blob: 31003338b978b78ef66d4579033cdd1d3ebd1a16 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
Alan Cox046c6882009-01-05 14:06:29 +00006 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 */
8
9#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070010#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/mm.h>
12#include <linux/shm.h>
13#include <linux/mman.h>
14#include <linux/pagemap.h>
15#include <linux/swap.h>
16#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/init.h>
19#include <linux/file.h>
20#include <linux/fs.h>
21#include <linux/personality.h>
22#include <linux/security.h>
23#include <linux/hugetlb.h>
24#include <linux/profile.h>
25#include <linux/module.h>
26#include <linux/mount.h>
27#include <linux/mempolicy.h>
28#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070029#include <linux/mmu_notifier.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020030#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/uaccess.h>
33#include <asm/cacheflush.h>
34#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020035#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Jan Beulich42b77722008-07-23 21:27:10 -070037#include "internal.h"
38
Kirill Korotaev3a459752006-09-07 14:17:04 +040039#ifndef arch_mmap_check
40#define arch_mmap_check(addr, len, flags) (0)
41#endif
42
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080043#ifndef arch_rebalance_pgtables
44#define arch_rebalance_pgtables(addr, len) (addr)
45#endif
46
Hugh Dickinse0da3822005-04-19 13:29:15 -070047static void unmap_region(struct mm_struct *mm,
48 struct vm_area_struct *vma, struct vm_area_struct *prev,
49 unsigned long start, unsigned long end);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * WARNING: the debugging will use recursive algorithms so never enable this
53 * unless you know what you are doing.
54 */
55#undef DEBUG_MM_RB
56
57/* description of effects of mapping type and prot in current implementation.
58 * this is due to the limited x86 page protection hardware. The expected
59 * behavior is in parens:
60 *
61 * map_type prot
62 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
63 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
64 * w: (no) no w: (no) no w: (yes) yes w: (no) no
65 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
66 *
67 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
68 * w: (no) no w: (no) no w: (copy) copy w: (no) no
69 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
70 *
71 */
72pgprot_t protection_map[16] = {
73 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
74 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
75};
76
Hugh Dickins804af2c2006-07-26 21:39:49 +010077pgprot_t vm_get_page_prot(unsigned long vm_flags)
78{
Dave Kleikampb845f312008-07-08 00:28:51 +100079 return __pgprot(pgprot_val(protection_map[vm_flags &
80 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
81 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010082}
83EXPORT_SYMBOL(vm_get_page_prot);
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
86int sysctl_overcommit_ratio = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070087int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070088struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90/*
91 * Check that a process has enough memory to allocate a new virtual
92 * mapping. 0 means there is enough memory for the allocation to
93 * succeed and -ENOMEM implies there is not.
94 *
95 * We currently support three overcommit policies, which are set via the
96 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
97 *
98 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
99 * Additional code 2002 Jul 20 by Robert Love.
100 *
101 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
102 *
103 * Note this is a helper function intended to be used by LSMs which
104 * wish to use this logic.
105 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700106int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107{
108 unsigned long free, allowed;
109
110 vm_acct_memory(pages);
111
112 /*
113 * Sometimes we want to use more memory than we have
114 */
115 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
116 return 0;
117
118 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
119 unsigned long n;
120
Christoph Lameter347ce432006-06-30 01:55:35 -0700121 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 free += nr_swap_pages;
123
124 /*
125 * Any slabs which are created with the
126 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
127 * which are reclaimable, under pressure. The dentry
128 * cache and most inode caches should fall into this
129 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700130 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 /*
133 * Leave the last 3% for root
134 */
135 if (!cap_sys_admin)
136 free -= free / 32;
137
138 if (free > pages)
139 return 0;
140
141 /*
142 * nr_free_pages() is very expensive on large systems,
143 * only call if we're about to fail.
144 */
145 n = nr_free_pages();
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700146
147 /*
148 * Leave reserved pages. The pages are not for anonymous pages.
149 */
150 if (n <= totalreserve_pages)
151 goto error;
152 else
153 n -= totalreserve_pages;
154
155 /*
156 * Leave the last 3% for root
157 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 if (!cap_sys_admin)
159 n -= n / 32;
160 free += n;
161
162 if (free > pages)
163 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700164
165 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 }
167
168 allowed = (totalram_pages - hugetlb_total_pages())
169 * sysctl_overcommit_ratio / 100;
170 /*
171 * Leave the last 3% for root
172 */
173 if (!cap_sys_admin)
174 allowed -= allowed / 32;
175 allowed += total_swap_pages;
176
177 /* Don't let a single process grow too big:
178 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700179 if (mm)
180 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700182 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700184error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 vm_unacct_memory(pages);
186
187 return -ENOMEM;
188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/*
191 * Requires inode->i_mapping->i_mmap_lock
192 */
193static void __remove_shared_vm_struct(struct vm_area_struct *vma,
194 struct file *file, struct address_space *mapping)
195{
196 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800197 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 if (vma->vm_flags & VM_SHARED)
199 mapping->i_mmap_writable--;
200
201 flush_dcache_mmap_lock(mapping);
202 if (unlikely(vma->vm_flags & VM_NONLINEAR))
203 list_del_init(&vma->shared.vm_set.list);
204 else
205 vma_prio_tree_remove(vma, &mapping->i_mmap);
206 flush_dcache_mmap_unlock(mapping);
207}
208
209/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700210 * Unlink a file-based vm structure from its prio_tree, to hide
211 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700213void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 struct file *file = vma->vm_file;
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 if (file) {
218 struct address_space *mapping = file->f_mapping;
219 spin_lock(&mapping->i_mmap_lock);
220 __remove_shared_vm_struct(vma, file, mapping);
221 spin_unlock(&mapping->i_mmap_lock);
222 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700223}
224
225/*
226 * Close a vm structure and free it, returning the next.
227 */
228static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
229{
230 struct vm_area_struct *next = vma->vm_next;
231
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700232 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (vma->vm_ops && vma->vm_ops->close)
234 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700235 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700236 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700237 if (vma->vm_flags & VM_EXECUTABLE)
238 removed_exe_file_vma(vma->vm_mm);
239 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700240 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700242 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100245SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 unsigned long rlim, retval;
248 unsigned long newbrk, oldbrk;
249 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700250 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 down_write(&mm->mmap_sem);
253
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700254#ifdef CONFIG_COMPAT_BRK
255 min_brk = mm->end_code;
256#else
257 min_brk = mm->start_brk;
258#endif
259 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700261
262 /*
263 * Check against rlimit here. If this check is done later after the test
264 * of oldbrk with newbrk then it can escape the test and let the data
265 * segment grow beyond its set limit the in case where the limit is
266 * not page aligned -Ram Gupta
267 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800268 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100269 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
270 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700271 goto out;
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 newbrk = PAGE_ALIGN(brk);
274 oldbrk = PAGE_ALIGN(mm->brk);
275 if (oldbrk == newbrk)
276 goto set_brk;
277
278 /* Always allow shrinking brk. */
279 if (brk <= mm->brk) {
280 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
281 goto set_brk;
282 goto out;
283 }
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 /* Check against existing mmap mappings. */
286 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
287 goto out;
288
289 /* Ok, looks good - let it rip. */
290 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
291 goto out;
292set_brk:
293 mm->brk = brk;
294out:
295 retval = mm->brk;
296 up_write(&mm->mmap_sem);
297 return retval;
298}
299
300#ifdef DEBUG_MM_RB
301static int browse_rb(struct rb_root *root)
302{
303 int i = 0, j;
304 struct rb_node *nd, *pn = NULL;
305 unsigned long prev = 0, pend = 0;
306
307 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
308 struct vm_area_struct *vma;
309 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
310 if (vma->vm_start < prev)
311 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
312 if (vma->vm_start < pend)
313 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
314 if (vma->vm_start > vma->vm_end)
315 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
316 i++;
317 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800318 prev = vma->vm_start;
319 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 }
321 j = 0;
322 for (nd = pn; nd; nd = rb_prev(nd)) {
323 j++;
324 }
325 if (i != j)
326 printk("backwards %d, forwards %d\n", j, i), i = 0;
327 return i;
328}
329
330void validate_mm(struct mm_struct *mm)
331{
332 int bug = 0;
333 int i = 0;
334 struct vm_area_struct *tmp = mm->mmap;
335 while (tmp) {
336 tmp = tmp->vm_next;
337 i++;
338 }
339 if (i != mm->map_count)
340 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
341 i = browse_rb(&mm->mm_rb);
342 if (i != mm->map_count)
343 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200344 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346#else
347#define validate_mm(mm) do { } while (0)
348#endif
349
350static struct vm_area_struct *
351find_vma_prepare(struct mm_struct *mm, unsigned long addr,
352 struct vm_area_struct **pprev, struct rb_node ***rb_link,
353 struct rb_node ** rb_parent)
354{
355 struct vm_area_struct * vma;
356 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
357
358 __rb_link = &mm->mm_rb.rb_node;
359 rb_prev = __rb_parent = NULL;
360 vma = NULL;
361
362 while (*__rb_link) {
363 struct vm_area_struct *vma_tmp;
364
365 __rb_parent = *__rb_link;
366 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
367
368 if (vma_tmp->vm_end > addr) {
369 vma = vma_tmp;
370 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700371 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 __rb_link = &__rb_parent->rb_left;
373 } else {
374 rb_prev = __rb_parent;
375 __rb_link = &__rb_parent->rb_right;
376 }
377 }
378
379 *pprev = NULL;
380 if (rb_prev)
381 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
382 *rb_link = __rb_link;
383 *rb_parent = __rb_parent;
384 return vma;
385}
386
387static inline void
388__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
389 struct vm_area_struct *prev, struct rb_node *rb_parent)
390{
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
441
442static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
443 struct vm_area_struct *prev, struct rb_node **rb_link,
444 struct rb_node *rb_parent)
445{
446 struct address_space *mapping = NULL;
447
448 if (vma->vm_file)
449 mapping = vma->vm_file->f_mapping;
450
451 if (mapping) {
452 spin_lock(&mapping->i_mmap_lock);
453 vma->vm_truncate_count = mapping->truncate_count;
454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
456 __vma_link(mm, vma, prev, rb_link, rb_parent);
457 __vma_link_file(vma);
458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (mapping)
460 spin_unlock(&mapping->i_mmap_lock);
461
462 mm->map_count++;
463 validate_mm(mm);
464}
465
466/*
467 * Helper for vma_adjust in the split_vma insert case:
468 * insert vm structure into list and rbtree and anon_vma,
469 * but it has already been inserted into prio_tree earlier.
470 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800471static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
ZhenwenXu48aae422009-01-06 14:40:21 -0800473 struct vm_area_struct *__vma, *prev;
474 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200477 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 __vma_link(mm, vma, prev, rb_link, rb_parent);
479 mm->map_count++;
480}
481
482static inline void
483__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
484 struct vm_area_struct *prev)
485{
486 prev->vm_next = vma->vm_next;
487 rb_erase(&vma->vm_rb, &mm->mm_rb);
488 if (mm->mmap_cache == vma)
489 mm->mmap_cache = prev;
490}
491
492/*
493 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
494 * is already present in an i_mmap tree without adjusting the tree.
495 * The following helper function should be used when such adjustments
496 * are necessary. The "insert" vma (if any) is to be inserted
497 * before we drop the necessary locks.
498 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800499int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
501{
502 struct mm_struct *mm = vma->vm_mm;
503 struct vm_area_struct *next = vma->vm_next;
504 struct vm_area_struct *importer = NULL;
505 struct address_space *mapping = NULL;
506 struct prio_tree_root *root = NULL;
Rik van Riel012f1802010-08-09 17:18:40 -0700507 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 long adjust_next = 0;
510 int remove_next = 0;
511
512 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700513 struct vm_area_struct *exporter = NULL;
514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 if (end >= next->vm_end) {
516 /*
517 * vma expands, overlapping all the next, and
518 * perhaps the one after too (mprotect case 6).
519 */
520again: remove_next = 1 + (end > next->vm_end);
521 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700522 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 importer = vma;
524 } else if (end > next->vm_start) {
525 /*
526 * vma expands, overlapping part of the next:
527 * mprotect case 5 shifting the boundary up.
528 */
529 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700530 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 importer = vma;
532 } else if (end < vma->vm_end) {
533 /*
534 * vma shrinks, and !insert tells it's not
535 * split_vma inserting another: so it must be
536 * mprotect case 4 shifting the boundary down.
537 */
538 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700539 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 importer = next;
541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Rik van Riel5beb4932010-03-05 13:42:07 -0800543 /*
544 * Easily overlooked: when mprotect shifts the boundary,
545 * make sure the expanding vma has anon_vma set if the
546 * shrinking vma had, to cover any anon pages imported.
547 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700548 if (exporter && exporter->anon_vma && !importer->anon_vma) {
549 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800550 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700551 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800552 }
553 }
554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (file) {
556 mapping = file->f_mapping;
557 if (!(vma->vm_flags & VM_NONLINEAR))
558 root = &mapping->i_mmap;
559 spin_lock(&mapping->i_mmap_lock);
560 if (importer &&
561 vma->vm_truncate_count != next->vm_truncate_count) {
562 /*
563 * unmap_mapping_range might be in progress:
564 * ensure that the expanding vma is rescanned.
565 */
566 importer->vm_truncate_count = 0;
567 }
568 if (insert) {
569 insert->vm_truncate_count = vma->vm_truncate_count;
570 /*
571 * Put into prio_tree now, so instantiated pages
572 * are visible to arm/parisc __flush_dcache_page
573 * throughout; but we cannot insert into address
574 * space until vma start or end is updated.
575 */
576 __vma_link_file(insert);
577 }
578 }
579
Rik van Riel012f1802010-08-09 17:18:40 -0700580 /*
581 * When changing only vma->vm_end, we don't really need anon_vma
582 * lock. This is a fairly rare case by itself, but the anon_vma
583 * lock may be shared between many sibling processes. Skipping
584 * the lock for brk adjustments makes a difference sometimes.
585 */
586 if (vma->anon_vma && (insert || importer || start != vma->vm_start)) {
587 anon_vma = vma->anon_vma;
588 anon_vma_lock(anon_vma);
589 }
590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 if (root) {
592 flush_dcache_mmap_lock(mapping);
593 vma_prio_tree_remove(vma, root);
594 if (adjust_next)
595 vma_prio_tree_remove(next, root);
596 }
597
598 vma->vm_start = start;
599 vma->vm_end = end;
600 vma->vm_pgoff = pgoff;
601 if (adjust_next) {
602 next->vm_start += adjust_next << PAGE_SHIFT;
603 next->vm_pgoff += adjust_next;
604 }
605
606 if (root) {
607 if (adjust_next)
608 vma_prio_tree_insert(next, root);
609 vma_prio_tree_insert(vma, root);
610 flush_dcache_mmap_unlock(mapping);
611 }
612
613 if (remove_next) {
614 /*
615 * vma_merge has merged next into vma, and needs
616 * us to remove next before dropping the locks.
617 */
618 __vma_unlink(mm, next, vma);
619 if (file)
620 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 } 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
Rik van Riel012f1802010-08-09 17:18:40 -0700630 if (anon_vma)
631 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 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 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800641 if (next->anon_vma)
642 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700644 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 kmem_cache_free(vm_area_cachep, next);
646 /*
647 * In mprotect's case 6 (see comments on vma_merge),
648 * we must remove another next too. It would clutter
649 * up the code too much to do both in one go.
650 */
651 if (remove_next == 2) {
652 next = vma->vm_next;
653 goto again;
654 }
655 }
656
657 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800658
659 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
662/*
663 * If the vma has a ->close operation then the driver probably needs to release
664 * per-vma resources, so we don't attempt to merge those.
665 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666static inline int is_mergeable_vma(struct vm_area_struct *vma,
667 struct file *file, unsigned long vm_flags)
668{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700669 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
670 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 return 0;
672 if (vma->vm_file != file)
673 return 0;
674 if (vma->vm_ops && vma->vm_ops->close)
675 return 0;
676 return 1;
677}
678
679static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
680 struct anon_vma *anon_vma2)
681{
682 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
683}
684
685/*
686 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
687 * in front of (at a lower virtual address and file offset than) the vma.
688 *
689 * We cannot merge two vmas if they have differently assigned (non-NULL)
690 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
691 *
692 * We don't check here for the merged mmap wrapping around the end of pagecache
693 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
694 * wrap, nor mmaps which cover the final page at index -1UL.
695 */
696static int
697can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
698 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
699{
700 if (is_mergeable_vma(vma, file, vm_flags) &&
701 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
702 if (vma->vm_pgoff == vm_pgoff)
703 return 1;
704 }
705 return 0;
706}
707
708/*
709 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
710 * beyond (at a higher virtual address and file offset than) the vma.
711 *
712 * We cannot merge two vmas if they have differently assigned (non-NULL)
713 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
714 */
715static int
716can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
717 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
718{
719 if (is_mergeable_vma(vma, file, vm_flags) &&
720 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
721 pgoff_t vm_pglen;
722 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
723 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
724 return 1;
725 }
726 return 0;
727}
728
729/*
730 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
731 * whether that can be merged with its predecessor or its successor.
732 * Or both (it neatly fills a hole).
733 *
734 * In most cases - when called for mmap, brk or mremap - [addr,end) is
735 * certain not to be mapped by the time vma_merge is called; but when
736 * called for mprotect, it is certain to be already mapped (either at
737 * an offset within prev, or at the start of next), and the flags of
738 * this area are about to be changed to vm_flags - and the no-change
739 * case has already been eliminated.
740 *
741 * The following mprotect cases have to be considered, where AAAA is
742 * the area passed down from mprotect_fixup, never extending beyond one
743 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
744 *
745 * AAAA AAAA AAAA AAAA
746 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
747 * cannot merge might become might become might become
748 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
749 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
750 * mremap move: PPPPNNNNNNNN 8
751 * AAAA
752 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
753 * might become case 1 below case 2 below case 3 below
754 *
755 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
756 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
757 */
758struct vm_area_struct *vma_merge(struct mm_struct *mm,
759 struct vm_area_struct *prev, unsigned long addr,
760 unsigned long end, unsigned long vm_flags,
761 struct anon_vma *anon_vma, struct file *file,
762 pgoff_t pgoff, struct mempolicy *policy)
763{
764 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
765 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800766 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 /*
769 * We later require that vma->vm_flags == vm_flags,
770 * so this tests vma->vm_flags & VM_SPECIAL, too.
771 */
772 if (vm_flags & VM_SPECIAL)
773 return NULL;
774
775 if (prev)
776 next = prev->vm_next;
777 else
778 next = mm->mmap;
779 area = next;
780 if (next && next->vm_end == end) /* cases 6, 7, 8 */
781 next = next->vm_next;
782
783 /*
784 * Can it merge with the predecessor?
785 */
786 if (prev && prev->vm_end == addr &&
787 mpol_equal(vma_policy(prev), policy) &&
788 can_vma_merge_after(prev, vm_flags,
789 anon_vma, file, pgoff)) {
790 /*
791 * OK, it can. Can we now merge in the successor as well?
792 */
793 if (next && end == next->vm_start &&
794 mpol_equal(policy, vma_policy(next)) &&
795 can_vma_merge_before(next, vm_flags,
796 anon_vma, file, pgoff+pglen) &&
797 is_mergeable_anon_vma(prev->anon_vma,
798 next->anon_vma)) {
799 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800800 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 next->vm_end, prev->vm_pgoff, NULL);
802 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800803 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800805 if (err)
806 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 return prev;
808 }
809
810 /*
811 * Can this new request be merged in front of next?
812 */
813 if (next && end == next->vm_start &&
814 mpol_equal(policy, vma_policy(next)) &&
815 can_vma_merge_before(next, vm_flags,
816 anon_vma, file, pgoff+pglen)) {
817 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800818 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 addr, prev->vm_pgoff, NULL);
820 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800821 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800823 if (err)
824 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 return area;
826 }
827
828 return NULL;
829}
830
831/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700832 * Rough compatbility check to quickly see if it's even worth looking
833 * at sharing an anon_vma.
834 *
835 * They need to have the same vm_file, and the flags can only differ
836 * in things that mprotect may change.
837 *
838 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
839 * we can merge the two vma's. For example, we refuse to merge a vma if
840 * there is a vm_ops->close() function, because that indicates that the
841 * driver is doing some kind of reference counting. But that doesn't
842 * really matter for the anon_vma sharing case.
843 */
844static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
845{
846 return a->vm_end == b->vm_start &&
847 mpol_equal(vma_policy(a), vma_policy(b)) &&
848 a->vm_file == b->vm_file &&
849 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
850 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
851}
852
853/*
854 * Do some basic sanity checking to see if we can re-use the anon_vma
855 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
856 * the same as 'old', the other will be the new one that is trying
857 * to share the anon_vma.
858 *
859 * NOTE! This runs with mm_sem held for reading, so it is possible that
860 * the anon_vma of 'old' is concurrently in the process of being set up
861 * by another page fault trying to merge _that_. But that's ok: if it
862 * is being set up, that automatically means that it will be a singleton
863 * acceptable for merging, so we can do all of this optimistically. But
864 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
865 *
866 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
867 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
868 * is to return an anon_vma that is "complex" due to having gone through
869 * a fork).
870 *
871 * We also make sure that the two vma's are compatible (adjacent,
872 * and with the same memory policies). That's all stable, even with just
873 * a read lock on the mm_sem.
874 */
875static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
876{
877 if (anon_vma_compatible(a, b)) {
878 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
879
880 if (anon_vma && list_is_singular(&old->anon_vma_chain))
881 return anon_vma;
882 }
883 return NULL;
884}
885
886/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
888 * neighbouring vmas for a suitable anon_vma, before it goes off
889 * to allocate a new anon_vma. It checks because a repetitive
890 * sequence of mprotects and faults may otherwise lead to distinct
891 * anon_vmas being allocated, preventing vma merge in subsequent
892 * mprotect.
893 */
894struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
895{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700896 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 near = vma->vm_next;
900 if (!near)
901 goto try_prev;
902
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700903 anon_vma = reusable_anon_vma(near, vma, near);
904 if (anon_vma)
905 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906try_prev:
907 /*
908 * It is potentially slow to have to call find_vma_prev here.
909 * But it's only on the first write fault on the vma, not
910 * every time, and we could devise a way to avoid it later
911 * (e.g. stash info in next's anon_vma_node when assigning
912 * an anon_vma, or when trying vma_merge). Another time.
913 */
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200914 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (!near)
916 goto none;
917
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700918 anon_vma = reusable_anon_vma(near, near, vma);
919 if (anon_vma)
920 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921none:
922 /*
923 * There's no absolute need to look only at touching neighbours:
924 * we could search further afield for "compatible" anon_vmas.
925 * But it would probably just be a waste of time searching,
926 * or lead to too many vmas hanging off the same anon_vma.
927 * We're trying to allow mprotect remerging later on,
928 * not trying to minimize memory used for anon_vmas.
929 */
930 return NULL;
931}
932
933#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700934void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 struct file *file, long pages)
936{
937 const unsigned long stack_flags
938 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 if (file) {
941 mm->shared_vm += pages;
942 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
943 mm->exec_vm += pages;
944 } else if (flags & stack_flags)
945 mm->stack_vm += pages;
946 if (flags & (VM_RESERVED|VM_IO))
947 mm->reserved_vm += pages;
948}
949#endif /* CONFIG_PROC_FS */
950
951/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700952 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 */
954
ZhenwenXu48aae422009-01-06 14:40:21 -0800955unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 unsigned long len, unsigned long prot,
957 unsigned long flags, unsigned long pgoff)
958{
959 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 struct inode *inode;
961 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700963 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 /*
966 * Does the application expect PROT_READ to imply PROT_EXEC?
967 *
968 * (the exception is when the underlying filesystem is noexec
969 * mounted, in which case we dont add PROT_EXEC.)
970 */
971 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800972 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 prot |= PROT_EXEC;
974
975 if (!len)
976 return -EINVAL;
977
Eric Paris7cd94142007-11-26 18:47:40 -0500978 if (!(flags & MAP_FIXED))
979 addr = round_hint_to_min(addr);
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 /* Careful about overflows.. */
982 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500983 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 return -ENOMEM;
985
986 /* offset overflow? */
987 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
988 return -EOVERFLOW;
989
990 /* Too many mappings? */
991 if (mm->map_count > sysctl_max_map_count)
992 return -ENOMEM;
993
994 /* Obtain the address to map to. we verify (or select) it and ensure
995 * that it represents a valid section of the address space.
996 */
997 addr = get_unmapped_area(file, addr, len, pgoff, flags);
998 if (addr & ~PAGE_MASK)
999 return addr;
1000
1001 /* Do simple checking here so the lower-level routines won't have
1002 * to. we assume access permissions have been handled by the open
1003 * of the memory object, so we don't do any here.
1004 */
1005 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1006 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1007
Huang Shijiecdf7b342009-09-21 17:03:36 -07001008 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 if (!can_do_mlock())
1010 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /* mlock MCL_FUTURE? */
1013 if (vm_flags & VM_LOCKED) {
1014 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001015 locked = len >> PAGE_SHIFT;
1016 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001017 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001018 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1020 return -EAGAIN;
1021 }
1022
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001023 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 if (file) {
1026 switch (flags & MAP_TYPE) {
1027 case MAP_SHARED:
1028 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1029 return -EACCES;
1030
1031 /*
1032 * Make sure we don't allow writing to an append-only
1033 * file..
1034 */
1035 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1036 return -EACCES;
1037
1038 /*
1039 * Make sure there are no mandatory locks on the file.
1040 */
1041 if (locks_verify_locked(inode))
1042 return -EAGAIN;
1043
1044 vm_flags |= VM_SHARED | VM_MAYSHARE;
1045 if (!(file->f_mode & FMODE_WRITE))
1046 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1047
1048 /* fall through */
1049 case MAP_PRIVATE:
1050 if (!(file->f_mode & FMODE_READ))
1051 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001052 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001053 if (vm_flags & VM_EXEC)
1054 return -EPERM;
1055 vm_flags &= ~VM_MAYEXEC;
1056 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001057
1058 if (!file->f_op || !file->f_op->mmap)
1059 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 break;
1061
1062 default:
1063 return -EINVAL;
1064 }
1065 } else {
1066 switch (flags & MAP_TYPE) {
1067 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001068 /*
1069 * Ignore pgoff.
1070 */
1071 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 vm_flags |= VM_SHARED | VM_MAYSHARE;
1073 break;
1074 case MAP_PRIVATE:
1075 /*
1076 * Set pgoff according to addr for anon_vma.
1077 */
1078 pgoff = addr >> PAGE_SHIFT;
1079 break;
1080 default:
1081 return -EINVAL;
1082 }
1083 }
1084
Eric Parised032182007-06-28 15:55:21 -04001085 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 if (error)
1087 return error;
Eric Parised032182007-06-28 15:55:21 -04001088
Mel Gorman5a6fe122009-02-10 14:02:27 +00001089 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001090}
1091EXPORT_SYMBOL(do_mmap_pgoff);
1092
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001093SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1094 unsigned long, prot, unsigned long, flags,
1095 unsigned long, fd, unsigned long, pgoff)
1096{
1097 struct file *file = NULL;
1098 unsigned long retval = -EBADF;
1099
1100 if (!(flags & MAP_ANONYMOUS)) {
1101 if (unlikely(flags & MAP_HUGETLB))
1102 return -EINVAL;
1103 file = fget(fd);
1104 if (!file)
1105 goto out;
1106 } else if (flags & MAP_HUGETLB) {
1107 struct user_struct *user = NULL;
1108 /*
1109 * VM_NORESERVE is used because the reservations will be
1110 * taken when vm_ops->mmap() is called
1111 * A dummy user value is used because we are not locking
1112 * memory so no accounting is necessary
1113 */
1114 len = ALIGN(len, huge_page_size(&default_hstate));
1115 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1116 &user, HUGETLB_ANONHUGE_INODE);
1117 if (IS_ERR(file))
1118 return PTR_ERR(file);
1119 }
1120
1121 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1122
1123 down_write(&current->mm->mmap_sem);
1124 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1125 up_write(&current->mm->mmap_sem);
1126
1127 if (file)
1128 fput(file);
1129out:
1130 return retval;
1131}
1132
Christoph Hellwiga4679372010-03-10 15:21:15 -08001133#ifdef __ARCH_WANT_SYS_OLD_MMAP
1134struct mmap_arg_struct {
1135 unsigned long addr;
1136 unsigned long len;
1137 unsigned long prot;
1138 unsigned long flags;
1139 unsigned long fd;
1140 unsigned long offset;
1141};
1142
1143SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1144{
1145 struct mmap_arg_struct a;
1146
1147 if (copy_from_user(&a, arg, sizeof(a)))
1148 return -EFAULT;
1149 if (a.offset & ~PAGE_MASK)
1150 return -EINVAL;
1151
1152 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1153 a.offset >> PAGE_SHIFT);
1154}
1155#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1156
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001157/*
1158 * Some shared mappigns will want the pages marked read-only
1159 * to track write events. If so, we'll downgrade vm_page_prot
1160 * to the private version (using protection_map[] without the
1161 * VM_SHARED bit).
1162 */
1163int vma_wants_writenotify(struct vm_area_struct *vma)
1164{
1165 unsigned int vm_flags = vma->vm_flags;
1166
1167 /* If it was private or non-writable, the write bit is already clear */
1168 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1169 return 0;
1170
1171 /* The backer wishes to know when pages are first written to? */
1172 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1173 return 1;
1174
1175 /* The open routine did something to the protections already? */
1176 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001177 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001178 return 0;
1179
1180 /* Specialty mapping? */
1181 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1182 return 0;
1183
1184 /* Can the mapping track the dirty pages? */
1185 return vma->vm_file && vma->vm_file->f_mapping &&
1186 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1187}
1188
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001189/*
1190 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001191 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001192 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001193static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001194{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001195 /*
1196 * hugetlb has its own accounting separate from the core VM
1197 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1198 */
1199 if (file && is_file_hugepages(file))
1200 return 0;
1201
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001202 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1203}
1204
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001205unsigned long mmap_region(struct file *file, unsigned long addr,
1206 unsigned long len, unsigned long flags,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001207 unsigned int vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001208{
1209 struct mm_struct *mm = current->mm;
1210 struct vm_area_struct *vma, *prev;
1211 int correct_wcount = 0;
1212 int error;
1213 struct rb_node **rb_link, *rb_parent;
1214 unsigned long charged = 0;
1215 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 /* Clear old maps */
1218 error = -ENOMEM;
1219munmap_back:
1220 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1221 if (vma && vma->vm_start < addr + len) {
1222 if (do_munmap(mm, addr, len))
1223 return -ENOMEM;
1224 goto munmap_back;
1225 }
1226
1227 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001228 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 return -ENOMEM;
1230
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001231 /*
1232 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001233 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001234 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001235 if ((flags & MAP_NORESERVE)) {
1236 /* We honor MAP_NORESERVE if allowed to overcommit */
1237 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1238 vm_flags |= VM_NORESERVE;
1239
1240 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1241 if (file && is_file_hugepages(file))
1242 vm_flags |= VM_NORESERVE;
1243 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001244
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001245 /*
1246 * Private writable mapping: check memory availability
1247 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001248 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001249 charged = len >> PAGE_SHIFT;
1250 if (security_vm_enough_memory(charged))
1251 return -ENOMEM;
1252 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 }
1254
1255 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001256 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001258 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1259 if (vma)
1260 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
1262 /*
1263 * Determine the object being mapped and call the appropriate
1264 * specific mapper. the address has already been validated, but
1265 * not unmapped, but the maps are removed from the list.
1266 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001267 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 if (!vma) {
1269 error = -ENOMEM;
1270 goto unacct_error;
1271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
1273 vma->vm_mm = mm;
1274 vma->vm_start = addr;
1275 vma->vm_end = addr + len;
1276 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001277 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001279 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
1281 if (file) {
1282 error = -EINVAL;
1283 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1284 goto free_vma;
1285 if (vm_flags & VM_DENYWRITE) {
1286 error = deny_write_access(file);
1287 if (error)
1288 goto free_vma;
1289 correct_wcount = 1;
1290 }
1291 vma->vm_file = file;
1292 get_file(file);
1293 error = file->f_op->mmap(file, vma);
1294 if (error)
1295 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001296 if (vm_flags & VM_EXECUTABLE)
1297 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001298
1299 /* Can addr have changed??
1300 *
1301 * Answer: Yes, several device drivers can do it in their
1302 * f_op->mmap method. -DaveM
1303 */
1304 addr = vma->vm_start;
1305 pgoff = vma->vm_pgoff;
1306 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 } else if (vm_flags & VM_SHARED) {
1308 error = shmem_zero_setup(vma);
1309 if (error)
1310 goto free_vma;
1311 }
1312
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001313 if (vma_wants_writenotify(vma)) {
1314 pgprot_t pprot = vma->vm_page_prot;
1315
1316 /* Can vma->vm_page_prot have changed??
1317 *
1318 * Answer: Yes, drivers may have changed it in their
1319 * f_op->mmap method.
1320 *
1321 * Ensures that vmas marked as uncached stay that way.
1322 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001323 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001324 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1325 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1326 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001327
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001328 vma_link(mm, vma, prev, rb_link, rb_parent);
1329 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001330
1331 /* Once vma denies write, undo our temporary denial count */
1332 if (correct_wcount)
1333 atomic_inc(&inode->i_writecount);
1334out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001335 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001338 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001340 if (!mlock_vma_pages_range(vma, addr, addr + len))
1341 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001342 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001343 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 return addr;
1345
1346unmap_and_free_vma:
1347 if (correct_wcount)
1348 atomic_inc(&inode->i_writecount);
1349 vma->vm_file = NULL;
1350 fput(file);
1351
1352 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001353 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1354 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355free_vma:
1356 kmem_cache_free(vm_area_cachep, vma);
1357unacct_error:
1358 if (charged)
1359 vm_unacct_memory(charged);
1360 return error;
1361}
1362
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363/* Get an address range which is currently unmapped.
1364 * For shmat() with addr=0.
1365 *
1366 * Ugly calling convention alert:
1367 * Return value with the low bits set means error value,
1368 * ie
1369 * if (ret & ~PAGE_MASK)
1370 * error = ret;
1371 *
1372 * This function "knows" that -ENOMEM has the bits set.
1373 */
1374#ifndef HAVE_ARCH_UNMAPPED_AREA
1375unsigned long
1376arch_get_unmapped_area(struct file *filp, unsigned long addr,
1377 unsigned long len, unsigned long pgoff, unsigned long flags)
1378{
1379 struct mm_struct *mm = current->mm;
1380 struct vm_area_struct *vma;
1381 unsigned long start_addr;
1382
1383 if (len > TASK_SIZE)
1384 return -ENOMEM;
1385
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001386 if (flags & MAP_FIXED)
1387 return addr;
1388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (addr) {
1390 addr = PAGE_ALIGN(addr);
1391 vma = find_vma(mm, addr);
1392 if (TASK_SIZE - len >= addr &&
1393 (!vma || addr + len <= vma->vm_start))
1394 return addr;
1395 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001396 if (len > mm->cached_hole_size) {
1397 start_addr = addr = mm->free_area_cache;
1398 } else {
1399 start_addr = addr = TASK_UNMAPPED_BASE;
1400 mm->cached_hole_size = 0;
1401 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403full_search:
1404 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1405 /* At this point: (!vma || addr < vma->vm_end). */
1406 if (TASK_SIZE - len < addr) {
1407 /*
1408 * Start a new search - just in case we missed
1409 * some holes.
1410 */
1411 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001412 addr = TASK_UNMAPPED_BASE;
1413 start_addr = addr;
1414 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 goto full_search;
1416 }
1417 return -ENOMEM;
1418 }
1419 if (!vma || addr + len <= vma->vm_start) {
1420 /*
1421 * Remember the place where we stopped the search:
1422 */
1423 mm->free_area_cache = addr + len;
1424 return addr;
1425 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001426 if (addr + mm->cached_hole_size < vma->vm_start)
1427 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 addr = vma->vm_end;
1429 }
1430}
1431#endif
1432
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001433void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434{
1435 /*
1436 * Is this a new hole at the lowest possible address?
1437 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001438 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1439 mm->free_area_cache = addr;
1440 mm->cached_hole_size = ~0UL;
1441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
1443
1444/*
1445 * This mmap-allocator allocates new areas top-down from below the
1446 * stack's low limit (the base):
1447 */
1448#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1449unsigned long
1450arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1451 const unsigned long len, const unsigned long pgoff,
1452 const unsigned long flags)
1453{
1454 struct vm_area_struct *vma;
1455 struct mm_struct *mm = current->mm;
1456 unsigned long addr = addr0;
1457
1458 /* requested length too big for entire address space */
1459 if (len > TASK_SIZE)
1460 return -ENOMEM;
1461
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001462 if (flags & MAP_FIXED)
1463 return addr;
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 /* requesting a specific address */
1466 if (addr) {
1467 addr = PAGE_ALIGN(addr);
1468 vma = find_vma(mm, addr);
1469 if (TASK_SIZE - len >= addr &&
1470 (!vma || addr + len <= vma->vm_start))
1471 return addr;
1472 }
1473
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001474 /* check if free_area_cache is useful for us */
1475 if (len <= mm->cached_hole_size) {
1476 mm->cached_hole_size = 0;
1477 mm->free_area_cache = mm->mmap_base;
1478 }
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 /* either no address requested or can't fit in requested address hole */
1481 addr = mm->free_area_cache;
1482
1483 /* make sure it can fit in the remaining address space */
Linus Torvalds49a43872005-05-18 15:39:33 -07001484 if (addr > len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 vma = find_vma(mm, addr-len);
1486 if (!vma || addr <= vma->vm_start)
1487 /* remember the address as a hint for next time */
1488 return (mm->free_area_cache = addr-len);
1489 }
1490
Chris Wright73219d12005-06-21 17:14:52 -07001491 if (mm->mmap_base < len)
1492 goto bottomup;
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 addr = mm->mmap_base-len;
1495
1496 do {
1497 /*
1498 * Lookup failure means no vma is above this address,
1499 * else if new region fits below vma->vm_start,
1500 * return with success:
1501 */
1502 vma = find_vma(mm, addr);
1503 if (!vma || addr+len <= vma->vm_start)
1504 /* remember the address as a hint for next time */
1505 return (mm->free_area_cache = addr);
1506
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001507 /* remember the largest hole we saw so far */
1508 if (addr + mm->cached_hole_size < vma->vm_start)
1509 mm->cached_hole_size = vma->vm_start - addr;
1510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 /* try just below the current vma->vm_start */
1512 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001513 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Chris Wright73219d12005-06-21 17:14:52 -07001515bottomup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 /*
1517 * A failed mmap() very likely causes application failure,
1518 * so fall back to the bottom-up function here. This scenario
1519 * can happen with large stack limits and large mmap()
1520 * allocations.
1521 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001522 mm->cached_hole_size = ~0UL;
1523 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1525 /*
1526 * Restore the topdown base:
1527 */
1528 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001529 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531 return addr;
1532}
1533#endif
1534
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001535void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
1537 /*
1538 * Is this a new hole at the highest possible address?
1539 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001540 if (addr > mm->free_area_cache)
1541 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
1543 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001544 if (mm->free_area_cache > mm->mmap_base)
1545 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546}
1547
1548unsigned long
1549get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1550 unsigned long pgoff, unsigned long flags)
1551{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001552 unsigned long (*get_area)(struct file *, unsigned long,
1553 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Al Viro9206de92009-12-03 15:23:11 -05001555 unsigned long error = arch_mmap_check(addr, len, flags);
1556 if (error)
1557 return error;
1558
1559 /* Careful about overflows.. */
1560 if (len > TASK_SIZE)
1561 return -ENOMEM;
1562
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001563 get_area = current->mm->get_unmapped_area;
1564 if (file && file->f_op && file->f_op->get_unmapped_area)
1565 get_area = file->f_op->get_unmapped_area;
1566 addr = get_area(file, addr, len, pgoff, flags);
1567 if (IS_ERR_VALUE(addr))
1568 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001570 if (addr > TASK_SIZE - len)
1571 return -ENOMEM;
1572 if (addr & ~PAGE_MASK)
1573 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001574
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001575 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576}
1577
1578EXPORT_SYMBOL(get_unmapped_area);
1579
1580/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001581struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582{
1583 struct vm_area_struct *vma = NULL;
1584
1585 if (mm) {
1586 /* Check the cache first. */
1587 /* (Cache hit rate is typically around 35%.) */
1588 vma = mm->mmap_cache;
1589 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1590 struct rb_node * rb_node;
1591
1592 rb_node = mm->mm_rb.rb_node;
1593 vma = NULL;
1594
1595 while (rb_node) {
1596 struct vm_area_struct * vma_tmp;
1597
1598 vma_tmp = rb_entry(rb_node,
1599 struct vm_area_struct, vm_rb);
1600
1601 if (vma_tmp->vm_end > addr) {
1602 vma = vma_tmp;
1603 if (vma_tmp->vm_start <= addr)
1604 break;
1605 rb_node = rb_node->rb_left;
1606 } else
1607 rb_node = rb_node->rb_right;
1608 }
1609 if (vma)
1610 mm->mmap_cache = vma;
1611 }
1612 }
1613 return vma;
1614}
1615
1616EXPORT_SYMBOL(find_vma);
1617
1618/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1619struct vm_area_struct *
1620find_vma_prev(struct mm_struct *mm, unsigned long addr,
1621 struct vm_area_struct **pprev)
1622{
1623 struct vm_area_struct *vma = NULL, *prev = NULL;
ZhenwenXu48aae422009-01-06 14:40:21 -08001624 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 if (!mm)
1626 goto out;
1627
1628 /* Guard against addr being lower than the first VMA */
1629 vma = mm->mmap;
1630
1631 /* Go through the RB tree quickly. */
1632 rb_node = mm->mm_rb.rb_node;
1633
1634 while (rb_node) {
1635 struct vm_area_struct *vma_tmp;
1636 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1637
1638 if (addr < vma_tmp->vm_end) {
1639 rb_node = rb_node->rb_left;
1640 } else {
1641 prev = vma_tmp;
1642 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1643 break;
1644 rb_node = rb_node->rb_right;
1645 }
1646 }
1647
1648out:
1649 *pprev = prev;
1650 return prev ? prev->vm_next : vma;
1651}
1652
1653/*
1654 * Verify that the stack growth is acceptable and
1655 * update accounting. This is shared with both the
1656 * grow-up and grow-down cases.
1657 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001658static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659{
1660 struct mm_struct *mm = vma->vm_mm;
1661 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001662 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
1664 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001665 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 return -ENOMEM;
1667
1668 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001669 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 return -ENOMEM;
1671
1672 /* mlock limit tests */
1673 if (vma->vm_flags & VM_LOCKED) {
1674 unsigned long locked;
1675 unsigned long limit;
1676 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001677 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1678 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 if (locked > limit && !capable(CAP_IPC_LOCK))
1680 return -ENOMEM;
1681 }
1682
Adam Litke0d59a012007-01-30 14:35:39 -08001683 /* Check to ensure the stack will not grow into a hugetlb-only region */
1684 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1685 vma->vm_end - size;
1686 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1687 return -EFAULT;
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 /*
1690 * Overcommit.. This must be the final test, as it will
1691 * update security statistics.
1692 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001693 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 return -ENOMEM;
1695
1696 /* Ok, everything looks good - let it rip */
1697 mm->total_vm += grow;
1698 if (vma->vm_flags & VM_LOCKED)
1699 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001700 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 return 0;
1702}
1703
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001704#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001706 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1707 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 */
Matthew Wilcox9ab88512005-11-18 16:16:42 -05001709#ifndef CONFIG_IA64
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001710static
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001711#endif
1712int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713{
1714 int error;
1715
1716 if (!(vma->vm_flags & VM_GROWSUP))
1717 return -EFAULT;
1718
1719 /*
1720 * We must make sure the anon_vma is allocated
1721 * so that the anon_vma locking is not a noop.
1722 */
1723 if (unlikely(anon_vma_prepare(vma)))
1724 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001725 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
1727 /*
1728 * vma->vm_start/vm_end cannot change under us because the caller
1729 * is required to hold the mmap_sem in read mode. We need the
1730 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001731 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 */
Helge Deller06b32f32006-12-19 19:28:33 +01001733 if (address < PAGE_ALIGN(address+4))
1734 address = PAGE_ALIGN(address+4);
1735 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001736 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001737 return -ENOMEM;
1738 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 error = 0;
1740
1741 /* Somebody else might have raced and expanded it already */
1742 if (address > vma->vm_end) {
1743 unsigned long size, grow;
1744
1745 size = address - vma->vm_start;
1746 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1747
1748 error = acct_stack_growth(vma, size, grow);
Eric B Munson3af9e852010-05-18 15:30:49 +01001749 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 vma->vm_end = address;
Eric B Munson3af9e852010-05-18 15:30:49 +01001751 perf_event_mmap(vma);
1752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001754 vma_unlock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 return error;
1756}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001757#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759/*
1760 * vma is the first one with address < vma->vm_start. Have to extend vma.
1761 */
Denys Vlasenkocb8f4882008-10-18 20:27:01 -07001762static int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001763 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764{
1765 int error;
1766
1767 /*
1768 * We must make sure the anon_vma is allocated
1769 * so that the anon_vma locking is not a noop.
1770 */
1771 if (unlikely(anon_vma_prepare(vma)))
1772 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001773
1774 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001775 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001776 if (error)
1777 return error;
1778
Rik van Rielbb4a3402010-08-09 17:18:37 -07001779 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
1781 /*
1782 * vma->vm_start/vm_end cannot change under us because the caller
1783 * is required to hold the mmap_sem in read mode. We need the
1784 * anon_vma lock to serialize against concurrent expand_stacks.
1785 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787 /* Somebody else might have raced and expanded it already */
1788 if (address < vma->vm_start) {
1789 unsigned long size, grow;
1790
1791 size = vma->vm_end - address;
1792 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1793
1794 error = acct_stack_growth(vma, size, grow);
1795 if (!error) {
1796 vma->vm_start = address;
1797 vma->vm_pgoff -= grow;
Eric B Munson3af9e852010-05-18 15:30:49 +01001798 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 }
1800 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001801 vma_unlock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 return error;
1803}
1804
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001805int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1806{
1807 return expand_downwards(vma, address);
1808}
1809
1810#ifdef CONFIG_STACK_GROWSUP
1811int expand_stack(struct vm_area_struct *vma, unsigned long address)
1812{
1813 return expand_upwards(vma, address);
1814}
1815
1816struct vm_area_struct *
1817find_extend_vma(struct mm_struct *mm, unsigned long addr)
1818{
1819 struct vm_area_struct *vma, *prev;
1820
1821 addr &= PAGE_MASK;
1822 vma = find_vma_prev(mm, addr, &prev);
1823 if (vma && (vma->vm_start <= addr))
1824 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001825 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001826 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001827 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001828 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001829 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001830 return prev;
1831}
1832#else
1833int expand_stack(struct vm_area_struct *vma, unsigned long address)
1834{
1835 return expand_downwards(vma, address);
1836}
1837
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838struct vm_area_struct *
1839find_extend_vma(struct mm_struct * mm, unsigned long addr)
1840{
1841 struct vm_area_struct * vma;
1842 unsigned long start;
1843
1844 addr &= PAGE_MASK;
1845 vma = find_vma(mm,addr);
1846 if (!vma)
1847 return NULL;
1848 if (vma->vm_start <= addr)
1849 return vma;
1850 if (!(vma->vm_flags & VM_GROWSDOWN))
1851 return NULL;
1852 start = vma->vm_start;
1853 if (expand_stack(vma, addr))
1854 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001855 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001856 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001857 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 return vma;
1859}
1860#endif
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001863 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001865 *
1866 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001868static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001870 /* Update high watermark before we lower total_vm */
1871 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001873 long nrpages = vma_pages(vma);
1874
1875 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001876 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001877 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001878 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 validate_mm(mm);
1880}
1881
1882/*
1883 * Get rid of page table information in the indicated region.
1884 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001885 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 */
1887static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001888 struct vm_area_struct *vma, struct vm_area_struct *prev,
1889 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001891 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 struct mmu_gather *tlb;
1893 unsigned long nr_accounted = 0;
1894
1895 lru_add_drain();
1896 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001897 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001898 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 vm_unacct_memory(nr_accounted);
Jan Beulich42b77722008-07-23 21:27:10 -07001900 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001901 next? next->vm_start: 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 tlb_finish_mmu(tlb, start, end);
1903}
1904
1905/*
1906 * Create a list of vma's touched by the unmap, removing them from the mm's
1907 * vma list as we go..
1908 */
1909static void
1910detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1911 struct vm_area_struct *prev, unsigned long end)
1912{
1913 struct vm_area_struct **insertion_point;
1914 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001915 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
1917 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1918 do {
1919 rb_erase(&vma->vm_rb, &mm->mm_rb);
1920 mm->map_count--;
1921 tail_vma = vma;
1922 vma = vma->vm_next;
1923 } while (vma && vma->vm_start < end);
1924 *insertion_point = vma;
1925 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001926 if (mm->unmap_area == arch_unmap_area)
1927 addr = prev ? prev->vm_end : mm->mmap_base;
1928 else
1929 addr = vma ? vma->vm_start : mm->mmap_base;
1930 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 mm->mmap_cache = NULL; /* Kill the cache. */
1932}
1933
1934/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001935 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1936 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001938static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 unsigned long addr, int new_below)
1940{
1941 struct mempolicy *pol;
1942 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001943 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
Andi Kleena5516432008-07-23 21:27:41 -07001945 if (is_vm_hugetlb_page(vma) && (addr &
1946 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 return -EINVAL;
1948
Christoph Lametere94b1762006-12-06 20:33:17 -08001949 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001951 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /* most fields are the same, copy all, and then fixup */
1954 *new = *vma;
1955
Rik van Riel5beb4932010-03-05 13:42:07 -08001956 INIT_LIST_HEAD(&new->anon_vma_chain);
1957
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 if (new_below)
1959 new->vm_end = addr;
1960 else {
1961 new->vm_start = addr;
1962 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1963 }
1964
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001965 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001967 err = PTR_ERR(pol);
1968 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 }
1970 vma_set_policy(new, pol);
1971
Rik van Riel5beb4932010-03-05 13:42:07 -08001972 if (anon_vma_clone(new, vma))
1973 goto out_free_mpol;
1974
Matt Helsley925d1c42008-04-29 01:01:36 -07001975 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001977 if (vma->vm_flags & VM_EXECUTABLE)
1978 added_exe_file_vma(mm);
1979 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
1981 if (new->vm_ops && new->vm_ops->open)
1982 new->vm_ops->open(new);
1983
1984 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08001985 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1987 else
Rik van Riel5beb4932010-03-05 13:42:07 -08001988 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
Rik van Riel5beb4932010-03-05 13:42:07 -08001990 /* Success. */
1991 if (!err)
1992 return 0;
1993
1994 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04001995 if (new->vm_ops && new->vm_ops->close)
1996 new->vm_ops->close(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08001997 if (new->vm_file) {
1998 if (vma->vm_flags & VM_EXECUTABLE)
1999 removed_exe_file_vma(mm);
2000 fput(new->vm_file);
2001 }
2002 out_free_mpol:
2003 mpol_put(pol);
2004 out_free_vma:
2005 kmem_cache_free(vm_area_cachep, new);
2006 out_err:
2007 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008}
2009
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002010/*
2011 * Split a vma into two pieces at address 'addr', a new vma is allocated
2012 * either for the first part or the tail.
2013 */
2014int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2015 unsigned long addr, int new_below)
2016{
2017 if (mm->map_count >= sysctl_max_map_count)
2018 return -ENOMEM;
2019
2020 return __split_vma(mm, vma, addr, new_below);
2021}
2022
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023/* Munmap is split into 2 main parts -- this part which finds
2024 * what needs doing, and the areas themselves, which do the
2025 * work. This now handles partial unmappings.
2026 * Jeremy Fitzhardinge <jeremy@goop.org>
2027 */
2028int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2029{
2030 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002031 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
2033 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2034 return -EINVAL;
2035
2036 if ((len = PAGE_ALIGN(len)) == 0)
2037 return -EINVAL;
2038
2039 /* Find the first overlapping VMA */
Hugh Dickins146425a2005-04-19 13:29:18 -07002040 vma = find_vma_prev(mm, start, &prev);
2041 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 return 0;
Hugh Dickins146425a2005-04-19 13:29:18 -07002043 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 /* if it doesn't overlap, we have nothing.. */
2046 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002047 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return 0;
2049
2050 /*
2051 * If we need to split any vma, do it now to save pain later.
2052 *
2053 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2054 * unmapped vm_area_struct will remain in use: so lower split_vma
2055 * places tmp vma above, and higher split_vma places tmp vma below.
2056 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002057 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002058 int error;
2059
2060 /*
2061 * Make sure that map_count on return from munmap() will
2062 * not exceed its limit; but let map_count go just above
2063 * its limit temporarily, to help free resources as expected.
2064 */
2065 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2066 return -ENOMEM;
2067
2068 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (error)
2070 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002071 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
2073
2074 /* Does it split the last one? */
2075 last = find_vma(mm, end);
2076 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002077 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (error)
2079 return error;
2080 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002081 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
2083 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002084 * unlock any mlock()ed ranges before detaching vmas
2085 */
2086 if (mm->locked_vm) {
2087 struct vm_area_struct *tmp = vma;
2088 while (tmp && tmp->vm_start < end) {
2089 if (tmp->vm_flags & VM_LOCKED) {
2090 mm->locked_vm -= vma_pages(tmp);
2091 munlock_vma_pages_all(tmp);
2092 }
2093 tmp = tmp->vm_next;
2094 }
2095 }
2096
2097 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 * Remove the vma's, and unmap the actual pages
2099 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002100 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2101 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
2103 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002104 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106 return 0;
2107}
2108
2109EXPORT_SYMBOL(do_munmap);
2110
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002111SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112{
2113 int ret;
2114 struct mm_struct *mm = current->mm;
2115
2116 profile_munmap(addr);
2117
2118 down_write(&mm->mmap_sem);
2119 ret = do_munmap(mm, addr, len);
2120 up_write(&mm->mmap_sem);
2121 return ret;
2122}
2123
2124static inline void verify_mm_writelocked(struct mm_struct *mm)
2125{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002126#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2128 WARN_ON(1);
2129 up_read(&mm->mmap_sem);
2130 }
2131#endif
2132}
2133
2134/*
2135 * this is really a simplified "do_mmap". it only handles
2136 * anonymous maps. eventually we may be able to do some
2137 * brk-specific accounting here.
2138 */
2139unsigned long do_brk(unsigned long addr, unsigned long len)
2140{
2141 struct mm_struct * mm = current->mm;
2142 struct vm_area_struct * vma, * prev;
2143 unsigned long flags;
2144 struct rb_node ** rb_link, * rb_parent;
2145 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002146 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
2148 len = PAGE_ALIGN(len);
2149 if (!len)
2150 return addr;
2151
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002152 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002153 if (error)
2154 return error;
2155
Kirill Korotaev3a459752006-09-07 14:17:04 +04002156 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2157
Al Viro2c6a10162009-12-03 19:40:46 -05002158 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2159 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002160 return error;
2161
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 /*
2163 * mlock MCL_FUTURE?
2164 */
2165 if (mm->def_flags & VM_LOCKED) {
2166 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002167 locked = len >> PAGE_SHIFT;
2168 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002169 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002170 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2172 return -EAGAIN;
2173 }
2174
2175 /*
2176 * mm->mmap_sem is required to protect against another thread
2177 * changing the mappings in case we sleep.
2178 */
2179 verify_mm_writelocked(mm);
2180
2181 /*
2182 * Clear old maps. this also does some error checking for us
2183 */
2184 munmap_back:
2185 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2186 if (vma && vma->vm_start < addr + len) {
2187 if (do_munmap(mm, addr, len))
2188 return -ENOMEM;
2189 goto munmap_back;
2190 }
2191
2192 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002193 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 return -ENOMEM;
2195
2196 if (mm->map_count > sysctl_max_map_count)
2197 return -ENOMEM;
2198
2199 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2200 return -ENOMEM;
2201
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002203 vma = vma_merge(mm, prev, addr, addr + len, flags,
2204 NULL, NULL, pgoff, NULL);
2205 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 goto out;
2207
2208 /*
2209 * create a vma struct for an anonymous mapping
2210 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002211 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 if (!vma) {
2213 vm_unacct_memory(len >> PAGE_SHIFT);
2214 return -ENOMEM;
2215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216
Rik van Riel5beb4932010-03-05 13:42:07 -08002217 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 vma->vm_mm = mm;
2219 vma->vm_start = addr;
2220 vma->vm_end = addr + len;
2221 vma->vm_pgoff = pgoff;
2222 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002223 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 vma_link(mm, vma, prev, rb_link, rb_parent);
2225out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002226 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 mm->total_vm += len >> PAGE_SHIFT;
2228 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002229 if (!mlock_vma_pages_range(vma, addr, addr + len))
2230 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 }
2232 return addr;
2233}
2234
2235EXPORT_SYMBOL(do_brk);
2236
2237/* Release all mmaps. */
2238void exit_mmap(struct mm_struct *mm)
2239{
2240 struct mmu_gather *tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002241 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 unsigned long nr_accounted = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07002243 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002245 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002246 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002247
Rik van Rielba470de2008-10-18 20:26:50 -07002248 if (mm->locked_vm) {
2249 vma = mm->mmap;
2250 while (vma) {
2251 if (vma->vm_flags & VM_LOCKED)
2252 munlock_vma_pages_all(vma);
2253 vma = vma->vm_next;
2254 }
2255 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002256
2257 arch_exit_mmap(mm);
2258
Rik van Rielba470de2008-10-18 20:26:50 -07002259 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002260 if (!vma) /* Can happen if dup_mmap() received an OOM */
2261 return;
2262
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 flush_cache_mm(mm);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002265 tlb = tlb_gather_mmu(mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002266 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002267 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Hugh Dickins508034a2005-10-29 18:16:30 -07002268 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 vm_unacct_memory(nr_accounted);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002270
Jan Beulich42b77722008-07-23 21:27:10 -07002271 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
Hugh Dickinsee39b372005-04-19 13:29:15 -07002272 tlb_finish_mmu(tlb, 0, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002275 * Walk the list again, actually closing and freeing it,
2276 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002278 while (vma)
2279 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002280
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002281 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282}
2283
2284/* Insert vm structure into process list sorted by address
2285 * and into the inode's i_mmap tree. If vm_file is non-NULL
2286 * then i_mmap_lock is taken here.
2287 */
2288int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2289{
2290 struct vm_area_struct * __vma, * prev;
2291 struct rb_node ** rb_link, * rb_parent;
2292
2293 /*
2294 * The vm_pgoff of a purely anonymous vma should be irrelevant
2295 * until its first write fault, when page's anon_vma and index
2296 * are set. But now set the vm_pgoff it will almost certainly
2297 * end up with (unless mremap moves it elsewhere before that
2298 * first wfault), so /proc/pid/maps tells a consistent story.
2299 *
2300 * By setting it to reflect the virtual start address of the
2301 * vma, merges and splits can happen in a seamless way, just
2302 * using the existing file pgoff checks and manipulations.
2303 * Similarly in do_mmap_pgoff and in do_brk.
2304 */
2305 if (!vma->vm_file) {
2306 BUG_ON(vma->anon_vma);
2307 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2308 }
2309 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2310 if (__vma && __vma->vm_start < vma->vm_end)
2311 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002312 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002313 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002314 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 vma_link(mm, vma, prev, rb_link, rb_parent);
2316 return 0;
2317}
2318
2319/*
2320 * Copy the vma structure to a new location in the same mm,
2321 * prior to moving page table entries, to effect an mremap move.
2322 */
2323struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2324 unsigned long addr, unsigned long len, pgoff_t pgoff)
2325{
2326 struct vm_area_struct *vma = *vmap;
2327 unsigned long vma_start = vma->vm_start;
2328 struct mm_struct *mm = vma->vm_mm;
2329 struct vm_area_struct *new_vma, *prev;
2330 struct rb_node **rb_link, *rb_parent;
2331 struct mempolicy *pol;
2332
2333 /*
2334 * If anonymous vma has not yet been faulted, update new pgoff
2335 * to match new location, to increase its chance of merging.
2336 */
2337 if (!vma->vm_file && !vma->anon_vma)
2338 pgoff = addr >> PAGE_SHIFT;
2339
2340 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2341 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2342 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2343 if (new_vma) {
2344 /*
2345 * Source vma may have been merged into new_vma
2346 */
2347 if (vma_start >= new_vma->vm_start &&
2348 vma_start < new_vma->vm_end)
2349 *vmap = new_vma;
2350 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002351 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 if (new_vma) {
2353 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002354 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002355 if (IS_ERR(pol))
2356 goto out_free_vma;
2357 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2358 if (anon_vma_clone(new_vma, vma))
2359 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 vma_set_policy(new_vma, pol);
2361 new_vma->vm_start = addr;
2362 new_vma->vm_end = addr + len;
2363 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002364 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002366 if (vma->vm_flags & VM_EXECUTABLE)
2367 added_exe_file_vma(mm);
2368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 if (new_vma->vm_ops && new_vma->vm_ops->open)
2370 new_vma->vm_ops->open(new_vma);
2371 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2372 }
2373 }
2374 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002375
2376 out_free_mempol:
2377 mpol_put(pol);
2378 out_free_vma:
2379 kmem_cache_free(vm_area_cachep, new_vma);
2380 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002382
2383/*
2384 * Return true if the calling process may expand its vm space by the passed
2385 * number of pages
2386 */
2387int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2388{
2389 unsigned long cur = mm->total_vm; /* pages */
2390 unsigned long lim;
2391
Jiri Slaby59e99e52010-03-05 13:41:44 -08002392 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002393
2394 if (cur + npages > lim)
2395 return 0;
2396 return 1;
2397}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002398
2399
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002400static int special_mapping_fault(struct vm_area_struct *vma,
2401 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002402{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002403 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002404 struct page **pages;
2405
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002406 /*
2407 * special mappings have no vm_file, and in that case, the mm
2408 * uses vm_pgoff internally. So we have to subtract it from here.
2409 * We are allowed to do this because we are the mm; do not copy
2410 * this code into drivers!
2411 */
2412 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002413
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002414 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2415 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002416
2417 if (*pages) {
2418 struct page *page = *pages;
2419 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002420 vmf->page = page;
2421 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002422 }
2423
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002424 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002425}
2426
2427/*
2428 * Having a close hook prevents vma merging regardless of flags.
2429 */
2430static void special_mapping_close(struct vm_area_struct *vma)
2431{
2432}
2433
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002434static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002435 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002436 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002437};
2438
2439/*
2440 * Called with mm->mmap_sem held for writing.
2441 * Insert a new vma covering the given region, with the given flags.
2442 * Its pages are supplied by the given array of struct page *.
2443 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2444 * The region past the last page supplied will always produce SIGBUS.
2445 * The array pointer and the pages it points to are assumed to stay alive
2446 * for as long as this mapping might exist.
2447 */
2448int install_special_mapping(struct mm_struct *mm,
2449 unsigned long addr, unsigned long len,
2450 unsigned long vm_flags, struct page **pages)
2451{
2452 struct vm_area_struct *vma;
2453
2454 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2455 if (unlikely(vma == NULL))
2456 return -ENOMEM;
2457
Rik van Riel5beb4932010-03-05 13:42:07 -08002458 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002459 vma->vm_mm = mm;
2460 vma->vm_start = addr;
2461 vma->vm_end = addr + len;
2462
Nick Piggin2f987352008-02-02 03:08:53 +01002463 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002464 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002465
2466 vma->vm_ops = &special_mapping_vmops;
2467 vma->vm_private_data = pages;
2468
2469 if (unlikely(insert_vm_struct(mm, vma))) {
2470 kmem_cache_free(vm_area_cachep, vma);
2471 return -ENOMEM;
2472 }
2473
2474 mm->total_vm += len >> PAGE_SHIFT;
2475
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002476 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002477
Roland McGrathfa5dc222007-02-08 14:20:41 -08002478 return 0;
2479}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002480
2481static DEFINE_MUTEX(mm_all_locks_mutex);
2482
Peter Zijlstra454ed842008-08-11 09:30:25 +02002483static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002484{
Rik van Riel012f1802010-08-09 17:18:40 -07002485 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002486 /*
2487 * The LSB of head.next can't change from under us
2488 * because we hold the mm_all_locks_mutex.
2489 */
Rik van Riel012f1802010-08-09 17:18:40 -07002490 spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002491 /*
2492 * We can safely modify head.next after taking the
Rik van Riel012f1802010-08-09 17:18:40 -07002493 * anon_vma->root->lock. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002494 * the same anon_vma we won't take it again.
2495 *
2496 * No need of atomic instructions here, head.next
2497 * can't change from under us thanks to the
Rik van Riel012f1802010-08-09 17:18:40 -07002498 * anon_vma->root->lock.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002499 */
2500 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002501 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002502 BUG();
2503 }
2504}
2505
Peter Zijlstra454ed842008-08-11 09:30:25 +02002506static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002507{
2508 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2509 /*
2510 * AS_MM_ALL_LOCKS can't change from under us because
2511 * we hold the mm_all_locks_mutex.
2512 *
2513 * Operations on ->flags have to be atomic because
2514 * even if AS_MM_ALL_LOCKS is stable thanks to the
2515 * mm_all_locks_mutex, there may be other cpus
2516 * changing other bitflags in parallel to us.
2517 */
2518 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2519 BUG();
Peter Zijlstra454ed842008-08-11 09:30:25 +02002520 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002521 }
2522}
2523
2524/*
2525 * This operation locks against the VM for all pte/vma/mm related
2526 * operations that could ever happen on a certain mm. This includes
2527 * vmtruncate, try_to_unmap, and all page faults.
2528 *
2529 * The caller must take the mmap_sem in write mode before calling
2530 * mm_take_all_locks(). The caller isn't allowed to release the
2531 * mmap_sem until mm_drop_all_locks() returns.
2532 *
2533 * mmap_sem in write mode is required in order to block all operations
2534 * that could modify pagetables and free pages without need of
2535 * altering the vma layout (for example populate_range() with
2536 * nonlinear vmas). It's also needed in write mode to avoid new
2537 * anon_vmas to be associated with existing vmas.
2538 *
2539 * A single task can't take more than one mm_take_all_locks() in a row
2540 * or it would deadlock.
2541 *
2542 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2543 * mapping->flags avoid to take the same lock twice, if more than one
2544 * vma in this mm is backed by the same anon_vma or address_space.
2545 *
2546 * We can take all the locks in random order because the VM code
2547 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2548 * takes more than one of them in a row. Secondly we're protected
2549 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2550 *
2551 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2552 * that may have to take thousand of locks.
2553 *
2554 * mm_take_all_locks() can fail if it's interrupted by signals.
2555 */
2556int mm_take_all_locks(struct mm_struct *mm)
2557{
2558 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002559 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002560 int ret = -EINTR;
2561
2562 BUG_ON(down_read_trylock(&mm->mmap_sem));
2563
2564 mutex_lock(&mm_all_locks_mutex);
2565
2566 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2567 if (signal_pending(current))
2568 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002569 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002570 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002571 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002572
2573 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2574 if (signal_pending(current))
2575 goto out_unlock;
2576 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002577 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2578 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002579 }
2580
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002581 ret = 0;
2582
2583out_unlock:
2584 if (ret)
2585 mm_drop_all_locks(mm);
2586
2587 return ret;
2588}
2589
2590static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2591{
Rik van Riel012f1802010-08-09 17:18:40 -07002592 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002593 /*
2594 * The LSB of head.next can't change to 0 from under
2595 * us because we hold the mm_all_locks_mutex.
2596 *
2597 * We must however clear the bitflag before unlocking
2598 * the vma so the users using the anon_vma->head will
2599 * never see our bitflag.
2600 *
2601 * No need of atomic instructions here, head.next
2602 * can't change from under us until we release the
Rik van Riel012f1802010-08-09 17:18:40 -07002603 * anon_vma->root->lock.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002604 */
2605 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002606 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002607 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002608 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002609 }
2610}
2611
2612static void vm_unlock_mapping(struct address_space *mapping)
2613{
2614 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2615 /*
2616 * AS_MM_ALL_LOCKS can't change to 0 from under us
2617 * because we hold the mm_all_locks_mutex.
2618 */
2619 spin_unlock(&mapping->i_mmap_lock);
2620 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2621 &mapping->flags))
2622 BUG();
2623 }
2624}
2625
2626/*
2627 * The mmap_sem cannot be released by the caller until
2628 * mm_drop_all_locks() returns.
2629 */
2630void mm_drop_all_locks(struct mm_struct *mm)
2631{
2632 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002633 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002634
2635 BUG_ON(down_read_trylock(&mm->mmap_sem));
2636 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2637
2638 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2639 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002640 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2641 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002642 if (vma->vm_file && vma->vm_file->f_mapping)
2643 vm_unlock_mapping(vma->vm_file->f_mapping);
2644 }
2645
2646 mutex_unlock(&mm_all_locks_mutex);
2647}
David Howells8feae132009-01-08 12:04:47 +00002648
2649/*
2650 * initialise the VMA slab
2651 */
2652void __init mmap_init(void)
2653{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002654 int ret;
2655
2656 ret = percpu_counter_init(&vm_committed_as, 0);
2657 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002658}