blob: a910c045cfd4241601306257c8822bda7f1ee05a [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31#include <asm/uaccess.h>
32#include <asm/cacheflush.h>
33#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020034#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Jan Beulich42b77722008-07-23 21:27:10 -070036#include "internal.h"
37
Kirill Korotaev3a459752006-09-07 14:17:04 +040038#ifndef arch_mmap_check
39#define arch_mmap_check(addr, len, flags) (0)
40#endif
41
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080042#ifndef arch_rebalance_pgtables
43#define arch_rebalance_pgtables(addr, len) (addr)
44#endif
45
Hugh Dickinse0da3822005-04-19 13:29:15 -070046static void unmap_region(struct mm_struct *mm,
47 struct vm_area_struct *vma, struct vm_area_struct *prev,
48 unsigned long start, unsigned long end);
49
Linus Torvalds1da177e2005-04-16 15:20:36 -070050/*
51 * WARNING: the debugging will use recursive algorithms so never enable this
52 * unless you know what you are doing.
53 */
54#undef DEBUG_MM_RB
55
56/* description of effects of mapping type and prot in current implementation.
57 * this is due to the limited x86 page protection hardware. The expected
58 * behavior is in parens:
59 *
60 * map_type prot
61 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
62 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
63 * w: (no) no w: (no) no w: (yes) yes w: (no) no
64 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
65 *
66 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
67 * w: (no) no w: (no) no w: (copy) copy w: (no) no
68 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
69 *
70 */
71pgprot_t protection_map[16] = {
72 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
73 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
74};
75
Hugh Dickins804af2c2006-07-26 21:39:49 +010076pgprot_t vm_get_page_prot(unsigned long vm_flags)
77{
Dave Kleikampb845f312008-07-08 00:28:51 +100078 return __pgprot(pgprot_val(protection_map[vm_flags &
79 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
80 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010081}
82EXPORT_SYMBOL(vm_get_page_prot);
83
Linus Torvalds1da177e2005-04-16 15:20:36 -070084int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
85int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070086int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Alan Cox80119ef2008-05-23 13:04:31 -070087atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
89/*
90 * Check that a process has enough memory to allocate a new virtual
91 * mapping. 0 means there is enough memory for the allocation to
92 * succeed and -ENOMEM implies there is not.
93 *
94 * We currently support three overcommit policies, which are set via the
95 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
96 *
97 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
98 * Additional code 2002 Jul 20 by Robert Love.
99 *
100 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
101 *
102 * Note this is a helper function intended to be used by LSMs which
103 * wish to use this logic.
104 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700105int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
107 unsigned long free, allowed;
108
109 vm_acct_memory(pages);
110
111 /*
112 * Sometimes we want to use more memory than we have
113 */
114 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
115 return 0;
116
117 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
118 unsigned long n;
119
Christoph Lameter347ce432006-06-30 01:55:35 -0700120 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 free += nr_swap_pages;
122
123 /*
124 * Any slabs which are created with the
125 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
126 * which are reclaimable, under pressure. The dentry
127 * cache and most inode caches should fall into this
128 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700129 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
131 /*
132 * Leave the last 3% for root
133 */
134 if (!cap_sys_admin)
135 free -= free / 32;
136
137 if (free > pages)
138 return 0;
139
140 /*
141 * nr_free_pages() is very expensive on large systems,
142 * only call if we're about to fail.
143 */
144 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700145
146 /*
147 * Leave reserved pages. The pages are not for anonymous pages.
148 */
149 if (n <= totalreserve_pages)
150 goto error;
151 else
152 n -= totalreserve_pages;
153
154 /*
155 * Leave the last 3% for root
156 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 if (!cap_sys_admin)
158 n -= n / 32;
159 free += n;
160
161 if (free > pages)
162 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700163
164 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 }
166
167 allowed = (totalram_pages - hugetlb_total_pages())
168 * sysctl_overcommit_ratio / 100;
169 /*
170 * Leave the last 3% for root
171 */
172 if (!cap_sys_admin)
173 allowed -= allowed / 32;
174 allowed += total_swap_pages;
175
176 /* Don't let a single process grow too big:
177 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700178 if (mm)
179 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Simon Derr2f60f8d2005-08-04 19:52:03 -0700181 /*
182 * cast `allowed' as a signed long because vm_committed_space
183 * sometimes has a negative value
184 */
Alan Cox80119ef2008-05-23 13:04:31 -0700185 if (atomic_long_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700187error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 vm_unacct_memory(pages);
189
190 return -ENOMEM;
191}
192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193/*
194 * Requires inode->i_mapping->i_mmap_lock
195 */
196static void __remove_shared_vm_struct(struct vm_area_struct *vma,
197 struct file *file, struct address_space *mapping)
198{
199 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800200 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 if (vma->vm_flags & VM_SHARED)
202 mapping->i_mmap_writable--;
203
204 flush_dcache_mmap_lock(mapping);
205 if (unlikely(vma->vm_flags & VM_NONLINEAR))
206 list_del_init(&vma->shared.vm_set.list);
207 else
208 vma_prio_tree_remove(vma, &mapping->i_mmap);
209 flush_dcache_mmap_unlock(mapping);
210}
211
212/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700213 * Unlink a file-based vm structure from its prio_tree, to hide
214 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700216void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
218 struct file *file = vma->vm_file;
219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 if (file) {
221 struct address_space *mapping = file->f_mapping;
222 spin_lock(&mapping->i_mmap_lock);
223 __remove_shared_vm_struct(vma, file, mapping);
224 spin_unlock(&mapping->i_mmap_lock);
225 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700226}
227
228/*
229 * Close a vm structure and free it, returning the next.
230 */
231static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
232{
233 struct vm_area_struct *next = vma->vm_next;
234
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700235 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 if (vma->vm_ops && vma->vm_ops->close)
237 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700238 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700239 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700240 if (vma->vm_flags & VM_EXECUTABLE)
241 removed_exe_file_vma(vma->vm_mm);
242 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700243 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700245 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248asmlinkage unsigned long sys_brk(unsigned long brk)
249{
250 unsigned long rlim, retval;
251 unsigned long newbrk, oldbrk;
252 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700253 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
255 down_write(&mm->mmap_sem);
256
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700257#ifdef CONFIG_COMPAT_BRK
258 min_brk = mm->end_code;
259#else
260 min_brk = mm->start_brk;
261#endif
262 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700264
265 /*
266 * Check against rlimit here. If this check is done later after the test
267 * of oldbrk with newbrk then it can escape the test and let the data
268 * segment grow beyond its set limit the in case where the limit is
269 * not page aligned -Ram Gupta
270 */
271 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100272 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
273 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700274 goto out;
275
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 newbrk = PAGE_ALIGN(brk);
277 oldbrk = PAGE_ALIGN(mm->brk);
278 if (oldbrk == newbrk)
279 goto set_brk;
280
281 /* Always allow shrinking brk. */
282 if (brk <= mm->brk) {
283 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
284 goto set_brk;
285 goto out;
286 }
287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 /* Check against existing mmap mappings. */
289 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
290 goto out;
291
292 /* Ok, looks good - let it rip. */
293 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
294 goto out;
295set_brk:
296 mm->brk = brk;
297out:
298 retval = mm->brk;
299 up_write(&mm->mmap_sem);
300 return retval;
301}
302
303#ifdef DEBUG_MM_RB
304static int browse_rb(struct rb_root *root)
305{
306 int i = 0, j;
307 struct rb_node *nd, *pn = NULL;
308 unsigned long prev = 0, pend = 0;
309
310 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
311 struct vm_area_struct *vma;
312 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
313 if (vma->vm_start < prev)
314 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
315 if (vma->vm_start < pend)
316 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
317 if (vma->vm_start > vma->vm_end)
318 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
319 i++;
320 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800321 prev = vma->vm_start;
322 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 }
324 j = 0;
325 for (nd = pn; nd; nd = rb_prev(nd)) {
326 j++;
327 }
328 if (i != j)
329 printk("backwards %d, forwards %d\n", j, i), i = 0;
330 return i;
331}
332
333void validate_mm(struct mm_struct *mm)
334{
335 int bug = 0;
336 int i = 0;
337 struct vm_area_struct *tmp = mm->mmap;
338 while (tmp) {
339 tmp = tmp->vm_next;
340 i++;
341 }
342 if (i != mm->map_count)
343 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
344 i = browse_rb(&mm->mm_rb);
345 if (i != mm->map_count)
346 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200347 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349#else
350#define validate_mm(mm) do { } while (0)
351#endif
352
353static struct vm_area_struct *
354find_vma_prepare(struct mm_struct *mm, unsigned long addr,
355 struct vm_area_struct **pprev, struct rb_node ***rb_link,
356 struct rb_node ** rb_parent)
357{
358 struct vm_area_struct * vma;
359 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
360
361 __rb_link = &mm->mm_rb.rb_node;
362 rb_prev = __rb_parent = NULL;
363 vma = NULL;
364
365 while (*__rb_link) {
366 struct vm_area_struct *vma_tmp;
367
368 __rb_parent = *__rb_link;
369 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
370
371 if (vma_tmp->vm_end > addr) {
372 vma = vma_tmp;
373 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700374 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 __rb_link = &__rb_parent->rb_left;
376 } else {
377 rb_prev = __rb_parent;
378 __rb_link = &__rb_parent->rb_right;
379 }
380 }
381
382 *pprev = NULL;
383 if (rb_prev)
384 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
385 *rb_link = __rb_link;
386 *rb_parent = __rb_parent;
387 return vma;
388}
389
390static inline void
391__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
392 struct vm_area_struct *prev, struct rb_node *rb_parent)
393{
394 if (prev) {
395 vma->vm_next = prev->vm_next;
396 prev->vm_next = vma;
397 } else {
398 mm->mmap = vma;
399 if (rb_parent)
400 vma->vm_next = rb_entry(rb_parent,
401 struct vm_area_struct, vm_rb);
402 else
403 vma->vm_next = NULL;
404 }
405}
406
407void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
408 struct rb_node **rb_link, struct rb_node *rb_parent)
409{
410 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
411 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
412}
413
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700414static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
ZhenwenXu48aae422009-01-06 14:40:21 -0800416 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
418 file = vma->vm_file;
419 if (file) {
420 struct address_space *mapping = file->f_mapping;
421
422 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800423 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 if (vma->vm_flags & VM_SHARED)
425 mapping->i_mmap_writable++;
426
427 flush_dcache_mmap_lock(mapping);
428 if (unlikely(vma->vm_flags & VM_NONLINEAR))
429 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
430 else
431 vma_prio_tree_insert(vma, &mapping->i_mmap);
432 flush_dcache_mmap_unlock(mapping);
433 }
434}
435
436static void
437__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
438 struct vm_area_struct *prev, struct rb_node **rb_link,
439 struct rb_node *rb_parent)
440{
441 __vma_link_list(mm, vma, prev, rb_parent);
442 __vma_link_rb(mm, vma, rb_link, rb_parent);
443 __anon_vma_link(vma);
444}
445
446static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
447 struct vm_area_struct *prev, struct rb_node **rb_link,
448 struct rb_node *rb_parent)
449{
450 struct address_space *mapping = NULL;
451
452 if (vma->vm_file)
453 mapping = vma->vm_file->f_mapping;
454
455 if (mapping) {
456 spin_lock(&mapping->i_mmap_lock);
457 vma->vm_truncate_count = mapping->truncate_count;
458 }
459 anon_vma_lock(vma);
460
461 __vma_link(mm, vma, prev, rb_link, rb_parent);
462 __vma_link_file(vma);
463
464 anon_vma_unlock(vma);
465 if (mapping)
466 spin_unlock(&mapping->i_mmap_lock);
467
468 mm->map_count++;
469 validate_mm(mm);
470}
471
472/*
473 * Helper for vma_adjust in the split_vma insert case:
474 * insert vm structure into list and rbtree and anon_vma,
475 * but it has already been inserted into prio_tree earlier.
476 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800477static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
ZhenwenXu48aae422009-01-06 14:40:21 -0800479 struct vm_area_struct *__vma, *prev;
480 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200483 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 __vma_link(mm, vma, prev, rb_link, rb_parent);
485 mm->map_count++;
486}
487
488static inline void
489__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
490 struct vm_area_struct *prev)
491{
492 prev->vm_next = vma->vm_next;
493 rb_erase(&vma->vm_rb, &mm->mm_rb);
494 if (mm->mmap_cache == vma)
495 mm->mmap_cache = prev;
496}
497
498/*
499 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
500 * is already present in an i_mmap tree without adjusting the tree.
501 * The following helper function should be used when such adjustments
502 * are necessary. The "insert" vma (if any) is to be inserted
503 * before we drop the necessary locks.
504 */
505void vma_adjust(struct vm_area_struct *vma, unsigned long start,
506 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
507{
508 struct mm_struct *mm = vma->vm_mm;
509 struct vm_area_struct *next = vma->vm_next;
510 struct vm_area_struct *importer = NULL;
511 struct address_space *mapping = NULL;
512 struct prio_tree_root *root = NULL;
513 struct file *file = vma->vm_file;
514 struct anon_vma *anon_vma = NULL;
515 long adjust_next = 0;
516 int remove_next = 0;
517
518 if (next && !insert) {
519 if (end >= next->vm_end) {
520 /*
521 * vma expands, overlapping all the next, and
522 * perhaps the one after too (mprotect case 6).
523 */
524again: remove_next = 1 + (end > next->vm_end);
525 end = next->vm_end;
526 anon_vma = next->anon_vma;
527 importer = vma;
528 } else if (end > next->vm_start) {
529 /*
530 * vma expands, overlapping part of the next:
531 * mprotect case 5 shifting the boundary up.
532 */
533 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
534 anon_vma = next->anon_vma;
535 importer = vma;
536 } else if (end < vma->vm_end) {
537 /*
538 * vma shrinks, and !insert tells it's not
539 * split_vma inserting another: so it must be
540 * mprotect case 4 shifting the boundary down.
541 */
542 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
543 anon_vma = next->anon_vma;
544 importer = next;
545 }
546 }
547
548 if (file) {
549 mapping = file->f_mapping;
550 if (!(vma->vm_flags & VM_NONLINEAR))
551 root = &mapping->i_mmap;
552 spin_lock(&mapping->i_mmap_lock);
553 if (importer &&
554 vma->vm_truncate_count != next->vm_truncate_count) {
555 /*
556 * unmap_mapping_range might be in progress:
557 * ensure that the expanding vma is rescanned.
558 */
559 importer->vm_truncate_count = 0;
560 }
561 if (insert) {
562 insert->vm_truncate_count = vma->vm_truncate_count;
563 /*
564 * Put into prio_tree now, so instantiated pages
565 * are visible to arm/parisc __flush_dcache_page
566 * throughout; but we cannot insert into address
567 * space until vma start or end is updated.
568 */
569 __vma_link_file(insert);
570 }
571 }
572
573 /*
574 * When changing only vma->vm_end, we don't really need
575 * anon_vma lock: but is that case worth optimizing out?
576 */
577 if (vma->anon_vma)
578 anon_vma = vma->anon_vma;
579 if (anon_vma) {
580 spin_lock(&anon_vma->lock);
581 /*
582 * Easily overlooked: when mprotect shifts the boundary,
583 * make sure the expanding vma has anon_vma set if the
584 * shrinking vma had, to cover any anon pages imported.
585 */
586 if (importer && !importer->anon_vma) {
587 importer->anon_vma = anon_vma;
588 __anon_vma_link(importer);
589 }
590 }
591
592 if (root) {
593 flush_dcache_mmap_lock(mapping);
594 vma_prio_tree_remove(vma, root);
595 if (adjust_next)
596 vma_prio_tree_remove(next, root);
597 }
598
599 vma->vm_start = start;
600 vma->vm_end = end;
601 vma->vm_pgoff = pgoff;
602 if (adjust_next) {
603 next->vm_start += adjust_next << PAGE_SHIFT;
604 next->vm_pgoff += adjust_next;
605 }
606
607 if (root) {
608 if (adjust_next)
609 vma_prio_tree_insert(next, root);
610 vma_prio_tree_insert(vma, root);
611 flush_dcache_mmap_unlock(mapping);
612 }
613
614 if (remove_next) {
615 /*
616 * vma_merge has merged next into vma, and needs
617 * us to remove next before dropping the locks.
618 */
619 __vma_unlink(mm, next, vma);
620 if (file)
621 __remove_shared_vm_struct(next, file, mapping);
622 if (next->anon_vma)
623 __anon_vma_merge(vma, next);
624 } else if (insert) {
625 /*
626 * split_vma has split insert from vma, and needs
627 * us to insert it before dropping the locks
628 * (it may either follow vma or precede it).
629 */
630 __insert_vm_struct(mm, insert);
631 }
632
633 if (anon_vma)
634 spin_unlock(&anon_vma->lock);
635 if (mapping)
636 spin_unlock(&mapping->i_mmap_lock);
637
638 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700639 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700641 if (next->vm_flags & VM_EXECUTABLE)
642 removed_exe_file_vma(mm);
643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700645 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 kmem_cache_free(vm_area_cachep, next);
647 /*
648 * In mprotect's case 6 (see comments on vma_merge),
649 * we must remove another next too. It would clutter
650 * up the code too much to do both in one go.
651 */
652 if (remove_next == 2) {
653 next = vma->vm_next;
654 goto again;
655 }
656 }
657
658 validate_mm(mm);
659}
660
661/*
662 * If the vma has a ->close operation then the driver probably needs to release
663 * per-vma resources, so we don't attempt to merge those.
664 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665static inline int is_mergeable_vma(struct vm_area_struct *vma,
666 struct file *file, unsigned long vm_flags)
667{
668 if (vma->vm_flags != vm_flags)
669 return 0;
670 if (vma->vm_file != file)
671 return 0;
672 if (vma->vm_ops && vma->vm_ops->close)
673 return 0;
674 return 1;
675}
676
677static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
678 struct anon_vma *anon_vma2)
679{
680 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
681}
682
683/*
684 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
685 * in front of (at a lower virtual address and file offset than) the vma.
686 *
687 * We cannot merge two vmas if they have differently assigned (non-NULL)
688 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
689 *
690 * We don't check here for the merged mmap wrapping around the end of pagecache
691 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
692 * wrap, nor mmaps which cover the final page at index -1UL.
693 */
694static int
695can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
696 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
697{
698 if (is_mergeable_vma(vma, file, vm_flags) &&
699 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
700 if (vma->vm_pgoff == vm_pgoff)
701 return 1;
702 }
703 return 0;
704}
705
706/*
707 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
708 * beyond (at a higher virtual address and file offset than) the vma.
709 *
710 * We cannot merge two vmas if they have differently assigned (non-NULL)
711 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
712 */
713static int
714can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
715 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
716{
717 if (is_mergeable_vma(vma, file, vm_flags) &&
718 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
719 pgoff_t vm_pglen;
720 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
721 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
722 return 1;
723 }
724 return 0;
725}
726
727/*
728 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
729 * whether that can be merged with its predecessor or its successor.
730 * Or both (it neatly fills a hole).
731 *
732 * In most cases - when called for mmap, brk or mremap - [addr,end) is
733 * certain not to be mapped by the time vma_merge is called; but when
734 * called for mprotect, it is certain to be already mapped (either at
735 * an offset within prev, or at the start of next), and the flags of
736 * this area are about to be changed to vm_flags - and the no-change
737 * case has already been eliminated.
738 *
739 * The following mprotect cases have to be considered, where AAAA is
740 * the area passed down from mprotect_fixup, never extending beyond one
741 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
742 *
743 * AAAA AAAA AAAA AAAA
744 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
745 * cannot merge might become might become might become
746 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
747 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
748 * mremap move: PPPPNNNNNNNN 8
749 * AAAA
750 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
751 * might become case 1 below case 2 below case 3 below
752 *
753 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
754 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
755 */
756struct vm_area_struct *vma_merge(struct mm_struct *mm,
757 struct vm_area_struct *prev, unsigned long addr,
758 unsigned long end, unsigned long vm_flags,
759 struct anon_vma *anon_vma, struct file *file,
760 pgoff_t pgoff, struct mempolicy *policy)
761{
762 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
763 struct vm_area_struct *area, *next;
764
765 /*
766 * We later require that vma->vm_flags == vm_flags,
767 * so this tests vma->vm_flags & VM_SPECIAL, too.
768 */
769 if (vm_flags & VM_SPECIAL)
770 return NULL;
771
772 if (prev)
773 next = prev->vm_next;
774 else
775 next = mm->mmap;
776 area = next;
777 if (next && next->vm_end == end) /* cases 6, 7, 8 */
778 next = next->vm_next;
779
780 /*
781 * Can it merge with the predecessor?
782 */
783 if (prev && prev->vm_end == addr &&
784 mpol_equal(vma_policy(prev), policy) &&
785 can_vma_merge_after(prev, vm_flags,
786 anon_vma, file, pgoff)) {
787 /*
788 * OK, it can. Can we now merge in the successor as well?
789 */
790 if (next && end == next->vm_start &&
791 mpol_equal(policy, vma_policy(next)) &&
792 can_vma_merge_before(next, vm_flags,
793 anon_vma, file, pgoff+pglen) &&
794 is_mergeable_anon_vma(prev->anon_vma,
795 next->anon_vma)) {
796 /* cases 1, 6 */
797 vma_adjust(prev, prev->vm_start,
798 next->vm_end, prev->vm_pgoff, NULL);
799 } else /* cases 2, 5, 7 */
800 vma_adjust(prev, prev->vm_start,
801 end, prev->vm_pgoff, NULL);
802 return prev;
803 }
804
805 /*
806 * Can this new request be merged in front of next?
807 */
808 if (next && end == next->vm_start &&
809 mpol_equal(policy, vma_policy(next)) &&
810 can_vma_merge_before(next, vm_flags,
811 anon_vma, file, pgoff+pglen)) {
812 if (prev && addr < prev->vm_end) /* case 4 */
813 vma_adjust(prev, prev->vm_start,
814 addr, prev->vm_pgoff, NULL);
815 else /* cases 3, 8 */
816 vma_adjust(area, addr, next->vm_end,
817 next->vm_pgoff - pglen, NULL);
818 return area;
819 }
820
821 return NULL;
822}
823
824/*
825 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
826 * neighbouring vmas for a suitable anon_vma, before it goes off
827 * to allocate a new anon_vma. It checks because a repetitive
828 * sequence of mprotects and faults may otherwise lead to distinct
829 * anon_vmas being allocated, preventing vma merge in subsequent
830 * mprotect.
831 */
832struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
833{
834 struct vm_area_struct *near;
835 unsigned long vm_flags;
836
837 near = vma->vm_next;
838 if (!near)
839 goto try_prev;
840
841 /*
842 * Since only mprotect tries to remerge vmas, match flags
843 * which might be mprotected into each other later on.
844 * Neither mlock nor madvise tries to remerge at present,
845 * so leave their flags as obstructing a merge.
846 */
847 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
848 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
849
850 if (near->anon_vma && vma->vm_end == near->vm_start &&
851 mpol_equal(vma_policy(vma), vma_policy(near)) &&
852 can_vma_merge_before(near, vm_flags,
853 NULL, vma->vm_file, vma->vm_pgoff +
854 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
855 return near->anon_vma;
856try_prev:
857 /*
858 * It is potentially slow to have to call find_vma_prev here.
859 * But it's only on the first write fault on the vma, not
860 * every time, and we could devise a way to avoid it later
861 * (e.g. stash info in next's anon_vma_node when assigning
862 * an anon_vma, or when trying vma_merge). Another time.
863 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200864 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 if (!near)
866 goto none;
867
868 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
869 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
870
871 if (near->anon_vma && near->vm_end == vma->vm_start &&
872 mpol_equal(vma_policy(near), vma_policy(vma)) &&
873 can_vma_merge_after(near, vm_flags,
874 NULL, vma->vm_file, vma->vm_pgoff))
875 return near->anon_vma;
876none:
877 /*
878 * There's no absolute need to look only at touching neighbours:
879 * we could search further afield for "compatible" anon_vmas.
880 * But it would probably just be a waste of time searching,
881 * or lead to too many vmas hanging off the same anon_vma.
882 * We're trying to allow mprotect remerging later on,
883 * not trying to minimize memory used for anon_vmas.
884 */
885 return NULL;
886}
887
888#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700889void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 struct file *file, long pages)
891{
892 const unsigned long stack_flags
893 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (file) {
896 mm->shared_vm += pages;
897 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
898 mm->exec_vm += pages;
899 } else if (flags & stack_flags)
900 mm->stack_vm += pages;
901 if (flags & (VM_RESERVED|VM_IO))
902 mm->reserved_vm += pages;
903}
904#endif /* CONFIG_PROC_FS */
905
906/*
907 * The caller must hold down_write(current->mm->mmap_sem).
908 */
909
ZhenwenXu48aae422009-01-06 14:40:21 -0800910unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 unsigned long len, unsigned long prot,
912 unsigned long flags, unsigned long pgoff)
913{
914 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 struct inode *inode;
916 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 int accountable = 1;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700919 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 /*
922 * Does the application expect PROT_READ to imply PROT_EXEC?
923 *
924 * (the exception is when the underlying filesystem is noexec
925 * mounted, in which case we dont add PROT_EXEC.)
926 */
927 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800928 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 prot |= PROT_EXEC;
930
931 if (!len)
932 return -EINVAL;
933
Eric Paris7cd94142007-11-26 18:47:40 -0500934 if (!(flags & MAP_FIXED))
935 addr = round_hint_to_min(addr);
936
Kirill Korotaev3a459752006-09-07 14:17:04 +0400937 error = arch_mmap_check(addr, len, flags);
938 if (error)
939 return error;
940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 /* Careful about overflows.. */
942 len = PAGE_ALIGN(len);
943 if (!len || len > TASK_SIZE)
944 return -ENOMEM;
945
946 /* offset overflow? */
947 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
948 return -EOVERFLOW;
949
950 /* Too many mappings? */
951 if (mm->map_count > sysctl_max_map_count)
952 return -ENOMEM;
953
954 /* Obtain the address to map to. we verify (or select) it and ensure
955 * that it represents a valid section of the address space.
956 */
957 addr = get_unmapped_area(file, addr, len, pgoff, flags);
958 if (addr & ~PAGE_MASK)
959 return addr;
960
961 /* Do simple checking here so the lower-level routines won't have
962 * to. we assume access permissions have been handled by the open
963 * of the memory object, so we don't do any here.
964 */
965 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
966 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
967
968 if (flags & MAP_LOCKED) {
969 if (!can_do_mlock())
970 return -EPERM;
971 vm_flags |= VM_LOCKED;
972 }
Rik van Rielba470de2008-10-18 20:26:50 -0700973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 /* mlock MCL_FUTURE? */
975 if (vm_flags & VM_LOCKED) {
976 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -0700977 locked = len >> PAGE_SHIFT;
978 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -0700980 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
982 return -EAGAIN;
983 }
984
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800985 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
987 if (file) {
988 switch (flags & MAP_TYPE) {
989 case MAP_SHARED:
990 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
991 return -EACCES;
992
993 /*
994 * Make sure we don't allow writing to an append-only
995 * file..
996 */
997 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
998 return -EACCES;
999
1000 /*
1001 * Make sure there are no mandatory locks on the file.
1002 */
1003 if (locks_verify_locked(inode))
1004 return -EAGAIN;
1005
1006 vm_flags |= VM_SHARED | VM_MAYSHARE;
1007 if (!(file->f_mode & FMODE_WRITE))
1008 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1009
1010 /* fall through */
1011 case MAP_PRIVATE:
1012 if (!(file->f_mode & FMODE_READ))
1013 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001014 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001015 if (vm_flags & VM_EXEC)
1016 return -EPERM;
1017 vm_flags &= ~VM_MAYEXEC;
1018 }
1019 if (is_file_hugepages(file))
1020 accountable = 0;
1021
1022 if (!file->f_op || !file->f_op->mmap)
1023 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 break;
1025
1026 default:
1027 return -EINVAL;
1028 }
1029 } else {
1030 switch (flags & MAP_TYPE) {
1031 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001032 /*
1033 * Ignore pgoff.
1034 */
1035 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 vm_flags |= VM_SHARED | VM_MAYSHARE;
1037 break;
1038 case MAP_PRIVATE:
1039 /*
1040 * Set pgoff according to addr for anon_vma.
1041 */
1042 pgoff = addr >> PAGE_SHIFT;
1043 break;
1044 default:
1045 return -EINVAL;
1046 }
1047 }
1048
Eric Parised032182007-06-28 15:55:21 -04001049 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (error)
1051 return error;
Eric Parised032182007-06-28 15:55:21 -04001052
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001053 return mmap_region(file, addr, len, flags, vm_flags, pgoff,
1054 accountable);
1055}
1056EXPORT_SYMBOL(do_mmap_pgoff);
1057
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001058/*
1059 * Some shared mappigns will want the pages marked read-only
1060 * to track write events. If so, we'll downgrade vm_page_prot
1061 * to the private version (using protection_map[] without the
1062 * VM_SHARED bit).
1063 */
1064int vma_wants_writenotify(struct vm_area_struct *vma)
1065{
1066 unsigned int vm_flags = vma->vm_flags;
1067
1068 /* If it was private or non-writable, the write bit is already clear */
1069 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1070 return 0;
1071
1072 /* The backer wishes to know when pages are first written to? */
1073 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1074 return 1;
1075
1076 /* The open routine did something to the protections already? */
1077 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001078 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001079 return 0;
1080
1081 /* Specialty mapping? */
1082 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1083 return 0;
1084
1085 /* Can the mapping track the dirty pages? */
1086 return vma->vm_file && vma->vm_file->f_mapping &&
1087 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1088}
1089
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001090unsigned long mmap_region(struct file *file, unsigned long addr,
1091 unsigned long len, unsigned long flags,
1092 unsigned int vm_flags, unsigned long pgoff,
1093 int accountable)
1094{
1095 struct mm_struct *mm = current->mm;
1096 struct vm_area_struct *vma, *prev;
1097 int correct_wcount = 0;
1098 int error;
1099 struct rb_node **rb_link, *rb_parent;
1100 unsigned long charged = 0;
1101 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 /* Clear old maps */
1104 error = -ENOMEM;
1105munmap_back:
1106 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1107 if (vma && vma->vm_start < addr + len) {
1108 if (do_munmap(mm, addr, len))
1109 return -ENOMEM;
1110 goto munmap_back;
1111 }
1112
1113 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001114 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 return -ENOMEM;
1116
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001117 if (flags & MAP_NORESERVE)
1118 vm_flags |= VM_NORESERVE;
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 if (accountable && (!(flags & MAP_NORESERVE) ||
1121 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1122 if (vm_flags & VM_SHARED) {
1123 /* Check memory availability in shmem_file_setup? */
1124 vm_flags |= VM_ACCOUNT;
1125 } else if (vm_flags & VM_WRITE) {
1126 /*
1127 * Private writable mapping: check memory availability
1128 */
1129 charged = len >> PAGE_SHIFT;
1130 if (security_vm_enough_memory(charged))
1131 return -ENOMEM;
1132 vm_flags |= VM_ACCOUNT;
1133 }
1134 }
1135
1136 /*
1137 * Can we just expand an old private anonymous mapping?
1138 * The VM_SHARED test is necessary because shmem_zero_setup
1139 * will create the file object for a shared anonymous map below.
1140 */
Rik van Rielba470de2008-10-18 20:26:50 -07001141 if (!file && !(vm_flags & VM_SHARED)) {
1142 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1143 NULL, NULL, pgoff, NULL);
1144 if (vma)
1145 goto out;
1146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
1148 /*
1149 * Determine the object being mapped and call the appropriate
1150 * specific mapper. the address has already been validated, but
1151 * not unmapped, but the maps are removed from the list.
1152 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001153 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (!vma) {
1155 error = -ENOMEM;
1156 goto unacct_error;
1157 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
1159 vma->vm_mm = mm;
1160 vma->vm_start = addr;
1161 vma->vm_end = addr + len;
1162 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001163 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 vma->vm_pgoff = pgoff;
1165
1166 if (file) {
1167 error = -EINVAL;
1168 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1169 goto free_vma;
1170 if (vm_flags & VM_DENYWRITE) {
1171 error = deny_write_access(file);
1172 if (error)
1173 goto free_vma;
1174 correct_wcount = 1;
1175 }
1176 vma->vm_file = file;
1177 get_file(file);
1178 error = file->f_op->mmap(file, vma);
1179 if (error)
1180 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001181 if (vm_flags & VM_EXECUTABLE)
1182 added_exe_file_vma(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 } else if (vm_flags & VM_SHARED) {
1184 error = shmem_zero_setup(vma);
1185 if (error)
1186 goto free_vma;
1187 }
1188
1189 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1190 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1191 * that memory reservation must be checked; but that reservation
1192 * belongs to shared memory object, not to vma: so now clear it.
1193 */
1194 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1195 vma->vm_flags &= ~VM_ACCOUNT;
1196
1197 /* Can addr have changed??
1198 *
1199 * Answer: Yes, several device drivers can do it in their
1200 * f_op->mmap method. -DaveM
1201 */
1202 addr = vma->vm_start;
1203 pgoff = vma->vm_pgoff;
1204 vm_flags = vma->vm_flags;
1205
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001206 if (vma_wants_writenotify(vma))
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001207 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001208
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001209 if (file && vma_merge(mm, prev, addr, vma->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001211 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 kmem_cache_free(vm_area_cachep, vma);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001213 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001214 if (vm_flags & VM_EXECUTABLE)
1215 removed_exe_file_vma(mm);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001216 } else {
1217 vma_link(mm, vma, prev, rb_link, rb_parent);
1218 file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 }
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001220
1221 /* Once vma denies write, undo our temporary denial count */
1222 if (correct_wcount)
1223 atomic_inc(&inode->i_writecount);
1224out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001226 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (vm_flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07001228 /*
1229 * makes pages present; downgrades, drops, reacquires mmap_sem
1230 */
1231 long nr_pages = mlock_vma_pages_range(vma, addr, addr + len);
1232 if (nr_pages < 0)
1233 return nr_pages; /* vma gone! */
1234 mm->locked_vm += (len >> PAGE_SHIFT) - nr_pages;
1235 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001236 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 return addr;
1238
1239unmap_and_free_vma:
1240 if (correct_wcount)
1241 atomic_inc(&inode->i_writecount);
1242 vma->vm_file = NULL;
1243 fput(file);
1244
1245 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001246 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1247 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248free_vma:
1249 kmem_cache_free(vm_area_cachep, vma);
1250unacct_error:
1251 if (charged)
1252 vm_unacct_memory(charged);
1253 return error;
1254}
1255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256/* Get an address range which is currently unmapped.
1257 * For shmat() with addr=0.
1258 *
1259 * Ugly calling convention alert:
1260 * Return value with the low bits set means error value,
1261 * ie
1262 * if (ret & ~PAGE_MASK)
1263 * error = ret;
1264 *
1265 * This function "knows" that -ENOMEM has the bits set.
1266 */
1267#ifndef HAVE_ARCH_UNMAPPED_AREA
1268unsigned long
1269arch_get_unmapped_area(struct file *filp, unsigned long addr,
1270 unsigned long len, unsigned long pgoff, unsigned long flags)
1271{
1272 struct mm_struct *mm = current->mm;
1273 struct vm_area_struct *vma;
1274 unsigned long start_addr;
1275
1276 if (len > TASK_SIZE)
1277 return -ENOMEM;
1278
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001279 if (flags & MAP_FIXED)
1280 return addr;
1281
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (addr) {
1283 addr = PAGE_ALIGN(addr);
1284 vma = find_vma(mm, addr);
1285 if (TASK_SIZE - len >= addr &&
1286 (!vma || addr + len <= vma->vm_start))
1287 return addr;
1288 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001289 if (len > mm->cached_hole_size) {
1290 start_addr = addr = mm->free_area_cache;
1291 } else {
1292 start_addr = addr = TASK_UNMAPPED_BASE;
1293 mm->cached_hole_size = 0;
1294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296full_search:
1297 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1298 /* At this point: (!vma || addr < vma->vm_end). */
1299 if (TASK_SIZE - len < addr) {
1300 /*
1301 * Start a new search - just in case we missed
1302 * some holes.
1303 */
1304 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001305 addr = TASK_UNMAPPED_BASE;
1306 start_addr = addr;
1307 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 goto full_search;
1309 }
1310 return -ENOMEM;
1311 }
1312 if (!vma || addr + len <= vma->vm_start) {
1313 /*
1314 * Remember the place where we stopped the search:
1315 */
1316 mm->free_area_cache = addr + len;
1317 return addr;
1318 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001319 if (addr + mm->cached_hole_size < vma->vm_start)
1320 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 addr = vma->vm_end;
1322 }
1323}
1324#endif
1325
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001326void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327{
1328 /*
1329 * Is this a new hole at the lowest possible address?
1330 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001331 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1332 mm->free_area_cache = addr;
1333 mm->cached_hole_size = ~0UL;
1334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335}
1336
1337/*
1338 * This mmap-allocator allocates new areas top-down from below the
1339 * stack's low limit (the base):
1340 */
1341#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1342unsigned long
1343arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1344 const unsigned long len, const unsigned long pgoff,
1345 const unsigned long flags)
1346{
1347 struct vm_area_struct *vma;
1348 struct mm_struct *mm = current->mm;
1349 unsigned long addr = addr0;
1350
1351 /* requested length too big for entire address space */
1352 if (len > TASK_SIZE)
1353 return -ENOMEM;
1354
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001355 if (flags & MAP_FIXED)
1356 return addr;
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 /* requesting a specific address */
1359 if (addr) {
1360 addr = PAGE_ALIGN(addr);
1361 vma = find_vma(mm, addr);
1362 if (TASK_SIZE - len >= addr &&
1363 (!vma || addr + len <= vma->vm_start))
1364 return addr;
1365 }
1366
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001367 /* check if free_area_cache is useful for us */
1368 if (len <= mm->cached_hole_size) {
1369 mm->cached_hole_size = 0;
1370 mm->free_area_cache = mm->mmap_base;
1371 }
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 /* either no address requested or can't fit in requested address hole */
1374 addr = mm->free_area_cache;
1375
1376 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001377 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 vma = find_vma(mm, addr-len);
1379 if (!vma || addr <= vma->vm_start)
1380 /* remember the address as a hint for next time */
1381 return (mm->free_area_cache = addr-len);
1382 }
1383
Chris Wright73219d12005-06-21 17:14:52 -07001384 if (mm->mmap_base < len)
1385 goto bottomup;
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 addr = mm->mmap_base-len;
1388
1389 do {
1390 /*
1391 * Lookup failure means no vma is above this address,
1392 * else if new region fits below vma->vm_start,
1393 * return with success:
1394 */
1395 vma = find_vma(mm, addr);
1396 if (!vma || addr+len <= vma->vm_start)
1397 /* remember the address as a hint for next time */
1398 return (mm->free_area_cache = addr);
1399
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001400 /* remember the largest hole we saw so far */
1401 if (addr + mm->cached_hole_size < vma->vm_start)
1402 mm->cached_hole_size = vma->vm_start - addr;
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 /* try just below the current vma->vm_start */
1405 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001406 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
Chris Wright73219d12005-06-21 17:14:52 -07001408bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 /*
1410 * A failed mmap() very likely causes application failure,
1411 * so fall back to the bottom-up function here. This scenario
1412 * can happen with large stack limits and large mmap()
1413 * allocations.
1414 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001415 mm->cached_hole_size = ~0UL;
1416 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1418 /*
1419 * Restore the topdown base:
1420 */
1421 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001422 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 return addr;
1425}
1426#endif
1427
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001428void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429{
1430 /*
1431 * Is this a new hole at the highest possible address?
1432 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001433 if (addr > mm->free_area_cache)
1434 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001437 if (mm->free_area_cache > mm->mmap_base)
1438 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439}
1440
1441unsigned long
1442get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1443 unsigned long pgoff, unsigned long flags)
1444{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001445 unsigned long (*get_area)(struct file *, unsigned long,
1446 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001448 get_area = current->mm->get_unmapped_area;
1449 if (file && file->f_op && file->f_op->get_unmapped_area)
1450 get_area = file->f_op->get_unmapped_area;
1451 addr = get_area(file, addr, len, pgoff, flags);
1452 if (IS_ERR_VALUE(addr))
1453 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001455 if (addr > TASK_SIZE - len)
1456 return -ENOMEM;
1457 if (addr & ~PAGE_MASK)
1458 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001459
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001460 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462
1463EXPORT_SYMBOL(get_unmapped_area);
1464
1465/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001466struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467{
1468 struct vm_area_struct *vma = NULL;
1469
1470 if (mm) {
1471 /* Check the cache first. */
1472 /* (Cache hit rate is typically around 35%.) */
1473 vma = mm->mmap_cache;
1474 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1475 struct rb_node * rb_node;
1476
1477 rb_node = mm->mm_rb.rb_node;
1478 vma = NULL;
1479
1480 while (rb_node) {
1481 struct vm_area_struct * vma_tmp;
1482
1483 vma_tmp = rb_entry(rb_node,
1484 struct vm_area_struct, vm_rb);
1485
1486 if (vma_tmp->vm_end > addr) {
1487 vma = vma_tmp;
1488 if (vma_tmp->vm_start <= addr)
1489 break;
1490 rb_node = rb_node->rb_left;
1491 } else
1492 rb_node = rb_node->rb_right;
1493 }
1494 if (vma)
1495 mm->mmap_cache = vma;
1496 }
1497 }
1498 return vma;
1499}
1500
1501EXPORT_SYMBOL(find_vma);
1502
1503/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1504struct vm_area_struct *
1505find_vma_prev(struct mm_struct *mm, unsigned long addr,
1506 struct vm_area_struct **pprev)
1507{
1508 struct vm_area_struct *vma = NULL, *prev = NULL;
ZhenwenXu48aae422009-01-06 14:40:21 -08001509 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 if (!mm)
1511 goto out;
1512
1513 /* Guard against addr being lower than the first VMA */
1514 vma = mm->mmap;
1515
1516 /* Go through the RB tree quickly. */
1517 rb_node = mm->mm_rb.rb_node;
1518
1519 while (rb_node) {
1520 struct vm_area_struct *vma_tmp;
1521 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1522
1523 if (addr < vma_tmp->vm_end) {
1524 rb_node = rb_node->rb_left;
1525 } else {
1526 prev = vma_tmp;
1527 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1528 break;
1529 rb_node = rb_node->rb_right;
1530 }
1531 }
1532
1533out:
1534 *pprev = prev;
1535 return prev ? prev->vm_next : vma;
1536}
1537
1538/*
1539 * Verify that the stack growth is acceptable and
1540 * update accounting. This is shared with both the
1541 * grow-up and grow-down cases.
1542 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001543static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544{
1545 struct mm_struct *mm = vma->vm_mm;
1546 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001547 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001550 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 return -ENOMEM;
1552
1553 /* Stack limit test */
1554 if (size > rlim[RLIMIT_STACK].rlim_cur)
1555 return -ENOMEM;
1556
1557 /* mlock limit tests */
1558 if (vma->vm_flags & VM_LOCKED) {
1559 unsigned long locked;
1560 unsigned long limit;
1561 locked = mm->locked_vm + grow;
1562 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1563 if (locked > limit && !capable(CAP_IPC_LOCK))
1564 return -ENOMEM;
1565 }
1566
Adam Litke0d59a012007-01-30 14:35:39 -08001567 /* Check to ensure the stack will not grow into a hugetlb-only region */
1568 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1569 vma->vm_end - size;
1570 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1571 return -EFAULT;
1572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 /*
1574 * Overcommit.. This must be the final test, as it will
1575 * update security statistics.
1576 */
1577 if (security_vm_enough_memory(grow))
1578 return -ENOMEM;
1579
1580 /* Ok, everything looks good - let it rip */
1581 mm->total_vm += grow;
1582 if (vma->vm_flags & VM_LOCKED)
1583 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001584 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 return 0;
1586}
1587
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001588#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001590 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1591 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001593#ifndef CONFIG_IA64
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001594static
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001595#endif
1596int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
1598 int error;
1599
1600 if (!(vma->vm_flags & VM_GROWSUP))
1601 return -EFAULT;
1602
1603 /*
1604 * We must make sure the anon_vma is allocated
1605 * so that the anon_vma locking is not a noop.
1606 */
1607 if (unlikely(anon_vma_prepare(vma)))
1608 return -ENOMEM;
1609 anon_vma_lock(vma);
1610
1611 /*
1612 * vma->vm_start/vm_end cannot change under us because the caller
1613 * is required to hold the mmap_sem in read mode. We need the
1614 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001615 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 */
Helge Deller06b32f32006-12-19 19:28:33 +01001617 if (address < PAGE_ALIGN(address+4))
1618 address = PAGE_ALIGN(address+4);
1619 else {
1620 anon_vma_unlock(vma);
1621 return -ENOMEM;
1622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 error = 0;
1624
1625 /* Somebody else might have raced and expanded it already */
1626 if (address > vma->vm_end) {
1627 unsigned long size, grow;
1628
1629 size = address - vma->vm_start;
1630 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1631
1632 error = acct_stack_growth(vma, size, grow);
1633 if (!error)
1634 vma->vm_end = address;
1635 }
1636 anon_vma_unlock(vma);
1637 return error;
1638}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001639#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1640
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641/*
1642 * vma is the first one with address < vma->vm_start. Have to extend vma.
1643 */
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001644static int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001645 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646{
1647 int error;
1648
1649 /*
1650 * We must make sure the anon_vma is allocated
1651 * so that the anon_vma locking is not a noop.
1652 */
1653 if (unlikely(anon_vma_prepare(vma)))
1654 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001655
1656 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001657 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001658 if (error)
1659 return error;
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 anon_vma_lock(vma);
1662
1663 /*
1664 * vma->vm_start/vm_end cannot change under us because the caller
1665 * is required to hold the mmap_sem in read mode. We need the
1666 * anon_vma lock to serialize against concurrent expand_stacks.
1667 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 /* Somebody else might have raced and expanded it already */
1670 if (address < vma->vm_start) {
1671 unsigned long size, grow;
1672
1673 size = vma->vm_end - address;
1674 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1675
1676 error = acct_stack_growth(vma, size, grow);
1677 if (!error) {
1678 vma->vm_start = address;
1679 vma->vm_pgoff -= grow;
1680 }
1681 }
1682 anon_vma_unlock(vma);
1683 return error;
1684}
1685
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001686int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1687{
1688 return expand_downwards(vma, address);
1689}
1690
1691#ifdef CONFIG_STACK_GROWSUP
1692int expand_stack(struct vm_area_struct *vma, unsigned long address)
1693{
1694 return expand_upwards(vma, address);
1695}
1696
1697struct vm_area_struct *
1698find_extend_vma(struct mm_struct *mm, unsigned long addr)
1699{
1700 struct vm_area_struct *vma, *prev;
1701
1702 addr &= PAGE_MASK;
1703 vma = find_vma_prev(mm, addr, &prev);
1704 if (vma && (vma->vm_start <= addr))
1705 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001706 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001707 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001708 if (prev->vm_flags & VM_LOCKED) {
1709 if (mlock_vma_pages_range(prev, addr, prev->vm_end) < 0)
1710 return NULL; /* vma gone! */
1711 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001712 return prev;
1713}
1714#else
1715int expand_stack(struct vm_area_struct *vma, unsigned long address)
1716{
1717 return expand_downwards(vma, address);
1718}
1719
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720struct vm_area_struct *
1721find_extend_vma(struct mm_struct * mm, unsigned long addr)
1722{
1723 struct vm_area_struct * vma;
1724 unsigned long start;
1725
1726 addr &= PAGE_MASK;
1727 vma = find_vma(mm,addr);
1728 if (!vma)
1729 return NULL;
1730 if (vma->vm_start <= addr)
1731 return vma;
1732 if (!(vma->vm_flags & VM_GROWSDOWN))
1733 return NULL;
1734 start = vma->vm_start;
1735 if (expand_stack(vma, addr))
1736 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001737 if (vma->vm_flags & VM_LOCKED) {
1738 if (mlock_vma_pages_range(vma, addr, start) < 0)
1739 return NULL; /* vma gone! */
1740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 return vma;
1742}
1743#endif
1744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001746 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001748 *
1749 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001751static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001753 /* Update high watermark before we lower total_vm */
1754 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001756 long nrpages = vma_pages(vma);
1757
1758 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001759 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001760 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001761 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 validate_mm(mm);
1763}
1764
1765/*
1766 * Get rid of page table information in the indicated region.
1767 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001768 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 */
1770static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001771 struct vm_area_struct *vma, struct vm_area_struct *prev,
1772 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001774 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 struct mmu_gather *tlb;
1776 unsigned long nr_accounted = 0;
1777
1778 lru_add_drain();
1779 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001780 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001781 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 vm_unacct_memory(nr_accounted);
Jan Beulich42b77722008-07-23 21:27:10 -07001783 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001784 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 tlb_finish_mmu(tlb, start, end);
1786}
1787
1788/*
1789 * Create a list of vma's touched by the unmap, removing them from the mm's
1790 * vma list as we go..
1791 */
1792static void
1793detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1794 struct vm_area_struct *prev, unsigned long end)
1795{
1796 struct vm_area_struct **insertion_point;
1797 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001798 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
1800 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1801 do {
1802 rb_erase(&vma->vm_rb, &mm->mm_rb);
1803 mm->map_count--;
1804 tail_vma = vma;
1805 vma = vma->vm_next;
1806 } while (vma && vma->vm_start < end);
1807 *insertion_point = vma;
1808 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001809 if (mm->unmap_area == arch_unmap_area)
1810 addr = prev ? prev->vm_end : mm->mmap_base;
1811 else
1812 addr = vma ? vma->vm_start : mm->mmap_base;
1813 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 mm->mmap_cache = NULL; /* Kill the cache. */
1815}
1816
1817/*
1818 * Split a vma into two pieces at address 'addr', a new vma is allocated
Michael Opdenacker59c51592007-05-09 08:57:56 +02001819 * either for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 */
1821int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1822 unsigned long addr, int new_below)
1823{
1824 struct mempolicy *pol;
1825 struct vm_area_struct *new;
1826
Andi Kleena5516432008-07-23 21:27:41 -07001827 if (is_vm_hugetlb_page(vma) && (addr &
1828 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 return -EINVAL;
1830
1831 if (mm->map_count >= sysctl_max_map_count)
1832 return -ENOMEM;
1833
Christoph Lametere94b1762006-12-06 20:33:17 -08001834 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 if (!new)
1836 return -ENOMEM;
1837
1838 /* most fields are the same, copy all, and then fixup */
1839 *new = *vma;
1840
1841 if (new_below)
1842 new->vm_end = addr;
1843 else {
1844 new->vm_start = addr;
1845 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1846 }
1847
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001848 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (IS_ERR(pol)) {
1850 kmem_cache_free(vm_area_cachep, new);
1851 return PTR_ERR(pol);
1852 }
1853 vma_set_policy(new, pol);
1854
Matt Helsley925d1c42008-04-29 01:01:36 -07001855 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001857 if (vma->vm_flags & VM_EXECUTABLE)
1858 added_exe_file_vma(mm);
1859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
1861 if (new->vm_ops && new->vm_ops->open)
1862 new->vm_ops->open(new);
1863
1864 if (new_below)
1865 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1866 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1867 else
1868 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1869
1870 return 0;
1871}
1872
1873/* Munmap is split into 2 main parts -- this part which finds
1874 * what needs doing, and the areas themselves, which do the
1875 * work. This now handles partial unmappings.
1876 * Jeremy Fitzhardinge <jeremy@goop.org>
1877 */
1878int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1879{
1880 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07001881 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
1883 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1884 return -EINVAL;
1885
1886 if ((len = PAGE_ALIGN(len)) == 0)
1887 return -EINVAL;
1888
1889 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07001890 vma = find_vma_prev(mm, start, &prev);
1891 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07001893 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
1895 /* if it doesn't overlap, we have nothing.. */
1896 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07001897 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 return 0;
1899
1900 /*
1901 * If we need to split any vma, do it now to save pain later.
1902 *
1903 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1904 * unmapped vm_area_struct will remain in use: so lower split_vma
1905 * places tmp vma above, and higher split_vma places tmp vma below.
1906 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001907 if (start > vma->vm_start) {
1908 int error = split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (error)
1910 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07001911 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
1913
1914 /* Does it split the last one? */
1915 last = find_vma(mm, end);
1916 if (last && end > last->vm_start) {
1917 int error = split_vma(mm, last, end, 1);
1918 if (error)
1919 return error;
1920 }
Hugh Dickins146425a2005-04-19 13:29:18 -07001921 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
1923 /*
Rik van Rielba470de2008-10-18 20:26:50 -07001924 * unlock any mlock()ed ranges before detaching vmas
1925 */
1926 if (mm->locked_vm) {
1927 struct vm_area_struct *tmp = vma;
1928 while (tmp && tmp->vm_start < end) {
1929 if (tmp->vm_flags & VM_LOCKED) {
1930 mm->locked_vm -= vma_pages(tmp);
1931 munlock_vma_pages_all(tmp);
1932 }
1933 tmp = tmp->vm_next;
1934 }
1935 }
1936
1937 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 * Remove the vma's, and unmap the actual pages
1939 */
Hugh Dickins146425a2005-04-19 13:29:18 -07001940 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1941 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
1943 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001944 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946 return 0;
1947}
1948
1949EXPORT_SYMBOL(do_munmap);
1950
1951asmlinkage long sys_munmap(unsigned long addr, size_t len)
1952{
1953 int ret;
1954 struct mm_struct *mm = current->mm;
1955
1956 profile_munmap(addr);
1957
1958 down_write(&mm->mmap_sem);
1959 ret = do_munmap(mm, addr, len);
1960 up_write(&mm->mmap_sem);
1961 return ret;
1962}
1963
1964static inline void verify_mm_writelocked(struct mm_struct *mm)
1965{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001966#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1968 WARN_ON(1);
1969 up_read(&mm->mmap_sem);
1970 }
1971#endif
1972}
1973
1974/*
1975 * this is really a simplified "do_mmap". it only handles
1976 * anonymous maps. eventually we may be able to do some
1977 * brk-specific accounting here.
1978 */
1979unsigned long do_brk(unsigned long addr, unsigned long len)
1980{
1981 struct mm_struct * mm = current->mm;
1982 struct vm_area_struct * vma, * prev;
1983 unsigned long flags;
1984 struct rb_node ** rb_link, * rb_parent;
1985 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04001986 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
1988 len = PAGE_ALIGN(len);
1989 if (!len)
1990 return addr;
1991
1992 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1993 return -EINVAL;
1994
Hugh Dickinscd2579d2006-11-14 13:43:38 +00001995 if (is_hugepage_only_range(mm, addr, len))
1996 return -EINVAL;
David Gibsoncb07c9a2006-11-14 02:03:38 -08001997
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001998 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05001999 if (error)
2000 return error;
2001
Kirill Korotaev3a459752006-09-07 14:17:04 +04002002 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2003
2004 error = arch_mmap_check(addr, len, flags);
2005 if (error)
2006 return error;
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 /*
2009 * mlock MCL_FUTURE?
2010 */
2011 if (mm->def_flags & VM_LOCKED) {
2012 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002013 locked = len >> PAGE_SHIFT;
2014 locked += mm->locked_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Chris Wright93ea1d02005-05-01 08:58:38 -07002016 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2018 return -EAGAIN;
2019 }
2020
2021 /*
2022 * mm->mmap_sem is required to protect against another thread
2023 * changing the mappings in case we sleep.
2024 */
2025 verify_mm_writelocked(mm);
2026
2027 /*
2028 * Clear old maps. this also does some error checking for us
2029 */
2030 munmap_back:
2031 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2032 if (vma && vma->vm_start < addr + len) {
2033 if (do_munmap(mm, addr, len))
2034 return -ENOMEM;
2035 goto munmap_back;
2036 }
2037
2038 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002039 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 return -ENOMEM;
2041
2042 if (mm->map_count > sysctl_max_map_count)
2043 return -ENOMEM;
2044
2045 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2046 return -ENOMEM;
2047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002049 vma = vma_merge(mm, prev, addr, addr + len, flags,
2050 NULL, NULL, pgoff, NULL);
2051 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 goto out;
2053
2054 /*
2055 * create a vma struct for an anonymous mapping
2056 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002057 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 if (!vma) {
2059 vm_unacct_memory(len >> PAGE_SHIFT);
2060 return -ENOMEM;
2061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
2063 vma->vm_mm = mm;
2064 vma->vm_start = addr;
2065 vma->vm_end = addr + len;
2066 vma->vm_pgoff = pgoff;
2067 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002068 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 vma_link(mm, vma, prev, rb_link, rb_parent);
2070out:
2071 mm->total_vm += len >> PAGE_SHIFT;
2072 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002073 if (!mlock_vma_pages_range(vma, addr, addr + len))
2074 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 }
2076 return addr;
2077}
2078
2079EXPORT_SYMBOL(do_brk);
2080
2081/* Release all mmaps. */
2082void exit_mmap(struct mm_struct *mm)
2083{
2084 struct mmu_gather *tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002085 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002087 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002089 /* mm's last user has gone, and its about to be pulled down */
2090 arch_exit_mmap(mm);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002091 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002092
Johannes Weinerdcd4a042009-01-06 14:40:31 -08002093 if (!mm->mmap) /* Can happen if dup_mmap() received an OOM */
2094 return;
2095
Rik van Rielba470de2008-10-18 20:26:50 -07002096 if (mm->locked_vm) {
2097 vma = mm->mmap;
2098 while (vma) {
2099 if (vma->vm_flags & VM_LOCKED)
2100 munlock_vma_pages_all(vma);
2101 vma = vma->vm_next;
2102 }
2103 }
2104 vma = mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002107 tlb = tlb_gather_mmu(mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002108 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002109 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002110 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 vm_unacct_memory(nr_accounted);
Jan Beulich42b77722008-07-23 21:27:10 -07002112 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07002113 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002116 * Walk the list again, actually closing and freeing it,
2117 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002119 while (vma)
2120 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002121
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002122 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123}
2124
2125/* Insert vm structure into process list sorted by address
2126 * and into the inode's i_mmap tree. If vm_file is non-NULL
2127 * then i_mmap_lock is taken here.
2128 */
2129int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2130{
2131 struct vm_area_struct * __vma, * prev;
2132 struct rb_node ** rb_link, * rb_parent;
2133
2134 /*
2135 * The vm_pgoff of a purely anonymous vma should be irrelevant
2136 * until its first write fault, when page's anon_vma and index
2137 * are set. But now set the vm_pgoff it will almost certainly
2138 * end up with (unless mremap moves it elsewhere before that
2139 * first wfault), so /proc/pid/maps tells a consistent story.
2140 *
2141 * By setting it to reflect the virtual start address of the
2142 * vma, merges and splits can happen in a seamless way, just
2143 * using the existing file pgoff checks and manipulations.
2144 * Similarly in do_mmap_pgoff and in do_brk.
2145 */
2146 if (!vma->vm_file) {
2147 BUG_ON(vma->anon_vma);
2148 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2149 }
2150 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2151 if (__vma && __vma->vm_start < vma->vm_end)
2152 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002153 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002154 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002155 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 vma_link(mm, vma, prev, rb_link, rb_parent);
2157 return 0;
2158}
2159
2160/*
2161 * Copy the vma structure to a new location in the same mm,
2162 * prior to moving page table entries, to effect an mremap move.
2163 */
2164struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2165 unsigned long addr, unsigned long len, pgoff_t pgoff)
2166{
2167 struct vm_area_struct *vma = *vmap;
2168 unsigned long vma_start = vma->vm_start;
2169 struct mm_struct *mm = vma->vm_mm;
2170 struct vm_area_struct *new_vma, *prev;
2171 struct rb_node **rb_link, *rb_parent;
2172 struct mempolicy *pol;
2173
2174 /*
2175 * If anonymous vma has not yet been faulted, update new pgoff
2176 * to match new location, to increase its chance of merging.
2177 */
2178 if (!vma->vm_file && !vma->anon_vma)
2179 pgoff = addr >> PAGE_SHIFT;
2180
2181 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2182 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2183 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2184 if (new_vma) {
2185 /*
2186 * Source vma may have been merged into new_vma
2187 */
2188 if (vma_start >= new_vma->vm_start &&
2189 vma_start < new_vma->vm_end)
2190 *vmap = new_vma;
2191 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002192 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 if (new_vma) {
2194 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002195 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 if (IS_ERR(pol)) {
2197 kmem_cache_free(vm_area_cachep, new_vma);
2198 return NULL;
2199 }
2200 vma_set_policy(new_vma, pol);
2201 new_vma->vm_start = addr;
2202 new_vma->vm_end = addr + len;
2203 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002204 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002206 if (vma->vm_flags & VM_EXECUTABLE)
2207 added_exe_file_vma(mm);
2208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 if (new_vma->vm_ops && new_vma->vm_ops->open)
2210 new_vma->vm_ops->open(new_vma);
2211 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2212 }
2213 }
2214 return new_vma;
2215}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002216
2217/*
2218 * Return true if the calling process may expand its vm space by the passed
2219 * number of pages
2220 */
2221int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2222{
2223 unsigned long cur = mm->total_vm; /* pages */
2224 unsigned long lim;
2225
2226 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2227
2228 if (cur + npages > lim)
2229 return 0;
2230 return 1;
2231}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002232
2233
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002234static int special_mapping_fault(struct vm_area_struct *vma,
2235 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002236{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002237 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002238 struct page **pages;
2239
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002240 /*
2241 * special mappings have no vm_file, and in that case, the mm
2242 * uses vm_pgoff internally. So we have to subtract it from here.
2243 * We are allowed to do this because we are the mm; do not copy
2244 * this code into drivers!
2245 */
2246 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002247
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002248 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2249 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002250
2251 if (*pages) {
2252 struct page *page = *pages;
2253 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002254 vmf->page = page;
2255 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002256 }
2257
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002258 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002259}
2260
2261/*
2262 * Having a close hook prevents vma merging regardless of flags.
2263 */
2264static void special_mapping_close(struct vm_area_struct *vma)
2265{
2266}
2267
2268static struct vm_operations_struct special_mapping_vmops = {
2269 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002270 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002271};
2272
2273/*
2274 * Called with mm->mmap_sem held for writing.
2275 * Insert a new vma covering the given region, with the given flags.
2276 * Its pages are supplied by the given array of struct page *.
2277 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2278 * The region past the last page supplied will always produce SIGBUS.
2279 * The array pointer and the pages it points to are assumed to stay alive
2280 * for as long as this mapping might exist.
2281 */
2282int install_special_mapping(struct mm_struct *mm,
2283 unsigned long addr, unsigned long len,
2284 unsigned long vm_flags, struct page **pages)
2285{
2286 struct vm_area_struct *vma;
2287
2288 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2289 if (unlikely(vma == NULL))
2290 return -ENOMEM;
2291
2292 vma->vm_mm = mm;
2293 vma->vm_start = addr;
2294 vma->vm_end = addr + len;
2295
Nick Piggin2f987352008-02-02 03:08:53 +01002296 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002297 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002298
2299 vma->vm_ops = &special_mapping_vmops;
2300 vma->vm_private_data = pages;
2301
2302 if (unlikely(insert_vm_struct(mm, vma))) {
2303 kmem_cache_free(vm_area_cachep, vma);
2304 return -ENOMEM;
2305 }
2306
2307 mm->total_vm += len >> PAGE_SHIFT;
2308
2309 return 0;
2310}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002311
2312static DEFINE_MUTEX(mm_all_locks_mutex);
2313
Peter Zijlstra454ed842008-08-11 09:30:25 +02002314static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002315{
2316 if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2317 /*
2318 * The LSB of head.next can't change from under us
2319 * because we hold the mm_all_locks_mutex.
2320 */
Peter Zijlstra454ed842008-08-11 09:30:25 +02002321 spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002322 /*
2323 * We can safely modify head.next after taking the
2324 * anon_vma->lock. If some other vma in this mm shares
2325 * the same anon_vma we won't take it again.
2326 *
2327 * No need of atomic instructions here, head.next
2328 * can't change from under us thanks to the
2329 * anon_vma->lock.
2330 */
2331 if (__test_and_set_bit(0, (unsigned long *)
2332 &anon_vma->head.next))
2333 BUG();
2334 }
2335}
2336
Peter Zijlstra454ed842008-08-11 09:30:25 +02002337static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002338{
2339 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2340 /*
2341 * AS_MM_ALL_LOCKS can't change from under us because
2342 * we hold the mm_all_locks_mutex.
2343 *
2344 * Operations on ->flags have to be atomic because
2345 * even if AS_MM_ALL_LOCKS is stable thanks to the
2346 * mm_all_locks_mutex, there may be other cpus
2347 * changing other bitflags in parallel to us.
2348 */
2349 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2350 BUG();
Peter Zijlstra454ed842008-08-11 09:30:25 +02002351 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002352 }
2353}
2354
2355/*
2356 * This operation locks against the VM for all pte/vma/mm related
2357 * operations that could ever happen on a certain mm. This includes
2358 * vmtruncate, try_to_unmap, and all page faults.
2359 *
2360 * The caller must take the mmap_sem in write mode before calling
2361 * mm_take_all_locks(). The caller isn't allowed to release the
2362 * mmap_sem until mm_drop_all_locks() returns.
2363 *
2364 * mmap_sem in write mode is required in order to block all operations
2365 * that could modify pagetables and free pages without need of
2366 * altering the vma layout (for example populate_range() with
2367 * nonlinear vmas). It's also needed in write mode to avoid new
2368 * anon_vmas to be associated with existing vmas.
2369 *
2370 * A single task can't take more than one mm_take_all_locks() in a row
2371 * or it would deadlock.
2372 *
2373 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2374 * mapping->flags avoid to take the same lock twice, if more than one
2375 * vma in this mm is backed by the same anon_vma or address_space.
2376 *
2377 * We can take all the locks in random order because the VM code
2378 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2379 * takes more than one of them in a row. Secondly we're protected
2380 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2381 *
2382 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2383 * that may have to take thousand of locks.
2384 *
2385 * mm_take_all_locks() can fail if it's interrupted by signals.
2386 */
2387int mm_take_all_locks(struct mm_struct *mm)
2388{
2389 struct vm_area_struct *vma;
2390 int ret = -EINTR;
2391
2392 BUG_ON(down_read_trylock(&mm->mmap_sem));
2393
2394 mutex_lock(&mm_all_locks_mutex);
2395
2396 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2397 if (signal_pending(current))
2398 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002399 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002400 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002401 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002402
2403 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2404 if (signal_pending(current))
2405 goto out_unlock;
2406 if (vma->anon_vma)
2407 vm_lock_anon_vma(mm, vma->anon_vma);
2408 }
2409
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002410 ret = 0;
2411
2412out_unlock:
2413 if (ret)
2414 mm_drop_all_locks(mm);
2415
2416 return ret;
2417}
2418
2419static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2420{
2421 if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2422 /*
2423 * The LSB of head.next can't change to 0 from under
2424 * us because we hold the mm_all_locks_mutex.
2425 *
2426 * We must however clear the bitflag before unlocking
2427 * the vma so the users using the anon_vma->head will
2428 * never see our bitflag.
2429 *
2430 * No need of atomic instructions here, head.next
2431 * can't change from under us until we release the
2432 * anon_vma->lock.
2433 */
2434 if (!__test_and_clear_bit(0, (unsigned long *)
2435 &anon_vma->head.next))
2436 BUG();
2437 spin_unlock(&anon_vma->lock);
2438 }
2439}
2440
2441static void vm_unlock_mapping(struct address_space *mapping)
2442{
2443 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2444 /*
2445 * AS_MM_ALL_LOCKS can't change to 0 from under us
2446 * because we hold the mm_all_locks_mutex.
2447 */
2448 spin_unlock(&mapping->i_mmap_lock);
2449 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2450 &mapping->flags))
2451 BUG();
2452 }
2453}
2454
2455/*
2456 * The mmap_sem cannot be released by the caller until
2457 * mm_drop_all_locks() returns.
2458 */
2459void mm_drop_all_locks(struct mm_struct *mm)
2460{
2461 struct vm_area_struct *vma;
2462
2463 BUG_ON(down_read_trylock(&mm->mmap_sem));
2464 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2465
2466 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2467 if (vma->anon_vma)
2468 vm_unlock_anon_vma(vma->anon_vma);
2469 if (vma->vm_file && vma->vm_file->f_mapping)
2470 vm_unlock_mapping(vma->vm_file->f_mapping);
2471 }
2472
2473 mutex_unlock(&mm_all_locks_mutex);
2474}