AAAA is
738 * the area passed down from mprotect_fixup, never extending beyond one
739 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
740 *
741 * AAAA AAAA AAAA AAAA
742 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
743 * cannot merge might become might become might become
744 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
745 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
746 * mremap move: PPPPNNNNNNNN 8
747 * AAAA
748 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
749 * might become case 1 below case 2 below case 3 below
750 *
751 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
752 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
753 */
754struct vm_area_struct *vma_merge(struct mm_struct *mm,
755 struct vm_area_struct *prev, unsigned long addr,
756 unsigned long end, unsigned long vm_flags,
757 struct anon_vma *anon_vma, struct file *file,
758 pgoff_t pgoff, struct mempolicy *policy)
759{
760 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
761 struct vm_area_struct *area, *next;
762
763 /*
764 * We later require that vma->vm_flags == vm_flags,
765 * so this tests vma->vm_flags & VM_SPECIAL, too.
766 */
767 if (vm_flags & VM_SPECIAL)
768 return NULL;
769
770 if (prev)
771 next = prev->vm_next;
772 else
773 next = mm->mmap;
774 area = next;
775 if (next && next->vm_end == end) /* cases 6, 7, 8 */
776 next = next->vm_next;
777
778 /*
779 * Can it merge with the predecessor?
780 */
781 if (prev && prev->vm_end == addr &&
782 mpol_equal(vma_policy(prev), policy) &&
783 can_vma_merge_after(prev, vm_flags,
784 anon_vma, file, pgoff)) {
785 /*
786 * OK, it can. Can we now merge in the successor as well?
787 */
788 if (next && end == next->vm_start &&
789 mpol_equal(policy, vma_policy(next)) &&
790 can_vma_merge_before(next, vm_flags,
791 anon_vma, file, pgoff+pglen) &&
792 is_mergeable_anon_vma(prev->anon_vma,
793 next->anon_vma)) {
794 /* cases 1, 6 */
795 vma_adjust(prev, prev->vm_start,
796 next->vm_end, prev->vm_pgoff, NULL);
797 } else /* cases 2, 5, 7 */
798 vma_adjust(prev, prev->vm_start,
799 end, prev->vm_pgoff, NULL);
800 return prev;
801 }
802
803 /*
804 * Can this new request be merged in front of next?
805 */
806 if (next && end == next->vm_start &&
807 mpol_equal(policy, vma_policy(next)) &&
808 can_vma_merge_before(next, vm_flags,
809 anon_vma, file, pgoff+pglen)) {
810 if (prev && addr < prev->vm_end) /* case 4 */
811 vma_adjust(prev, prev->vm_start,
812 addr, prev->vm_pgoff, NULL);
813 else /* cases 3, 8 */
814 vma_adjust(area, addr, next->vm_end,
815 next->vm_pgoff - pglen, NULL);
816 return area;
817 }
818
819 return NULL;
820}
821
822/*
823 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
824 * neighbouring vmas for a suitable anon_vma, before it goes off
825 * to allocate a new anon_vma. It checks because a repetitive
826 * sequence of mprotects and faults may otherwise lead to distinct
827 * anon_vmas being allocated, preventing vma merge in subsequent
828 * mprotect.
829 */
830struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
831{
832 struct vm_area_struct *near;
833 unsigned long vm_flags;
834
835 near = vma->vm_next;
836 if (!near)
837 goto try_prev;
838
839 /*
840 * Since only mprotect tries to remerge vmas, match flags
841 * which might be mprotected into each other later on.
842 * Neither mlock nor madvise tries to remerge at present,
843 * so leave their flags as obstructing a merge.
844 */
845 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
846 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
847
848 if (near->anon_vma && vma->vm_end == near->vm_start &&
849 mpol_equal(vma_policy(vma), vma_policy(near)) &&
850 can_vma_merge_before(near, vm_flags,
851 NULL, vma->vm_file, vma->vm_pgoff +
852 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
853 return near->anon_vma;
854try_prev:
855 /*
856 * It is potentially slow to have to call find_vma_prev here.
857 * But it's only on the first write fault on the vma, not
858 * every time, and we could devise a way to avoid it later
859 * (e.g. stash info in next's anon_vma_node when assigning
860 * an anon_vma, or when trying vma_merge). Another time.
861 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200862 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (!near)
864 goto none;
865
866 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
867 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
868
869 if (near->anon_vma && near->vm_end == vma->vm_start &&
870 mpol_equal(vma_policy(near), vma_policy(vma)) &&
871 can_vma_merge_after(near, vm_flags,
872 NULL, vma->vm_file, vma->vm_pgoff))
873 return near->anon_vma;
874none:
875 /*
876 * There's no absolute need to look only at touching neighbours:
877 * we could search further afield for "compatible" anon_vmas.
878 * But it would probably just be a waste of time searching,
879 * or lead to too many vmas hanging off the same anon_vma.
880 * We're trying to allow mprotect remerging later on,
881 * not trying to minimize memory used for anon_vmas.
882 */
883 return NULL;
884}
885
886#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700887void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 struct file *file, long pages)
889{
890 const unsigned long stack_flags
891 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
892
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (file) {
894 mm->shared_vm += pages;
895 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
896 mm->exec_vm += pages;
897 } else if (flags & stack_flags)
898 mm->stack_vm += pages;
899 if (flags & (VM_RESERVED|VM_IO))
900 mm->reserved_vm += pages;
901}
902#endif /* CONFIG_PROC_FS */
903
904/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700905 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 */
907
ZhenwenXu48aae422009-01-06 14:40:21 -0800908unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 unsigned long len, unsigned long prot,
910 unsigned long flags, unsigned long pgoff)
911{
912 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 struct inode *inode;
914 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700916 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 /*
919 * Does the application expect PROT_READ to imply PROT_EXEC?
920 *
921 * (the exception is when the underlying filesystem is noexec
922 * mounted, in which case we dont add PROT_EXEC.)
923 */
924 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800925 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 prot |= PROT_EXEC;
927
928 if (!len)
929 return -EINVAL;
930
Eric Paris7cd94142007-11-26 18:47:40 -0500931 if (!(flags & MAP_FIXED))
932 addr = round_hint_to_min(addr);
933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 /* Careful about overflows.. */
935 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500936 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return -ENOMEM;
938
939 /* offset overflow? */
940 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
941 return -EOVERFLOW;
942
943 /* Too many mappings? */
944 if (mm->map_count > sysctl_max_map_count)
945 return -ENOMEM;
946
947 /* Obtain the address to map to. we verify (or select) it and ensure
948 * that it represents a valid section of the address space.
949 */
950 addr = get_unmapped_area(file, addr, len, pgoff, flags);
951 if (addr & ~PAGE_MASK)
952 return addr;
953
954 /* Do simple checking here so the lower-level routines won't have
955 * to. we assume access permissions have been handled by the open
956 * of the memory object, so we don't do any here.
957 */
958 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
959 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
960
Huang Shijiecdf7b342009-09-21 17:03:36 -0700961 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 if (!can_do_mlock())
963 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -0700964
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 /* mlock MCL_FUTURE? */
966 if (vm_flags & VM_LOCKED) {
967 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700968 locked = len >> PAGE_SHIFT;
969 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700971 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
973 return -EAGAIN;
974 }
975
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800976 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
978 if (file) {
979 switch (flags & MAP_TYPE) {
980 case MAP_SHARED:
981 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
982 return -EACCES;
983
984 /*
985 * Make sure we don't allow writing to an append-only
986 * file..
987 */
988 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
989 return -EACCES;
990
991 /*
992 * Make sure there are no mandatory locks on the file.
993 */
994 if (locks_verify_locked(inode))
995 return -EAGAIN;
996
997 vm_flags |= VM_SHARED | VM_MAYSHARE;
998 if (!(file->f_mode & FMODE_WRITE))
999 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1000
1001 /* fall through */
1002 case MAP_PRIVATE:
1003 if (!(file->f_mode & FMODE_READ))
1004 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001005 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001006 if (vm_flags & VM_EXEC)
1007 return -EPERM;
1008 vm_flags &= ~VM_MAYEXEC;
1009 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001010
1011 if (!file->f_op || !file->f_op->mmap)
1012 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 break;
1014
1015 default:
1016 return -EINVAL;
1017 }
1018 } else {
1019 switch (flags & MAP_TYPE) {
1020 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001021 /*
1022 * Ignore pgoff.
1023 */
1024 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 vm_flags |= VM_SHARED | VM_MAYSHARE;
1026 break;
1027 case MAP_PRIVATE:
1028 /*
1029 * Set pgoff according to addr for anon_vma.
1030 */
1031 pgoff = addr >> PAGE_SHIFT;
1032 break;
1033 default:
1034 return -EINVAL;
1035 }
1036 }
1037
Eric Parised032182007-06-28 15:55:21 -04001038 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 if (error)
1040 return error;
Eric Parised032182007-06-28 15:55:21 -04001041
Mel Gorman5a6fe122009-02-10 14:02:27 +00001042 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001043}
1044EXPORT_SYMBOL(do_mmap_pgoff);
1045
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001046SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1047 unsigned long, prot, unsigned long, flags,
1048 unsigned long, fd, unsigned long, pgoff)
1049{
1050 struct file *file = NULL;
1051 unsigned long retval = -EBADF;
1052
1053 if (!(flags & MAP_ANONYMOUS)) {
1054 if (unlikely(flags & MAP_HUGETLB))
1055 return -EINVAL;
1056 file = fget(fd);
1057 if (!file)
1058 goto out;
1059 } else if (flags & MAP_HUGETLB) {
1060 struct user_struct *user = NULL;
1061 /*
1062 * VM_NORESERVE is used because the reservations will be
1063 * taken when vm_ops->mmap() is called
1064 * A dummy user value is used because we are not locking
1065 * memory so no accounting is necessary
1066 */
1067 len = ALIGN(len, huge_page_size(&default_hstate));
1068 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1069 &user, HUGETLB_ANONHUGE_INODE);
1070 if (IS_ERR(file))
1071 return PTR_ERR(file);
1072 }
1073
1074 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1075
1076 down_write(&current->mm->mmap_sem);
1077 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1078 up_write(&current->mm->mmap_sem);
1079
1080 if (file)
1081 fput(file);
1082out:
1083 return retval;
1084}
1085
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001086/*
1087 * Some shared mappigns will want the pages marked read-only
1088 * to track write events. If so, we'll downgrade vm_page_prot
1089 * to the private version (using protection_map[] without the
1090 * VM_SHARED bit).
1091 */
1092int vma_wants_writenotify(struct vm_area_struct *vma)
1093{
1094 unsigned int vm_flags = vma->vm_flags;
1095
1096 /* If it was private or non-writable, the write bit is already clear */
1097 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1098 return 0;
1099
1100 /* The backer wishes to know when pages are first written to? */
1101 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1102 return 1;
1103
1104 /* The open routine did something to the protections already? */
1105 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001106 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001107 return 0;
1108
1109 /* Specialty mapping? */
1110 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1111 return 0;
1112
1113 /* Can the mapping track the dirty pages? */
1114 return vma->vm_file && vma->vm_file->f_mapping &&
1115 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1116}
1117
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001118/*
1119 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001120 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001121 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001122static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001123{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001124 /*
1125 * hugetlb has its own accounting separate from the core VM
1126 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1127 */
1128 if (file && is_file_hugepages(file))
1129 return 0;
1130
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001131 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1132}
1133
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001134unsigned long mmap_region(struct file *file, unsigned long addr,
1135 unsigned long len, unsigned long flags,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001136 unsigned int vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001137{
1138 struct mm_struct *mm = current->mm;
1139 struct vm_area_struct *vma, *prev;
1140 int correct_wcount = 0;
1141 int error;
1142 struct rb_node **rb_link, *rb_parent;
1143 unsigned long charged = 0;
1144 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 /* Clear old maps */
1147 error = -ENOMEM;
1148munmap_back:
1149 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1150 if (vma && vma->vm_start < addr + len) {
1151 if (do_munmap(mm, addr, len))
1152 return -ENOMEM;
1153 goto munmap_back;
1154 }
1155
1156 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001157 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 return -ENOMEM;
1159
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001160 /*
1161 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001162 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001163 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001164 if ((flags & MAP_NORESERVE)) {
1165 /* We honor MAP_NORESERVE if allowed to overcommit */
1166 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1167 vm_flags |= VM_NORESERVE;
1168
1169 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1170 if (file && is_file_hugepages(file))
1171 vm_flags |= VM_NORESERVE;
1172 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001173
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001174 /*
1175 * Private writable mapping: check memory availability
1176 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001177 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001178 charged = len >> PAGE_SHIFT;
1179 if (security_vm_enough_memory(charged))
1180 return -ENOMEM;
1181 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
1183
1184 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001185 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001187 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1188 if (vma)
1189 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 /*
1192 * Determine the object being mapped and call the appropriate
1193 * specific mapper. the address has already been validated, but
1194 * not unmapped, but the maps are removed from the list.
1195 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001196 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 if (!vma) {
1198 error = -ENOMEM;
1199 goto unacct_error;
1200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 vma->vm_mm = mm;
1203 vma->vm_start = addr;
1204 vma->vm_end = addr + len;
1205 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001206 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 vma->vm_pgoff = pgoff;
1208
1209 if (file) {
1210 error = -EINVAL;
1211 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1212 goto free_vma;
1213 if (vm_flags & VM_DENYWRITE) {
1214 error = deny_write_access(file);
1215 if (error)
1216 goto free_vma;
1217 correct_wcount = 1;
1218 }
1219 vma->vm_file = file;
1220 get_file(file);
1221 error = file->f_op->mmap(file, vma);
1222 if (error)
1223 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001224 if (vm_flags & VM_EXECUTABLE)
1225 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001226
1227 /* Can addr have changed??
1228 *
1229 * Answer: Yes, several device drivers can do it in their
1230 * f_op->mmap method. -DaveM
1231 */
1232 addr = vma->vm_start;
1233 pgoff = vma->vm_pgoff;
1234 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 } else if (vm_flags & VM_SHARED) {
1236 error = shmem_zero_setup(vma);
1237 if (error)
1238 goto free_vma;
1239 }
1240
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001241 if (vma_wants_writenotify(vma)) {
1242 pgprot_t pprot = vma->vm_page_prot;
1243
1244 /* Can vma->vm_page_prot have changed??
1245 *
1246 * Answer: Yes, drivers may have changed it in their
1247 * f_op->mmap method.
1248 *
1249 * Ensures that vmas marked as uncached stay that way.
1250 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001251 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001252 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1253 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1254 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001255
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001256 vma_link(mm, vma, prev, rb_link, rb_parent);
1257 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001258
1259 /* Once vma denies write, undo our temporary denial count */
1260 if (correct_wcount)
1261 atomic_inc(&inode->i_writecount);
1262out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001263 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001264
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001266 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (vm_flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07001268 /*
1269 * makes pages present; downgrades, drops, reacquires mmap_sem
1270 */
1271 long nr_pages = mlock_vma_pages_range(vma, addr, addr + len);
1272 if (nr_pages < 0)
1273 return nr_pages; /* vma gone! */
1274 mm->locked_vm += (len >> PAGE_SHIFT) - nr_pages;
1275 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001276 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 return addr;
1278
1279unmap_and_free_vma:
1280 if (correct_wcount)
1281 atomic_inc(&inode->i_writecount);
1282 vma->vm_file = NULL;
1283 fput(file);
1284
1285 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001286 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1287 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288free_vma:
1289 kmem_cache_free(vm_area_cachep, vma);
1290unacct_error:
1291 if (charged)
1292 vm_unacct_memory(charged);
1293 return error;
1294}
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296/* Get an address range which is currently unmapped.
1297 * For shmat() with addr=0.
1298 *
1299 * Ugly calling convention alert:
1300 * Return value with the low bits set means error value,
1301 * ie
1302 * if (ret & ~PAGE_MASK)
1303 * error = ret;
1304 *
1305 * This function "knows" that -ENOMEM has the bits set.
1306 */
1307#ifndef HAVE_ARCH_UNMAPPED_AREA
1308unsigned long
1309arch_get_unmapped_area(struct file *filp, unsigned long addr,
1310 unsigned long len, unsigned long pgoff, unsigned long flags)
1311{
1312 struct mm_struct *mm = current->mm;
1313 struct vm_area_struct *vma;
1314 unsigned long start_addr;
1315
1316 if (len > TASK_SIZE)
1317 return -ENOMEM;
1318
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001319 if (flags & MAP_FIXED)
1320 return addr;
1321
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 if (addr) {
1323 addr = PAGE_ALIGN(addr);
1324 vma = find_vma(mm, addr);
1325 if (TASK_SIZE - len >= addr &&
1326 (!vma || addr + len <= vma->vm_start))
1327 return addr;
1328 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001329 if (len > mm->cached_hole_size) {
1330 start_addr = addr = mm->free_area_cache;
1331 } else {
1332 start_addr = addr = TASK_UNMAPPED_BASE;
1333 mm->cached_hole_size = 0;
1334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
1336full_search:
1337 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1338 /* At this point: (!vma || addr < vma->vm_end). */
1339 if (TASK_SIZE - len < addr) {
1340 /*
1341 * Start a new search - just in case we missed
1342 * some holes.
1343 */
1344 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001345 addr = TASK_UNMAPPED_BASE;
1346 start_addr = addr;
1347 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 goto full_search;
1349 }
1350 return -ENOMEM;
1351 }
1352 if (!vma || addr + len <= vma->vm_start) {
1353 /*
1354 * Remember the place where we stopped the search:
1355 */
1356 mm->free_area_cache = addr + len;
1357 return addr;
1358 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001359 if (addr + mm->cached_hole_size < vma->vm_start)
1360 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 addr = vma->vm_end;
1362 }
1363}
1364#endif
1365
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001366void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
1368 /*
1369 * Is this a new hole at the lowest possible address?
1370 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001371 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1372 mm->free_area_cache = addr;
1373 mm->cached_hole_size = ~0UL;
1374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
1377/*
1378 * This mmap-allocator allocates new areas top-down from below the
1379 * stack's low limit (the base):
1380 */
1381#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1382unsigned long
1383arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1384 const unsigned long len, const unsigned long pgoff,
1385 const unsigned long flags)
1386{
1387 struct vm_area_struct *vma;
1388 struct mm_struct *mm = current->mm;
1389 unsigned long addr = addr0;
1390
1391 /* requested length too big for entire address space */
1392 if (len > TASK_SIZE)
1393 return -ENOMEM;
1394
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001395 if (flags & MAP_FIXED)
1396 return addr;
1397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 /* requesting a specific address */
1399 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 }
1406
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001407 /* check if free_area_cache is useful for us */
1408 if (len <= mm->cached_hole_size) {
1409 mm->cached_hole_size = 0;
1410 mm->free_area_cache = mm->mmap_base;
1411 }
1412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 /* either no address requested or can't fit in requested address hole */
1414 addr = mm->free_area_cache;
1415
1416 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001417 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 vma = find_vma(mm, addr-len);
1419 if (!vma || addr <= vma->vm_start)
1420 /* remember the address as a hint for next time */
1421 return (mm->free_area_cache = addr-len);
1422 }
1423
Chris Wright73219d12005-06-21 17:14:52 -07001424 if (mm->mmap_base < len)
1425 goto bottomup;
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 addr = mm->mmap_base-len;
1428
1429 do {
1430 /*
1431 * Lookup failure means no vma is above this address,
1432 * else if new region fits below vma->vm_start,
1433 * return with success:
1434 */
1435 vma = find_vma(mm, addr);
1436 if (!vma || addr+len <= vma->vm_start)
1437 /* remember the address as a hint for next time */
1438 return (mm->free_area_cache = addr);
1439
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001440 /* remember the largest hole we saw so far */
1441 if (addr + mm->cached_hole_size < vma->vm_start)
1442 mm->cached_hole_size = vma->vm_start - addr;
1443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 /* try just below the current vma->vm_start */
1445 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001446 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Chris Wright73219d12005-06-21 17:14:52 -07001448bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 /*
1450 * A failed mmap() very likely causes application failure,
1451 * so fall back to the bottom-up function here. This scenario
1452 * can happen with large stack limits and large mmap()
1453 * allocations.
1454 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001455 mm->cached_hole_size = ~0UL;
1456 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1458 /*
1459 * Restore the topdown base:
1460 */
1461 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001462 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
1464 return addr;
1465}
1466#endif
1467
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001468void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
1470 /*
1471 * Is this a new hole at the highest possible address?
1472 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001473 if (addr > mm->free_area_cache)
1474 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001477 if (mm->free_area_cache > mm->mmap_base)
1478 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479}
1480
1481unsigned long
1482get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1483 unsigned long pgoff, unsigned long flags)
1484{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001485 unsigned long (*get_area)(struct file *, unsigned long,
1486 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Al Viro9206de92009-12-03 15:23:11 -05001488 unsigned long error = arch_mmap_check(addr, len, flags);
1489 if (error)
1490 return error;
1491
1492 /* Careful about overflows.. */
1493 if (len > TASK_SIZE)
1494 return -ENOMEM;
1495
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001496 get_area = current->mm->get_unmapped_area;
1497 if (file && file->f_op && file->f_op->get_unmapped_area)
1498 get_area = file->f_op->get_unmapped_area;
1499 addr = get_area(file, addr, len, pgoff, flags);
1500 if (IS_ERR_VALUE(addr))
1501 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001503 if (addr > TASK_SIZE - len)
1504 return -ENOMEM;
1505 if (addr & ~PAGE_MASK)
1506 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001507
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001508 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509}
1510
1511EXPORT_SYMBOL(get_unmapped_area);
1512
1513/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001514struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515{
1516 struct vm_area_struct *vma = NULL;
1517
1518 if (mm) {
1519 /* Check the cache first. */
1520 /* (Cache hit rate is typically around 35%.) */
1521 vma = mm->mmap_cache;
1522 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1523 struct rb_node * rb_node;
1524
1525 rb_node = mm->mm_rb.rb_node;
1526 vma = NULL;
1527
1528 while (rb_node) {
1529 struct vm_area_struct * vma_tmp;
1530
1531 vma_tmp = rb_entry(rb_node,
1532 struct vm_area_struct, vm_rb);
1533
1534 if (vma_tmp->vm_end > addr) {
1535 vma = vma_tmp;
1536 if (vma_tmp->vm_start <= addr)
1537 break;
1538 rb_node = rb_node->rb_left;
1539 } else
1540 rb_node = rb_node->rb_right;
1541 }
1542 if (vma)
1543 mm->mmap_cache = vma;
1544 }
1545 }
1546 return vma;
1547}
1548
1549EXPORT_SYMBOL(find_vma);
1550
1551/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1552struct vm_area_struct *
1553find_vma_prev(struct mm_struct *mm, unsigned long addr,
1554 struct vm_area_struct **pprev)
1555{
1556 struct vm_area_struct *vma = NULL, *prev = NULL;
ZhenwenXu48aae422009-01-06 14:40:21 -08001557 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (!mm)
1559 goto out;
1560
1561 /* Guard against addr being lower than the first VMA */
1562 vma = mm->mmap;
1563
1564 /* Go through the RB tree quickly. */
1565 rb_node = mm->mm_rb.rb_node;
1566
1567 while (rb_node) {
1568 struct vm_area_struct *vma_tmp;
1569 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1570
1571 if (addr < vma_tmp->vm_end) {
1572 rb_node = rb_node->rb_left;
1573 } else {
1574 prev = vma_tmp;
1575 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1576 break;
1577 rb_node = rb_node->rb_right;
1578 }
1579 }
1580
1581out:
1582 *pprev = prev;
1583 return prev ? prev->vm_next : vma;
1584}
1585
1586/*
1587 * Verify that the stack growth is acceptable and
1588 * update accounting. This is shared with both the
1589 * grow-up and grow-down cases.
1590 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001591static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592{
1593 struct mm_struct *mm = vma->vm_mm;
1594 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001595 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001598 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 return -ENOMEM;
1600
1601 /* Stack limit test */
1602 if (size > rlim[RLIMIT_STACK].rlim_cur)
1603 return -ENOMEM;
1604
1605 /* mlock limit tests */
1606 if (vma->vm_flags & VM_LOCKED) {
1607 unsigned long locked;
1608 unsigned long limit;
1609 locked = mm->locked_vm + grow;
1610 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1611 if (locked > limit && !capable(CAP_IPC_LOCK))
1612 return -ENOMEM;
1613 }
1614
Adam Litke0d59a012007-01-30 14:35:39 -08001615 /* Check to ensure the stack will not grow into a hugetlb-only region */
1616 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1617 vma->vm_end - size;
1618 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1619 return -EFAULT;
1620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 /*
1622 * Overcommit.. This must be the final test, as it will
1623 * update security statistics.
1624 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001625 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 return -ENOMEM;
1627
1628 /* Ok, everything looks good - let it rip */
1629 mm->total_vm += grow;
1630 if (vma->vm_flags & VM_LOCKED)
1631 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001632 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 return 0;
1634}
1635
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001636#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001638 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1639 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001641#ifndef CONFIG_IA64
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001642static
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001643#endif
1644int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
1646 int error;
1647
1648 if (!(vma->vm_flags & VM_GROWSUP))
1649 return -EFAULT;
1650
1651 /*
1652 * We must make sure the anon_vma is allocated
1653 * so that the anon_vma locking is not a noop.
1654 */
1655 if (unlikely(anon_vma_prepare(vma)))
1656 return -ENOMEM;
1657 anon_vma_lock(vma);
1658
1659 /*
1660 * vma->vm_start/vm_end cannot change under us because the caller
1661 * is required to hold the mmap_sem in read mode. We need the
1662 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001663 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 */
Helge Deller06b32f32006-12-19 19:28:33 +01001665 if (address < PAGE_ALIGN(address+4))
1666 address = PAGE_ALIGN(address+4);
1667 else {
1668 anon_vma_unlock(vma);
1669 return -ENOMEM;
1670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 error = 0;
1672
1673 /* Somebody else might have raced and expanded it already */
1674 if (address > vma->vm_end) {
1675 unsigned long size, grow;
1676
1677 size = address - vma->vm_start;
1678 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1679
1680 error = acct_stack_growth(vma, size, grow);
1681 if (!error)
1682 vma->vm_end = address;
1683 }
1684 anon_vma_unlock(vma);
1685 return error;
1686}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001687#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689/*
1690 * vma is the first one with address < vma->vm_start. Have to extend vma.
1691 */
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001692static int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001693 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694{
1695 int error;
1696
1697 /*
1698 * We must make sure the anon_vma is allocated
1699 * so that the anon_vma locking is not a noop.
1700 */
1701 if (unlikely(anon_vma_prepare(vma)))
1702 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001703
1704 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001705 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001706 if (error)
1707 return error;
1708
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 anon_vma_lock(vma);
1710
1711 /*
1712 * vma->vm_start/vm_end cannot change under us because the caller
1713 * is required to hold the mmap_sem in read mode. We need the
1714 * anon_vma lock to serialize against concurrent expand_stacks.
1715 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716
1717 /* Somebody else might have raced and expanded it already */
1718 if (address < vma->vm_start) {
1719 unsigned long size, grow;
1720
1721 size = vma->vm_end - address;
1722 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1723
1724 error = acct_stack_growth(vma, size, grow);
1725 if (!error) {
1726 vma->vm_start = address;
1727 vma->vm_pgoff -= grow;
1728 }
1729 }
1730 anon_vma_unlock(vma);
1731 return error;
1732}
1733
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001734int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1735{
1736 return expand_downwards(vma, address);
1737}
1738
1739#ifdef CONFIG_STACK_GROWSUP
1740int expand_stack(struct vm_area_struct *vma, unsigned long address)
1741{
1742 return expand_upwards(vma, address);
1743}
1744
1745struct vm_area_struct *
1746find_extend_vma(struct mm_struct *mm, unsigned long addr)
1747{
1748 struct vm_area_struct *vma, *prev;
1749
1750 addr &= PAGE_MASK;
1751 vma = find_vma_prev(mm, addr, &prev);
1752 if (vma && (vma->vm_start <= addr))
1753 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001754 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001755 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001756 if (prev->vm_flags & VM_LOCKED) {
1757 if (mlock_vma_pages_range(prev, addr, prev->vm_end) < 0)
1758 return NULL; /* vma gone! */
1759 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001760 return prev;
1761}
1762#else
1763int expand_stack(struct vm_area_struct *vma, unsigned long address)
1764{
1765 return expand_downwards(vma, address);
1766}
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768struct vm_area_struct *
1769find_extend_vma(struct mm_struct * mm, unsigned long addr)
1770{
1771 struct vm_area_struct * vma;
1772 unsigned long start;
1773
1774 addr &= PAGE_MASK;
1775 vma = find_vma(mm,addr);
1776 if (!vma)
1777 return NULL;
1778 if (vma->vm_start <= addr)
1779 return vma;
1780 if (!(vma->vm_flags & VM_GROWSDOWN))
1781 return NULL;
1782 start = vma->vm_start;
1783 if (expand_stack(vma, addr))
1784 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001785 if (vma->vm_flags & VM_LOCKED) {
1786 if (mlock_vma_pages_range(vma, addr, start) < 0)
1787 return NULL; /* vma gone! */
1788 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 return vma;
1790}
1791#endif
1792
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001794 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001796 *
1797 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001799static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001801 /* Update high watermark before we lower total_vm */
1802 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001804 long nrpages = vma_pages(vma);
1805
1806 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001807 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001808 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001809 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 validate_mm(mm);
1811}
1812
1813/*
1814 * Get rid of page table information in the indicated region.
1815 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001816 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 */
1818static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001819 struct vm_area_struct *vma, struct vm_area_struct *prev,
1820 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001822 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 struct mmu_gather *tlb;
1824 unsigned long nr_accounted = 0;
1825
1826 lru_add_drain();
1827 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001828 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001829 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 vm_unacct_memory(nr_accounted);
Jan Beulich42b77722008-07-23 21:27:10 -07001831 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001832 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 tlb_finish_mmu(tlb, start, end);
1834}
1835
1836/*
1837 * Create a list of vma's touched by the unmap, removing them from the mm's
1838 * vma list as we go..
1839 */
1840static void
1841detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1842 struct vm_area_struct *prev, unsigned long end)
1843{
1844 struct vm_area_struct **insertion_point;
1845 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001846 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1849 do {
1850 rb_erase(&vma->vm_rb, &mm->mm_rb);
1851 mm->map_count--;
1852 tail_vma = vma;
1853 vma = vma->vm_next;
1854 } while (vma && vma->vm_start < end);
1855 *insertion_point = vma;
1856 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001857 if (mm->unmap_area == arch_unmap_area)
1858 addr = prev ? prev->vm_end : mm->mmap_base;
1859 else
1860 addr = vma ? vma->vm_start : mm->mmap_base;
1861 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 mm->mmap_cache = NULL; /* Kill the cache. */
1863}
1864
1865/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001866 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1867 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001869static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 unsigned long addr, int new_below)
1871{
1872 struct mempolicy *pol;
1873 struct vm_area_struct *new;
1874
Andi Kleena5516432008-07-23 21:27:41 -07001875 if (is_vm_hugetlb_page(vma) && (addr &
1876 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 return -EINVAL;
1878
Christoph Lametere94b1762006-12-06 20:33:17 -08001879 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 if (!new)
1881 return -ENOMEM;
1882
1883 /* most fields are the same, copy all, and then fixup */
1884 *new = *vma;
1885
1886 if (new_below)
1887 new->vm_end = addr;
1888 else {
1889 new->vm_start = addr;
1890 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1891 }
1892
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001893 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 if (IS_ERR(pol)) {
1895 kmem_cache_free(vm_area_cachep, new);
1896 return PTR_ERR(pol);
1897 }
1898 vma_set_policy(new, pol);
1899
Matt Helsley925d1c42008-04-29 01:01:36 -07001900 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001902 if (vma->vm_flags & VM_EXECUTABLE)
1903 added_exe_file_vma(mm);
1904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
1906 if (new->vm_ops && new->vm_ops->open)
1907 new->vm_ops->open(new);
1908
1909 if (new_below)
1910 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1911 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1912 else
1913 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1914
1915 return 0;
1916}
1917
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001918/*
1919 * Split a vma into two pieces at address 'addr', a new vma is allocated
1920 * either for the first part or the tail.
1921 */
1922int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1923 unsigned long addr, int new_below)
1924{
1925 if (mm->map_count >= sysctl_max_map_count)
1926 return -ENOMEM;
1927
1928 return __split_vma(mm, vma, addr, new_below);
1929}
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931/* Munmap is split into 2 main parts -- this part which finds
1932 * what needs doing, and the areas themselves, which do the
1933 * work. This now handles partial unmappings.
1934 * Jeremy Fitzhardinge <jeremy@goop.org>
1935 */
1936int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1937{
1938 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001939 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
1941 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1942 return -EINVAL;
1943
1944 if ((len = PAGE_ALIGN(len)) == 0)
1945 return -EINVAL;
1946
1947 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001948 vma = find_vma_prev(mm, start, &prev);
1949 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001951 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /* if it doesn't overlap, we have nothing.. */
1954 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001955 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 return 0;
1957
1958 /*
1959 * If we need to split any vma, do it now to save pain later.
1960 *
1961 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1962 * unmapped vm_area_struct will remain in use: so lower split_vma
1963 * places tmp vma above, and higher split_vma places tmp vma below.
1964 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001965 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001966 int error;
1967
1968 /*
1969 * Make sure that map_count on return from munmap() will
1970 * not exceed its limit; but let map_count go just above
1971 * its limit temporarily, to help free resources as expected.
1972 */
1973 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
1974 return -ENOMEM;
1975
1976 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 if (error)
1978 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001979 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 }
1981
1982 /* Does it split the last one? */
1983 last = find_vma(mm, end);
1984 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001985 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 if (error)
1987 return error;
1988 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001989 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
1991 /*
Rik van Rielba470de2008-10-18 20:26:50 -07001992 * unlock any mlock()ed ranges before detaching vmas
1993 */
1994 if (mm->locked_vm) {
1995 struct vm_area_struct *tmp = vma;
1996 while (tmp && tmp->vm_start < end) {
1997 if (tmp->vm_flags & VM_LOCKED) {
1998 mm->locked_vm -= vma_pages(tmp);
1999 munlock_vma_pages_all(tmp);
2000 }
2001 tmp = tmp->vm_next;
2002 }
2003 }
2004
2005 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 * Remove the vma's, and unmap the actual pages
2007 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002008 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2009 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
2011 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002012 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014 return 0;
2015}
2016
2017EXPORT_SYMBOL(do_munmap);
2018
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002019SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020{
2021 int ret;
2022 struct mm_struct *mm = current->mm;
2023
2024 profile_munmap(addr);
2025
2026 down_write(&mm->mmap_sem);
2027 ret = do_munmap(mm, addr, len);
2028 up_write(&mm->mmap_sem);
2029 return ret;
2030}
2031
2032static inline void verify_mm_writelocked(struct mm_struct *mm)
2033{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002034#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2036 WARN_ON(1);
2037 up_read(&mm->mmap_sem);
2038 }
2039#endif
2040}
2041
2042/*
2043 * this is really a simplified "do_mmap". it only handles
2044 * anonymous maps. eventually we may be able to do some
2045 * brk-specific accounting here.
2046 */
2047unsigned long do_brk(unsigned long addr, unsigned long len)
2048{
2049 struct mm_struct * mm = current->mm;
2050 struct vm_area_struct * vma, * prev;
2051 unsigned long flags;
2052 struct rb_node ** rb_link, * rb_parent;
2053 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002054 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
2056 len = PAGE_ALIGN(len);
2057 if (!len)
2058 return addr;
2059
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002060 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002061 if (error)
2062 return error;
2063
Kirill Korotaev3a459752006-09-07 14:17:04 +04002064 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2065
Al Viro2c6a10162009-12-03 19:40:46 -05002066 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2067 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002068 return error;
2069
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 /*
2071 * mlock MCL_FUTURE?
2072 */
2073 if (mm->def_flags & VM_LOCKED) {
2074 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002075 locked = len >> PAGE_SHIFT;
2076 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07002078 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2080 return -EAGAIN;
2081 }
2082
2083 /*
2084 * mm->mmap_sem is required to protect against another thread
2085 * changing the mappings in case we sleep.
2086 */
2087 verify_mm_writelocked(mm);
2088
2089 /*
2090 * Clear old maps. this also does some error checking for us
2091 */
2092 munmap_back:
2093 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2094 if (vma && vma->vm_start < addr + len) {
2095 if (do_munmap(mm, addr, len))
2096 return -ENOMEM;
2097 goto munmap_back;
2098 }
2099
2100 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002101 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 return -ENOMEM;
2103
2104 if (mm->map_count > sysctl_max_map_count)
2105 return -ENOMEM;
2106
2107 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2108 return -ENOMEM;
2109
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002111 vma = vma_merge(mm, prev, addr, addr + len, flags,
2112 NULL, NULL, pgoff, NULL);
2113 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 goto out;
2115
2116 /*
2117 * create a vma struct for an anonymous mapping
2118 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002119 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 if (!vma) {
2121 vm_unacct_memory(len >> PAGE_SHIFT);
2122 return -ENOMEM;
2123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
2125 vma->vm_mm = mm;
2126 vma->vm_start = addr;
2127 vma->vm_end = addr + len;
2128 vma->vm_pgoff = pgoff;
2129 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002130 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 vma_link(mm, vma, prev, rb_link, rb_parent);
2132out:
2133 mm->total_vm += len >> PAGE_SHIFT;
2134 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002135 if (!mlock_vma_pages_range(vma, addr, addr + len))
2136 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 }
2138 return addr;
2139}
2140
2141EXPORT_SYMBOL(do_brk);
2142
2143/* Release all mmaps. */
2144void exit_mmap(struct mm_struct *mm)
2145{
2146 struct mmu_gather *tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002147 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002149 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002151 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002152 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002153
Rik van Rielba470de2008-10-18 20:26:50 -07002154 if (mm->locked_vm) {
2155 vma = mm->mmap;
2156 while (vma) {
2157 if (vma->vm_flags & VM_LOCKED)
2158 munlock_vma_pages_all(vma);
2159 vma = vma->vm_next;
2160 }
2161 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002162
2163 arch_exit_mmap(mm);
2164
Rik van Rielba470de2008-10-18 20:26:50 -07002165 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002166 if (!vma) /* Can happen if dup_mmap() received an OOM */
2167 return;
2168
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002171 tlb = tlb_gather_mmu(mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002172 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002173 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002174 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 vm_unacct_memory(nr_accounted);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002176
Jan Beulich42b77722008-07-23 21:27:10 -07002177 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07002178 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002181 * Walk the list again, actually closing and freeing it,
2182 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002184 while (vma)
2185 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002186
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002187 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188}
2189
2190/* Insert vm structure into process list sorted by address
2191 * and into the inode's i_mmap tree. If vm_file is non-NULL
2192 * then i_mmap_lock is taken here.
2193 */
2194int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2195{
2196 struct vm_area_struct * __vma, * prev;
2197 struct rb_node ** rb_link, * rb_parent;
2198
2199 /*
2200 * The vm_pgoff of a purely anonymous vma should be irrelevant
2201 * until its first write fault, when page's anon_vma and index
2202 * are set. But now set the vm_pgoff it will almost certainly
2203 * end up with (unless mremap moves it elsewhere before that
2204 * first wfault), so /proc/pid/maps tells a consistent story.
2205 *
2206 * By setting it to reflect the virtual start address of the
2207 * vma, merges and splits can happen in a seamless way, just
2208 * using the existing file pgoff checks and manipulations.
2209 * Similarly in do_mmap_pgoff and in do_brk.
2210 */
2211 if (!vma->vm_file) {
2212 BUG_ON(vma->anon_vma);
2213 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2214 }
2215 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2216 if (__vma && __vma->vm_start < vma->vm_end)
2217 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002218 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002219 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002220 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 vma_link(mm, vma, prev, rb_link, rb_parent);
2222 return 0;
2223}
2224
2225/*
2226 * Copy the vma structure to a new location in the same mm,
2227 * prior to moving page table entries, to effect an mremap move.
2228 */
2229struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2230 unsigned long addr, unsigned long len, pgoff_t pgoff)
2231{
2232 struct vm_area_struct *vma = *vmap;
2233 unsigned long vma_start = vma->vm_start;
2234 struct mm_struct *mm = vma->vm_mm;
2235 struct vm_area_struct *new_vma, *prev;
2236 struct rb_node **rb_link, *rb_parent;
2237 struct mempolicy *pol;
2238
2239 /*
2240 * If anonymous vma has not yet been faulted, update new pgoff
2241 * to match new location, to increase its chance of merging.
2242 */
2243 if (!vma->vm_file && !vma->anon_vma)
2244 pgoff = addr >> PAGE_SHIFT;
2245
2246 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2247 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2248 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2249 if (new_vma) {
2250 /*
2251 * Source vma may have been merged into new_vma
2252 */
2253 if (vma_start >= new_vma->vm_start &&
2254 vma_start < new_vma->vm_end)
2255 *vmap = new_vma;
2256 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002257 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 if (new_vma) {
2259 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002260 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 if (IS_ERR(pol)) {
2262 kmem_cache_free(vm_area_cachep, new_vma);
2263 return NULL;
2264 }
2265 vma_set_policy(new_vma, pol);
2266 new_vma->vm_start = addr;
2267 new_vma->vm_end = addr + len;
2268 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002269 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002271 if (vma->vm_flags & VM_EXECUTABLE)
2272 added_exe_file_vma(mm);
2273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 if (new_vma->vm_ops && new_vma->vm_ops->open)
2275 new_vma->vm_ops->open(new_vma);
2276 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2277 }
2278 }
2279 return new_vma;
2280}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002281
2282/*
2283 * Return true if the calling process may expand its vm space by the passed
2284 * number of pages
2285 */
2286int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2287{
2288 unsigned long cur = mm->total_vm; /* pages */
2289 unsigned long lim;
2290
2291 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2292
2293 if (cur + npages > lim)
2294 return 0;
2295 return 1;
2296}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002297
2298
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002299static int special_mapping_fault(struct vm_area_struct *vma,
2300 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002301{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002302 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002303 struct page **pages;
2304
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002305 /*
2306 * special mappings have no vm_file, and in that case, the mm
2307 * uses vm_pgoff internally. So we have to subtract it from here.
2308 * We are allowed to do this because we are the mm; do not copy
2309 * this code into drivers!
2310 */
2311 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002312
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002313 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2314 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002315
2316 if (*pages) {
2317 struct page *page = *pages;
2318 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002319 vmf->page = page;
2320 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002321 }
2322
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002323 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002324}
2325
2326/*
2327 * Having a close hook prevents vma merging regardless of flags.
2328 */
2329static void special_mapping_close(struct vm_area_struct *vma)
2330{
2331}
2332
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002333static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002334 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002335 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002336};
2337
2338/*
2339 * Called with mm->mmap_sem held for writing.
2340 * Insert a new vma covering the given region, with the given flags.
2341 * Its pages are supplied by the given array of struct page *.
2342 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2343 * The region past the last page supplied will always produce SIGBUS.
2344 * The array pointer and the pages it points to are assumed to stay alive
2345 * for as long as this mapping might exist.
2346 */
2347int install_special_mapping(struct mm_struct *mm,
2348 unsigned long addr, unsigned long len,
2349 unsigned long vm_flags, struct page **pages)
2350{
2351 struct vm_area_struct *vma;
2352
2353 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2354 if (unlikely(vma == NULL))
2355 return -ENOMEM;
2356
2357 vma->vm_mm = mm;
2358 vma->vm_start = addr;
2359 vma->vm_end = addr + len;
2360
Nick Piggin2f987352008-02-02 03:08:53 +01002361 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002362 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002363
2364 vma->vm_ops = &special_mapping_vmops;
2365 vma->vm_private_data = pages;
2366
2367 if (unlikely(insert_vm_struct(mm, vma))) {
2368 kmem_cache_free(vm_area_cachep, vma);
2369 return -ENOMEM;
2370 }
2371
2372 mm->total_vm += len >> PAGE_SHIFT;
2373
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002374 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002375
Roland McGrathfa5dc222007-02-08 14:20:41 -08002376 return 0;
2377}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002378
2379static DEFINE_MUTEX(mm_all_locks_mutex);
2380
Peter Zijlstra454ed842008-08-11 09:30:25 +02002381static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002382{
2383 if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2384 /*
2385 * The LSB of head.next can't change from under us
2386 * because we hold the mm_all_locks_mutex.
2387 */
Peter Zijlstra454ed842008-08-11 09:30:25 +02002388 spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002389 /*
2390 * We can safely modify head.next after taking the
2391 * anon_vma->lock. If some other vma in this mm shares
2392 * the same anon_vma we won't take it again.
2393 *
2394 * No need of atomic instructions here, head.next
2395 * can't change from under us thanks to the
2396 * anon_vma->lock.
2397 */
2398 if (__test_and_set_bit(0, (unsigned long *)
2399 &anon_vma->head.next))
2400 BUG();
2401 }
2402}
2403
Peter Zijlstra454ed842008-08-11 09:30:25 +02002404static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002405{
2406 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2407 /*
2408 * AS_MM_ALL_LOCKS can't change from under us because
2409 * we hold the mm_all_locks_mutex.
2410 *
2411 * Operations on ->flags have to be atomic because
2412 * even if AS_MM_ALL_LOCKS is stable thanks to the
2413 * mm_all_locks_mutex, there may be other cpus
2414 * changing other bitflags in parallel to us.
2415 */
2416 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2417 BUG();
Peter Zijlstra454ed842008-08-11 09:30:25 +02002418 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002419 }
2420}
2421
2422/*
2423 * This operation locks against the VM for all pte/vma/mm related
2424 * operations that could ever happen on a certain mm. This includes
2425 * vmtruncate, try_to_unmap, and all page faults.
2426 *
2427 * The caller must take the mmap_sem in write mode before calling
2428 * mm_take_all_locks(). The caller isn't allowed to release the
2429 * mmap_sem until mm_drop_all_locks() returns.
2430 *
2431 * mmap_sem in write mode is required in order to block all operations
2432 * that could modify pagetables and free pages without need of
2433 * altering the vma layout (for example populate_range() with
2434 * nonlinear vmas). It's also needed in write mode to avoid new
2435 * anon_vmas to be associated with existing vmas.
2436 *
2437 * A single task can't take more than one mm_take_all_locks() in a row
2438 * or it would deadlock.
2439 *
2440 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2441 * mapping->flags avoid to take the same lock twice, if more than one
2442 * vma in this mm is backed by the same anon_vma or address_space.
2443 *
2444 * We can take all the locks in random order because the VM code
2445 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2446 * takes more than one of them in a row. Secondly we're protected
2447 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2448 *
2449 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2450 * that may have to take thousand of locks.
2451 *
2452 * mm_take_all_locks() can fail if it's interrupted by signals.
2453 */
2454int mm_take_all_locks(struct mm_struct *mm)
2455{
2456 struct vm_area_struct *vma;
2457 int ret = -EINTR;
2458
2459 BUG_ON(down_read_trylock(&mm->mmap_sem));
2460
2461 mutex_lock(&mm_all_locks_mutex);
2462
2463 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2464 if (signal_pending(current))
2465 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002466 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002467 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002468 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002469
2470 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2471 if (signal_pending(current))
2472 goto out_unlock;
2473 if (vma->anon_vma)
2474 vm_lock_anon_vma(mm, vma->anon_vma);
2475 }
2476
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002477 ret = 0;
2478
2479out_unlock:
2480 if (ret)
2481 mm_drop_all_locks(mm);
2482
2483 return ret;
2484}
2485
2486static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2487{
2488 if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2489 /*
2490 * The LSB of head.next can't change to 0 from under
2491 * us because we hold the mm_all_locks_mutex.
2492 *
2493 * We must however clear the bitflag before unlocking
2494 * the vma so the users using the anon_vma->head will
2495 * never see our bitflag.
2496 *
2497 * No need of atomic instructions here, head.next
2498 * can't change from under us until we release the
2499 * anon_vma->lock.
2500 */
2501 if (!__test_and_clear_bit(0, (unsigned long *)
2502 &anon_vma->head.next))
2503 BUG();
2504 spin_unlock(&anon_vma->lock);
2505 }
2506}
2507
2508static void vm_unlock_mapping(struct address_space *mapping)
2509{
2510 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2511 /*
2512 * AS_MM_ALL_LOCKS can't change to 0 from under us
2513 * because we hold the mm_all_locks_mutex.
2514 */
2515 spin_unlock(&mapping->i_mmap_lock);
2516 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2517 &mapping->flags))
2518 BUG();
2519 }
2520}
2521
2522/*
2523 * The mmap_sem cannot be released by the caller until
2524 * mm_drop_all_locks() returns.
2525 */
2526void mm_drop_all_locks(struct mm_struct *mm)
2527{
2528 struct vm_area_struct *vma;
2529
2530 BUG_ON(down_read_trylock(&mm->mmap_sem));
2531 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2532
2533 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2534 if (vma->anon_vma)
2535 vm_unlock_anon_vma(vma->anon_vma);
2536 if (vma->vm_file && vma->vm_file->f_mapping)
2537 vm_unlock_mapping(vma->vm_file->f_mapping);
2538 }
2539
2540 mutex_unlock(&mm_all_locks_mutex);
2541}
David Howells8feae132009-01-08 12:04:47 +00002542
2543/*
2544 * initialise the VMA slab
2545 */
2546void __init mmap_init(void)
2547{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002548 int ret;
2549
2550 ret = percpu_counter_init(&vm_committed_as, 0);
2551 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002552}