blob: 4a9c2a391e28efe523e1a3aaaf9ab06d24627e86 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
Alan Cox046c6882009-01-05 14:06:29 +00006 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 */
8
9#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070010#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/mm.h>
12#include <linux/shm.h>
13#include <linux/mman.h>
14#include <linux/pagemap.h>
15#include <linux/swap.h>
16#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/init.h>
19#include <linux/file.h>
20#include <linux/fs.h>
21#include <linux/personality.h>
22#include <linux/security.h>
23#include <linux/hugetlb.h>
24#include <linux/profile.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040025#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/mount.h>
27#include <linux/mempolicy.h>
28#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070029#include <linux/mmu_notifier.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020030#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040031#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080032#include <linux/khugepaged.h>
Srikar Dronamraju2b144492012-02-09 14:56:42 +053033#include <linux/uprobes.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35#include <asm/uaccess.h>
36#include <asm/cacheflush.h>
37#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020038#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Jan Beulich42b77722008-07-23 21:27:10 -070040#include "internal.h"
41
Kirill Korotaev3a459752006-09-07 14:17:04 +040042#ifndef arch_mmap_check
43#define arch_mmap_check(addr, len, flags) (0)
44#endif
45
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080046#ifndef arch_rebalance_pgtables
47#define arch_rebalance_pgtables(addr, len) (addr)
48#endif
49
Hugh Dickinse0da3822005-04-19 13:29:15 -070050static void unmap_region(struct mm_struct *mm,
51 struct vm_area_struct *vma, struct vm_area_struct *prev,
52 unsigned long start, unsigned long end);
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054/*
55 * WARNING: the debugging will use recursive algorithms so never enable this
56 * unless you know what you are doing.
57 */
58#undef DEBUG_MM_RB
59
60/* description of effects of mapping type and prot in current implementation.
61 * this is due to the limited x86 page protection hardware. The expected
62 * behavior is in parens:
63 *
64 * map_type prot
65 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
66 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
67 * w: (no) no w: (no) no w: (yes) yes w: (no) no
68 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
69 *
70 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (copy) copy w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
73 *
74 */
75pgprot_t protection_map[16] = {
76 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
77 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
78};
79
Hugh Dickins804af2c2006-07-26 21:39:49 +010080pgprot_t vm_get_page_prot(unsigned long vm_flags)
81{
Dave Kleikampb845f312008-07-08 00:28:51 +100082 return __pgprot(pgprot_val(protection_map[vm_flags &
83 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
84 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010085}
86EXPORT_SYMBOL(vm_get_page_prot);
87
Shaohua Li34679d72011-05-24 17:11:18 -070088int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
89int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070090int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Shaohua Li34679d72011-05-24 17:11:18 -070091/*
92 * Make sure vm_committed_as in one cacheline and not cacheline shared with
93 * other variables. It can be updated by several CPUs frequently.
94 */
95struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97/*
98 * Check that a process has enough memory to allocate a new virtual
99 * mapping. 0 means there is enough memory for the allocation to
100 * succeed and -ENOMEM implies there is not.
101 *
102 * We currently support three overcommit policies, which are set via the
103 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
104 *
105 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
106 * Additional code 2002 Jul 20 by Robert Love.
107 *
108 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
109 *
110 * Note this is a helper function intended to be used by LSMs which
111 * wish to use this logic.
112 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700113int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
115 unsigned long free, allowed;
116
117 vm_acct_memory(pages);
118
119 /*
120 * Sometimes we want to use more memory than we have
121 */
122 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
123 return 0;
124
125 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700126 free = global_page_state(NR_FREE_PAGES);
127 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Dmitry Finkc15bef32011-07-25 17:12:19 -0700129 /*
130 * shmem pages shouldn't be counted as free in this
131 * case, they can't be purged, only swapped out, and
132 * that won't affect the overall amount of available
133 * memory in the system.
134 */
135 free -= global_page_state(NR_SHMEM);
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 free += nr_swap_pages;
138
139 /*
140 * Any slabs which are created with the
141 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
142 * which are reclaimable, under pressure. The dentry
143 * cache and most inode caches should fall into this
144 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700145 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700148 * Leave reserved pages. The pages are not for anonymous pages.
149 */
150 if (free <= totalreserve_pages)
151 goto error;
152 else
153 free -= totalreserve_pages;
154
155 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 * Leave the last 3% for root
157 */
158 if (!cap_sys_admin)
159 free -= free / 32;
160
161 if (free > pages)
162 return 0;
163
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700164 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 }
166
167 allowed = (totalram_pages - hugetlb_total_pages())
168 * sysctl_overcommit_ratio / 100;
169 /*
170 * Leave the last 3% for root
171 */
172 if (!cap_sys_admin)
173 allowed -= allowed / 32;
174 allowed += total_swap_pages;
175
176 /* Don't let a single process grow too big:
177 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700178 if (mm)
179 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700181 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700183error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 vm_unacct_memory(pages);
185
186 return -ENOMEM;
187}
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700190 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 */
192static void __remove_shared_vm_struct(struct vm_area_struct *vma,
193 struct file *file, struct address_space *mapping)
194{
195 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800196 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 if (vma->vm_flags & VM_SHARED)
198 mapping->i_mmap_writable--;
199
200 flush_dcache_mmap_lock(mapping);
201 if (unlikely(vma->vm_flags & VM_NONLINEAR))
202 list_del_init(&vma->shared.vm_set.list);
203 else
204 vma_prio_tree_remove(vma, &mapping->i_mmap);
205 flush_dcache_mmap_unlock(mapping);
206}
207
208/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700209 * Unlink a file-based vm structure from its prio_tree, to hide
210 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700212void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 struct file *file = vma->vm_file;
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (file) {
217 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700218 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700220 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700222}
223
224/*
225 * Close a vm structure and free it, returning the next.
226 */
227static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
228{
229 struct vm_area_struct *next = vma->vm_next;
230
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700231 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 if (vma->vm_ops && vma->vm_ops->close)
233 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700234 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700235 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700236 if (vma->vm_flags & VM_EXECUTABLE)
237 removed_exe_file_vma(vma->vm_mm);
238 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700239 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700241 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700244static unsigned long do_brk(unsigned long addr, unsigned long len);
245
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100246SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
248 unsigned long rlim, retval;
249 unsigned long newbrk, oldbrk;
250 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700251 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
253 down_write(&mm->mmap_sem);
254
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700255#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800256 /*
257 * CONFIG_COMPAT_BRK can still be overridden by setting
258 * randomize_va_space to 2, which will still cause mm->start_brk
259 * to be arbitrarily shifted
260 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700261 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800262 min_brk = mm->start_brk;
263 else
264 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700265#else
266 min_brk = mm->start_brk;
267#endif
268 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700270
271 /*
272 * Check against rlimit here. If this check is done later after the test
273 * of oldbrk with newbrk then it can escape the test and let the data
274 * segment grow beyond its set limit the in case where the limit is
275 * not page aligned -Ram Gupta
276 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800277 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100278 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
279 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700280 goto out;
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 newbrk = PAGE_ALIGN(brk);
283 oldbrk = PAGE_ALIGN(mm->brk);
284 if (oldbrk == newbrk)
285 goto set_brk;
286
287 /* Always allow shrinking brk. */
288 if (brk <= mm->brk) {
289 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
290 goto set_brk;
291 goto out;
292 }
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 /* Check against existing mmap mappings. */
295 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
296 goto out;
297
298 /* Ok, looks good - let it rip. */
299 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
300 goto out;
301set_brk:
302 mm->brk = brk;
303out:
304 retval = mm->brk;
305 up_write(&mm->mmap_sem);
306 return retval;
307}
308
309#ifdef DEBUG_MM_RB
310static int browse_rb(struct rb_root *root)
311{
312 int i = 0, j;
313 struct rb_node *nd, *pn = NULL;
314 unsigned long prev = 0, pend = 0;
315
316 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
317 struct vm_area_struct *vma;
318 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
319 if (vma->vm_start < prev)
320 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
321 if (vma->vm_start < pend)
322 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
323 if (vma->vm_start > vma->vm_end)
324 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
325 i++;
326 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800327 prev = vma->vm_start;
328 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
330 j = 0;
331 for (nd = pn; nd; nd = rb_prev(nd)) {
332 j++;
333 }
334 if (i != j)
335 printk("backwards %d, forwards %d\n", j, i), i = 0;
336 return i;
337}
338
339void validate_mm(struct mm_struct *mm)
340{
341 int bug = 0;
342 int i = 0;
343 struct vm_area_struct *tmp = mm->mmap;
344 while (tmp) {
345 tmp = tmp->vm_next;
346 i++;
347 }
348 if (i != mm->map_count)
349 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
350 i = browse_rb(&mm->mm_rb);
351 if (i != mm->map_count)
352 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200353 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355#else
356#define validate_mm(mm) do { } while (0)
357#endif
358
359static struct vm_area_struct *
360find_vma_prepare(struct mm_struct *mm, unsigned long addr,
361 struct vm_area_struct **pprev, struct rb_node ***rb_link,
362 struct rb_node ** rb_parent)
363{
364 struct vm_area_struct * vma;
365 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
366
367 __rb_link = &mm->mm_rb.rb_node;
368 rb_prev = __rb_parent = NULL;
369 vma = NULL;
370
371 while (*__rb_link) {
372 struct vm_area_struct *vma_tmp;
373
374 __rb_parent = *__rb_link;
375 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
376
377 if (vma_tmp->vm_end > addr) {
378 vma = vma_tmp;
379 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700380 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 __rb_link = &__rb_parent->rb_left;
382 } else {
383 rb_prev = __rb_parent;
384 __rb_link = &__rb_parent->rb_right;
385 }
386 }
387
388 *pprev = NULL;
389 if (rb_prev)
390 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
391 *rb_link = __rb_link;
392 *rb_parent = __rb_parent;
393 return vma;
394}
395
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
397 struct rb_node **rb_link, struct rb_node *rb_parent)
398{
399 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
400 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
401}
402
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700403static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
ZhenwenXu48aae422009-01-06 14:40:21 -0800405 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 file = vma->vm_file;
408 if (file) {
409 struct address_space *mapping = file->f_mapping;
410
411 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800412 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (vma->vm_flags & VM_SHARED)
414 mapping->i_mmap_writable++;
415
416 flush_dcache_mmap_lock(mapping);
417 if (unlikely(vma->vm_flags & VM_NONLINEAR))
418 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
419 else
420 vma_prio_tree_insert(vma, &mapping->i_mmap);
421 flush_dcache_mmap_unlock(mapping);
422 }
423}
424
425static void
426__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
427 struct vm_area_struct *prev, struct rb_node **rb_link,
428 struct rb_node *rb_parent)
429{
430 __vma_link_list(mm, vma, prev, rb_parent);
431 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433
434static void 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 struct address_space *mapping = NULL;
439
440 if (vma->vm_file)
441 mapping = vma->vm_file->f_mapping;
442
Peter Zijlstra97a89412011-05-24 17:12:04 -0700443 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700444 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 __vma_link(mm, vma, prev, rb_link, rb_parent);
447 __vma_link_file(vma);
448
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700450 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 mm->map_count++;
453 validate_mm(mm);
454}
455
456/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700457 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
458 * mm's list and rbtree. It has already been inserted into the prio_tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800460static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
ZhenwenXu48aae422009-01-06 14:40:21 -0800462 struct vm_area_struct *__vma, *prev;
463 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200466 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 __vma_link(mm, vma, prev, rb_link, rb_parent);
468 mm->map_count++;
469}
470
471static inline void
472__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
473 struct vm_area_struct *prev)
474{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700475 struct vm_area_struct *next = vma->vm_next;
476
477 prev->vm_next = next;
478 if (next)
479 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 rb_erase(&vma->vm_rb, &mm->mm_rb);
481 if (mm->mmap_cache == vma)
482 mm->mmap_cache = prev;
483}
484
485/*
486 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
487 * is already present in an i_mmap tree without adjusting the tree.
488 * The following helper function should be used when such adjustments
489 * are necessary. The "insert" vma (if any) is to be inserted
490 * before we drop the necessary locks.
491 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800492int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
494{
495 struct mm_struct *mm = vma->vm_mm;
496 struct vm_area_struct *next = vma->vm_next;
497 struct vm_area_struct *importer = NULL;
498 struct address_space *mapping = NULL;
499 struct prio_tree_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700500 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 long adjust_next = 0;
503 int remove_next = 0;
504
505 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700506 struct vm_area_struct *exporter = NULL;
507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (end >= next->vm_end) {
509 /*
510 * vma expands, overlapping all the next, and
511 * perhaps the one after too (mprotect case 6).
512 */
513again: remove_next = 1 + (end > next->vm_end);
514 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700515 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 importer = vma;
517 } else if (end > next->vm_start) {
518 /*
519 * vma expands, overlapping part of the next:
520 * mprotect case 5 shifting the boundary up.
521 */
522 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700523 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 importer = vma;
525 } else if (end < vma->vm_end) {
526 /*
527 * vma shrinks, and !insert tells it's not
528 * split_vma inserting another: so it must be
529 * mprotect case 4 shifting the boundary down.
530 */
531 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700532 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 importer = next;
534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
Rik van Riel5beb4932010-03-05 13:42:07 -0800536 /*
537 * Easily overlooked: when mprotect shifts the boundary,
538 * make sure the expanding vma has anon_vma set if the
539 * shrinking vma had, to cover any anon pages imported.
540 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700541 if (exporter && exporter->anon_vma && !importer->anon_vma) {
542 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800543 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700544 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800545 }
546 }
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 if (file) {
549 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530550 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530552 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530553
554 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530555 uprobe_munmap(next, next->vm_start,
556 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530557 }
558
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700559 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 /*
562 * Put into prio_tree now, so instantiated pages
563 * are visible to arm/parisc __flush_dcache_page
564 * throughout; but we cannot insert into address
565 * space until vma start or end is updated.
566 */
567 __vma_link_file(insert);
568 }
569 }
570
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800571 vma_adjust_trans_huge(vma, start, end, adjust_next);
572
Rik van Riel012f18002010-08-09 17:18:40 -0700573 /*
574 * When changing only vma->vm_end, we don't really need anon_vma
575 * lock. This is a fairly rare case by itself, but the anon_vma
576 * lock may be shared between many sibling processes. Skipping
577 * the lock for brk adjustments makes a difference sometimes.
578 */
Shaohua Li5f70b962011-05-24 17:11:19 -0700579 if (vma->anon_vma && (importer || start != vma->vm_start)) {
Rik van Riel012f18002010-08-09 17:18:40 -0700580 anon_vma = vma->anon_vma;
581 anon_vma_lock(anon_vma);
582 }
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (root) {
585 flush_dcache_mmap_lock(mapping);
586 vma_prio_tree_remove(vma, root);
587 if (adjust_next)
588 vma_prio_tree_remove(next, root);
589 }
590
591 vma->vm_start = start;
592 vma->vm_end = end;
593 vma->vm_pgoff = pgoff;
594 if (adjust_next) {
595 next->vm_start += adjust_next << PAGE_SHIFT;
596 next->vm_pgoff += adjust_next;
597 }
598
599 if (root) {
600 if (adjust_next)
601 vma_prio_tree_insert(next, root);
602 vma_prio_tree_insert(vma, root);
603 flush_dcache_mmap_unlock(mapping);
604 }
605
606 if (remove_next) {
607 /*
608 * vma_merge has merged next into vma, and needs
609 * us to remove next before dropping the locks.
610 */
611 __vma_unlink(mm, next, vma);
612 if (file)
613 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 } else if (insert) {
615 /*
616 * split_vma has split insert from vma, and needs
617 * us to insert it before dropping the locks
618 * (it may either follow vma or precede it).
619 */
620 __insert_vm_struct(mm, insert);
621 }
622
Rik van Riel012f18002010-08-09 17:18:40 -0700623 if (anon_vma)
624 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700626 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530628 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100629 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530630
631 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100632 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530633 }
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700636 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530637 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700639 if (next->vm_flags & VM_EXECUTABLE)
640 removed_exe_file_vma(mm);
641 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800642 if (next->anon_vma)
643 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700645 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 kmem_cache_free(vm_area_cachep, next);
647 /*
648 * In mprotect's case 6 (see comments on vma_merge),
649 * we must remove another next too. It would clutter
650 * up the code too much to do both in one go.
651 */
652 if (remove_next == 2) {
653 next = vma->vm_next;
654 goto again;
655 }
656 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530657 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100658 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800661
662 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665/*
666 * If the vma has a ->close operation then the driver probably needs to release
667 * per-vma resources, so we don't attempt to merge those.
668 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669static inline int is_mergeable_vma(struct vm_area_struct *vma,
670 struct file *file, unsigned long vm_flags)
671{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700672 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
673 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 return 0;
675 if (vma->vm_file != file)
676 return 0;
677 if (vma->vm_ops && vma->vm_ops->close)
678 return 0;
679 return 1;
680}
681
682static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700683 struct anon_vma *anon_vma2,
684 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
Shaohua Li965f55d2011-05-24 17:11:20 -0700686 /*
687 * The list_is_singular() test is to avoid merging VMA cloned from
688 * parents. This can improve scalability caused by anon_vma lock.
689 */
690 if ((!anon_vma1 || !anon_vma2) && (!vma ||
691 list_is_singular(&vma->anon_vma_chain)))
692 return 1;
693 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
696/*
697 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
698 * in front of (at a lower virtual address and file offset than) the vma.
699 *
700 * We cannot merge two vmas if they have differently assigned (non-NULL)
701 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
702 *
703 * We don't check here for the merged mmap wrapping around the end of pagecache
704 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
705 * wrap, nor mmaps which cover the final page at index -1UL.
706 */
707static int
708can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
709 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
710{
711 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700712 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (vma->vm_pgoff == vm_pgoff)
714 return 1;
715 }
716 return 0;
717}
718
719/*
720 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
721 * beyond (at a higher virtual address and file offset than) the vma.
722 *
723 * We cannot merge two vmas if they have differently assigned (non-NULL)
724 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
725 */
726static int
727can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
728 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
729{
730 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700731 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 pgoff_t vm_pglen;
733 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
734 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
735 return 1;
736 }
737 return 0;
738}
739
740/*
741 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
742 * whether that can be merged with its predecessor or its successor.
743 * Or both (it neatly fills a hole).
744 *
745 * In most cases - when called for mmap, brk or mremap - [addr,end) is
746 * certain not to be mapped by the time vma_merge is called; but when
747 * called for mprotect, it is certain to be already mapped (either at
748 * an offset within prev, or at the start of next), and the flags of
749 * this area are about to be changed to vm_flags - and the no-change
750 * case has already been eliminated.
751 *
752 * The following mprotect cases have to be considered, where AAAA is
753 * the area passed down from mprotect_fixup, never extending beyond one
754 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
755 *
756 * AAAA AAAA AAAA AAAA
757 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
758 * cannot merge might become might become might become
759 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
760 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
761 * mremap move: PPPPNNNNNNNN 8
762 * AAAA
763 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
764 * might become case 1 below case 2 below case 3 below
765 *
766 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
767 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
768 */
769struct vm_area_struct *vma_merge(struct mm_struct *mm,
770 struct vm_area_struct *prev, unsigned long addr,
771 unsigned long end, unsigned long vm_flags,
772 struct anon_vma *anon_vma, struct file *file,
773 pgoff_t pgoff, struct mempolicy *policy)
774{
775 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
776 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800777 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779 /*
780 * We later require that vma->vm_flags == vm_flags,
781 * so this tests vma->vm_flags & VM_SPECIAL, too.
782 */
783 if (vm_flags & VM_SPECIAL)
784 return NULL;
785
786 if (prev)
787 next = prev->vm_next;
788 else
789 next = mm->mmap;
790 area = next;
791 if (next && next->vm_end == end) /* cases 6, 7, 8 */
792 next = next->vm_next;
793
794 /*
795 * Can it merge with the predecessor?
796 */
797 if (prev && prev->vm_end == addr &&
798 mpol_equal(vma_policy(prev), policy) &&
799 can_vma_merge_after(prev, vm_flags,
800 anon_vma, file, pgoff)) {
801 /*
802 * OK, it can. Can we now merge in the successor as well?
803 */
804 if (next && end == next->vm_start &&
805 mpol_equal(policy, vma_policy(next)) &&
806 can_vma_merge_before(next, vm_flags,
807 anon_vma, file, pgoff+pglen) &&
808 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700809 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800811 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 next->vm_end, prev->vm_pgoff, NULL);
813 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800814 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800816 if (err)
817 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800818 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return prev;
820 }
821
822 /*
823 * Can this new request be merged in front of next?
824 */
825 if (next && end == next->vm_start &&
826 mpol_equal(policy, vma_policy(next)) &&
827 can_vma_merge_before(next, vm_flags,
828 anon_vma, file, pgoff+pglen)) {
829 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800830 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 addr, prev->vm_pgoff, NULL);
832 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800833 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800835 if (err)
836 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800837 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 return area;
839 }
840
841 return NULL;
842}
843
844/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700845 * Rough compatbility check to quickly see if it's even worth looking
846 * at sharing an anon_vma.
847 *
848 * They need to have the same vm_file, and the flags can only differ
849 * in things that mprotect may change.
850 *
851 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
852 * we can merge the two vma's. For example, we refuse to merge a vma if
853 * there is a vm_ops->close() function, because that indicates that the
854 * driver is doing some kind of reference counting. But that doesn't
855 * really matter for the anon_vma sharing case.
856 */
857static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
858{
859 return a->vm_end == b->vm_start &&
860 mpol_equal(vma_policy(a), vma_policy(b)) &&
861 a->vm_file == b->vm_file &&
862 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
863 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
864}
865
866/*
867 * Do some basic sanity checking to see if we can re-use the anon_vma
868 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
869 * the same as 'old', the other will be the new one that is trying
870 * to share the anon_vma.
871 *
872 * NOTE! This runs with mm_sem held for reading, so it is possible that
873 * the anon_vma of 'old' is concurrently in the process of being set up
874 * by another page fault trying to merge _that_. But that's ok: if it
875 * is being set up, that automatically means that it will be a singleton
876 * acceptable for merging, so we can do all of this optimistically. But
877 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
878 *
879 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
880 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
881 * is to return an anon_vma that is "complex" due to having gone through
882 * a fork).
883 *
884 * We also make sure that the two vma's are compatible (adjacent,
885 * and with the same memory policies). That's all stable, even with just
886 * a read lock on the mm_sem.
887 */
888static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
889{
890 if (anon_vma_compatible(a, b)) {
891 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
892
893 if (anon_vma && list_is_singular(&old->anon_vma_chain))
894 return anon_vma;
895 }
896 return NULL;
897}
898
899/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
901 * neighbouring vmas for a suitable anon_vma, before it goes off
902 * to allocate a new anon_vma. It checks because a repetitive
903 * sequence of mprotects and faults may otherwise lead to distinct
904 * anon_vmas being allocated, preventing vma merge in subsequent
905 * mprotect.
906 */
907struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
908{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700909 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 near = vma->vm_next;
913 if (!near)
914 goto try_prev;
915
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700916 anon_vma = reusable_anon_vma(near, vma, near);
917 if (anon_vma)
918 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -0700920 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (!near)
922 goto none;
923
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700924 anon_vma = reusable_anon_vma(near, near, vma);
925 if (anon_vma)
926 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927none:
928 /*
929 * There's no absolute need to look only at touching neighbours:
930 * we could search further afield for "compatible" anon_vmas.
931 * But it would probably just be a waste of time searching,
932 * or lead to too many vmas hanging off the same anon_vma.
933 * We're trying to allow mprotect remerging later on,
934 * not trying to minimize memory used for anon_vmas.
935 */
936 return NULL;
937}
938
939#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700940void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 struct file *file, long pages)
942{
943 const unsigned long stack_flags
944 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 if (file) {
947 mm->shared_vm += pages;
948 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
949 mm->exec_vm += pages;
950 } else if (flags & stack_flags)
951 mm->stack_vm += pages;
952 if (flags & (VM_RESERVED|VM_IO))
953 mm->reserved_vm += pages;
954}
955#endif /* CONFIG_PROC_FS */
956
957/*
Al Viro40401532012-02-13 03:58:52 +0000958 * If a hint addr is less than mmap_min_addr change hint to be as
959 * low as possible but still greater than mmap_min_addr
960 */
961static inline unsigned long round_hint_to_min(unsigned long hint)
962{
963 hint &= PAGE_MASK;
964 if (((void *)hint != NULL) &&
965 (hint < mmap_min_addr))
966 return PAGE_ALIGN(mmap_min_addr);
967 return hint;
968}
969
970/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700971 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 */
973
Linus Torvalds6be5ceb2012-04-20 17:13:58 -0700974static unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 unsigned long len, unsigned long prot,
976 unsigned long flags, unsigned long pgoff)
977{
978 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 struct inode *inode;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900980 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700982 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /*
985 * Does the application expect PROT_READ to imply PROT_EXEC?
986 *
987 * (the exception is when the underlying filesystem is noexec
988 * mounted, in which case we dont add PROT_EXEC.)
989 */
990 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800991 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 prot |= PROT_EXEC;
993
994 if (!len)
995 return -EINVAL;
996
Eric Paris7cd94142007-11-26 18:47:40 -0500997 if (!(flags & MAP_FIXED))
998 addr = round_hint_to_min(addr);
999
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 /* Careful about overflows.. */
1001 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001002 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 return -ENOMEM;
1004
1005 /* offset overflow? */
1006 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1007 return -EOVERFLOW;
1008
1009 /* Too many mappings? */
1010 if (mm->map_count > sysctl_max_map_count)
1011 return -ENOMEM;
1012
1013 /* Obtain the address to map to. we verify (or select) it and ensure
1014 * that it represents a valid section of the address space.
1015 */
1016 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1017 if (addr & ~PAGE_MASK)
1018 return addr;
1019
1020 /* Do simple checking here so the lower-level routines won't have
1021 * to. we assume access permissions have been handled by the open
1022 * of the memory object, so we don't do any here.
1023 */
1024 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1025 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1026
Huang Shijiecdf7b342009-09-21 17:03:36 -07001027 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 if (!can_do_mlock())
1029 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 /* mlock MCL_FUTURE? */
1032 if (vm_flags & VM_LOCKED) {
1033 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001034 locked = len >> PAGE_SHIFT;
1035 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001036 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001037 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1039 return -EAGAIN;
1040 }
1041
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001042 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 if (file) {
1045 switch (flags & MAP_TYPE) {
1046 case MAP_SHARED:
1047 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1048 return -EACCES;
1049
1050 /*
1051 * Make sure we don't allow writing to an append-only
1052 * file..
1053 */
1054 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1055 return -EACCES;
1056
1057 /*
1058 * Make sure there are no mandatory locks on the file.
1059 */
1060 if (locks_verify_locked(inode))
1061 return -EAGAIN;
1062
1063 vm_flags |= VM_SHARED | VM_MAYSHARE;
1064 if (!(file->f_mode & FMODE_WRITE))
1065 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1066
1067 /* fall through */
1068 case MAP_PRIVATE:
1069 if (!(file->f_mode & FMODE_READ))
1070 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001071 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001072 if (vm_flags & VM_EXEC)
1073 return -EPERM;
1074 vm_flags &= ~VM_MAYEXEC;
1075 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001076
1077 if (!file->f_op || !file->f_op->mmap)
1078 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 break;
1080
1081 default:
1082 return -EINVAL;
1083 }
1084 } else {
1085 switch (flags & MAP_TYPE) {
1086 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001087 /*
1088 * Ignore pgoff.
1089 */
1090 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 vm_flags |= VM_SHARED | VM_MAYSHARE;
1092 break;
1093 case MAP_PRIVATE:
1094 /*
1095 * Set pgoff according to addr for anon_vma.
1096 */
1097 pgoff = addr >> PAGE_SHIFT;
1098 break;
1099 default:
1100 return -EINVAL;
1101 }
1102 }
1103
Eric Parised032182007-06-28 15:55:21 -04001104 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 if (error)
1106 return error;
Eric Parised032182007-06-28 15:55:21 -04001107
Mel Gorman5a6fe122009-02-10 14:02:27 +00001108 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001109}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001110
1111unsigned long do_mmap(struct file *file, unsigned long addr,
1112 unsigned long len, unsigned long prot,
1113 unsigned long flag, unsigned long offset)
1114{
1115 if (unlikely(offset + PAGE_ALIGN(len) < offset))
1116 return -EINVAL;
1117 if (unlikely(offset & ~PAGE_MASK))
1118 return -EINVAL;
1119 return do_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
1120}
1121EXPORT_SYMBOL(do_mmap);
1122
1123unsigned long vm_mmap(struct file *file, unsigned long addr,
1124 unsigned long len, unsigned long prot,
1125 unsigned long flag, unsigned long offset)
1126{
1127 unsigned long ret;
1128 struct mm_struct *mm = current->mm;
1129
1130 down_write(&mm->mmap_sem);
1131 ret = do_mmap(file, addr, len, prot, flag, offset);
1132 up_write(&mm->mmap_sem);
1133 return ret;
1134}
1135EXPORT_SYMBOL(vm_mmap);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001136
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001137SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1138 unsigned long, prot, unsigned long, flags,
1139 unsigned long, fd, unsigned long, pgoff)
1140{
1141 struct file *file = NULL;
1142 unsigned long retval = -EBADF;
1143
1144 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001145 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001146 if (unlikely(flags & MAP_HUGETLB))
1147 return -EINVAL;
1148 file = fget(fd);
1149 if (!file)
1150 goto out;
1151 } else if (flags & MAP_HUGETLB) {
1152 struct user_struct *user = NULL;
1153 /*
1154 * VM_NORESERVE is used because the reservations will be
1155 * taken when vm_ops->mmap() is called
1156 * A dummy user value is used because we are not locking
1157 * memory so no accounting is necessary
1158 */
Steven Truelove40716e22012-03-21 16:34:14 -07001159 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1160 VM_NORESERVE, &user,
1161 HUGETLB_ANONHUGE_INODE);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001162 if (IS_ERR(file))
1163 return PTR_ERR(file);
1164 }
1165
1166 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1167
1168 down_write(&current->mm->mmap_sem);
1169 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1170 up_write(&current->mm->mmap_sem);
1171
1172 if (file)
1173 fput(file);
1174out:
1175 return retval;
1176}
1177
Christoph Hellwiga4679372010-03-10 15:21:15 -08001178#ifdef __ARCH_WANT_SYS_OLD_MMAP
1179struct mmap_arg_struct {
1180 unsigned long addr;
1181 unsigned long len;
1182 unsigned long prot;
1183 unsigned long flags;
1184 unsigned long fd;
1185 unsigned long offset;
1186};
1187
1188SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1189{
1190 struct mmap_arg_struct a;
1191
1192 if (copy_from_user(&a, arg, sizeof(a)))
1193 return -EFAULT;
1194 if (a.offset & ~PAGE_MASK)
1195 return -EINVAL;
1196
1197 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1198 a.offset >> PAGE_SHIFT);
1199}
1200#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1201
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001202/*
1203 * Some shared mappigns will want the pages marked read-only
1204 * to track write events. If so, we'll downgrade vm_page_prot
1205 * to the private version (using protection_map[] without the
1206 * VM_SHARED bit).
1207 */
1208int vma_wants_writenotify(struct vm_area_struct *vma)
1209{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001210 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001211
1212 /* If it was private or non-writable, the write bit is already clear */
1213 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1214 return 0;
1215
1216 /* The backer wishes to know when pages are first written to? */
1217 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1218 return 1;
1219
1220 /* The open routine did something to the protections already? */
1221 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001222 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001223 return 0;
1224
1225 /* Specialty mapping? */
1226 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1227 return 0;
1228
1229 /* Can the mapping track the dirty pages? */
1230 return vma->vm_file && vma->vm_file->f_mapping &&
1231 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1232}
1233
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001234/*
1235 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001236 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001237 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001238static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001239{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001240 /*
1241 * hugetlb has its own accounting separate from the core VM
1242 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1243 */
1244 if (file && is_file_hugepages(file))
1245 return 0;
1246
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001247 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1248}
1249
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001250unsigned long mmap_region(struct file *file, unsigned long addr,
1251 unsigned long len, unsigned long flags,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001252 vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001253{
1254 struct mm_struct *mm = current->mm;
1255 struct vm_area_struct *vma, *prev;
1256 int correct_wcount = 0;
1257 int error;
1258 struct rb_node **rb_link, *rb_parent;
1259 unsigned long charged = 0;
1260 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1261
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 /* Clear old maps */
1263 error = -ENOMEM;
1264munmap_back:
1265 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1266 if (vma && vma->vm_start < addr + len) {
1267 if (do_munmap(mm, addr, len))
1268 return -ENOMEM;
1269 goto munmap_back;
1270 }
1271
1272 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001273 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 return -ENOMEM;
1275
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001276 /*
1277 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001278 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001279 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001280 if ((flags & MAP_NORESERVE)) {
1281 /* We honor MAP_NORESERVE if allowed to overcommit */
1282 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1283 vm_flags |= VM_NORESERVE;
1284
1285 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1286 if (file && is_file_hugepages(file))
1287 vm_flags |= VM_NORESERVE;
1288 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001289
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001290 /*
1291 * Private writable mapping: check memory availability
1292 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001293 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001294 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001295 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001296 return -ENOMEM;
1297 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 }
1299
1300 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001301 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001303 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1304 if (vma)
1305 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 /*
1308 * Determine the object being mapped and call the appropriate
1309 * specific mapper. the address has already been validated, but
1310 * not unmapped, but the maps are removed from the list.
1311 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001312 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 if (!vma) {
1314 error = -ENOMEM;
1315 goto unacct_error;
1316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 vma->vm_mm = mm;
1319 vma->vm_start = addr;
1320 vma->vm_end = addr + len;
1321 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001322 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001324 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Hugh Dickinsce8fea72012-03-06 12:28:52 -08001326 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1330 goto free_vma;
1331 if (vm_flags & VM_DENYWRITE) {
1332 error = deny_write_access(file);
1333 if (error)
1334 goto free_vma;
1335 correct_wcount = 1;
1336 }
1337 vma->vm_file = file;
1338 get_file(file);
1339 error = file->f_op->mmap(file, vma);
1340 if (error)
1341 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001342 if (vm_flags & VM_EXECUTABLE)
1343 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001344
1345 /* Can addr have changed??
1346 *
1347 * Answer: Yes, several device drivers can do it in their
1348 * f_op->mmap method. -DaveM
1349 */
1350 addr = vma->vm_start;
1351 pgoff = vma->vm_pgoff;
1352 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 } else if (vm_flags & VM_SHARED) {
Al Viro835ee792012-03-05 06:39:47 +00001354 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1355 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 error = shmem_zero_setup(vma);
1357 if (error)
1358 goto free_vma;
1359 }
1360
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001361 if (vma_wants_writenotify(vma)) {
1362 pgprot_t pprot = vma->vm_page_prot;
1363
1364 /* Can vma->vm_page_prot have changed??
1365 *
1366 * Answer: Yes, drivers may have changed it in their
1367 * f_op->mmap method.
1368 *
1369 * Ensures that vmas marked as uncached stay that way.
1370 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001371 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001372 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1373 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1374 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001375
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001376 vma_link(mm, vma, prev, rb_link, rb_parent);
1377 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001378
1379 /* Once vma denies write, undo our temporary denial count */
1380 if (correct_wcount)
1381 atomic_inc(&inode->i_writecount);
1382out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001383 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001386 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001388 if (!mlock_vma_pages_range(vma, addr, addr + len))
1389 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001390 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001391 make_pages_present(addr, addr + len);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301392
Ingo Molnar7b2d81d2012-02-17 09:27:41 +01001393 if (file && uprobe_mmap(vma))
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301394 /* matching probes but cannot insert */
1395 goto unmap_and_free_vma;
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 return addr;
1398
1399unmap_and_free_vma:
1400 if (correct_wcount)
1401 atomic_inc(&inode->i_writecount);
1402 vma->vm_file = NULL;
1403 fput(file);
1404
1405 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001406 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1407 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408free_vma:
1409 kmem_cache_free(vm_area_cachep, vma);
1410unacct_error:
1411 if (charged)
1412 vm_unacct_memory(charged);
1413 return error;
1414}
1415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416/* Get an address range which is currently unmapped.
1417 * For shmat() with addr=0.
1418 *
1419 * Ugly calling convention alert:
1420 * Return value with the low bits set means error value,
1421 * ie
1422 * if (ret & ~PAGE_MASK)
1423 * error = ret;
1424 *
1425 * This function "knows" that -ENOMEM has the bits set.
1426 */
1427#ifndef HAVE_ARCH_UNMAPPED_AREA
1428unsigned long
1429arch_get_unmapped_area(struct file *filp, unsigned long addr,
1430 unsigned long len, unsigned long pgoff, unsigned long flags)
1431{
1432 struct mm_struct *mm = current->mm;
1433 struct vm_area_struct *vma;
1434 unsigned long start_addr;
1435
1436 if (len > TASK_SIZE)
1437 return -ENOMEM;
1438
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001439 if (flags & MAP_FIXED)
1440 return addr;
1441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 if (addr) {
1443 addr = PAGE_ALIGN(addr);
1444 vma = find_vma(mm, addr);
1445 if (TASK_SIZE - len >= addr &&
1446 (!vma || addr + len <= vma->vm_start))
1447 return addr;
1448 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001449 if (len > mm->cached_hole_size) {
1450 start_addr = addr = mm->free_area_cache;
1451 } else {
1452 start_addr = addr = TASK_UNMAPPED_BASE;
1453 mm->cached_hole_size = 0;
1454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456full_search:
1457 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1458 /* At this point: (!vma || addr < vma->vm_end). */
1459 if (TASK_SIZE - len < addr) {
1460 /*
1461 * Start a new search - just in case we missed
1462 * some holes.
1463 */
1464 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001465 addr = TASK_UNMAPPED_BASE;
1466 start_addr = addr;
1467 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 goto full_search;
1469 }
1470 return -ENOMEM;
1471 }
1472 if (!vma || addr + len <= vma->vm_start) {
1473 /*
1474 * Remember the place where we stopped the search:
1475 */
1476 mm->free_area_cache = addr + len;
1477 return addr;
1478 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001479 if (addr + mm->cached_hole_size < vma->vm_start)
1480 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 addr = vma->vm_end;
1482 }
1483}
1484#endif
1485
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001486void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
1488 /*
1489 * Is this a new hole at the lowest possible address?
1490 */
Xiao Guangrongf44d2192012-03-21 16:33:56 -07001491 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001492 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
1494
1495/*
1496 * This mmap-allocator allocates new areas top-down from below the
1497 * stack's low limit (the base):
1498 */
1499#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1500unsigned long
1501arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1502 const unsigned long len, const unsigned long pgoff,
1503 const unsigned long flags)
1504{
1505 struct vm_area_struct *vma;
1506 struct mm_struct *mm = current->mm;
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001507 unsigned long addr = addr0, start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
1509 /* requested length too big for entire address space */
1510 if (len > TASK_SIZE)
1511 return -ENOMEM;
1512
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001513 if (flags & MAP_FIXED)
1514 return addr;
1515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 /* requesting a specific address */
1517 if (addr) {
1518 addr = PAGE_ALIGN(addr);
1519 vma = find_vma(mm, addr);
1520 if (TASK_SIZE - len >= addr &&
1521 (!vma || addr + len <= vma->vm_start))
1522 return addr;
1523 }
1524
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001525 /* check if free_area_cache is useful for us */
1526 if (len <= mm->cached_hole_size) {
1527 mm->cached_hole_size = 0;
1528 mm->free_area_cache = mm->mmap_base;
1529 }
1530
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001531try_again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 /* either no address requested or can't fit in requested address hole */
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001533 start_addr = addr = mm->free_area_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001535 if (addr < len)
1536 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001538 addr -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 do {
1540 /*
1541 * Lookup failure means no vma is above this address,
1542 * else if new region fits below vma->vm_start,
1543 * return with success:
1544 */
1545 vma = find_vma(mm, addr);
1546 if (!vma || addr+len <= vma->vm_start)
1547 /* remember the address as a hint for next time */
1548 return (mm->free_area_cache = addr);
1549
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001550 /* remember the largest hole we saw so far */
1551 if (addr + mm->cached_hole_size < vma->vm_start)
1552 mm->cached_hole_size = vma->vm_start - addr;
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 /* try just below the current vma->vm_start */
1555 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001556 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001558fail:
1559 /*
1560 * if hint left us with no space for the requested
1561 * mapping then try again:
1562 *
1563 * Note: this is different with the case of bottomup
1564 * which does the fully line-search, but we use find_vma
1565 * here that causes some holes skipped.
1566 */
1567 if (start_addr != mm->mmap_base) {
1568 mm->free_area_cache = mm->mmap_base;
1569 mm->cached_hole_size = 0;
1570 goto try_again;
1571 }
1572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 /*
1574 * A failed mmap() very likely causes application failure,
1575 * so fall back to the bottom-up function here. This scenario
1576 * can happen with large stack limits and large mmap()
1577 * allocations.
1578 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001579 mm->cached_hole_size = ~0UL;
1580 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1582 /*
1583 * Restore the topdown base:
1584 */
1585 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001586 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
1588 return addr;
1589}
1590#endif
1591
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001592void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593{
1594 /*
1595 * Is this a new hole at the highest possible address?
1596 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001597 if (addr > mm->free_area_cache)
1598 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001601 if (mm->free_area_cache > mm->mmap_base)
1602 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603}
1604
1605unsigned long
1606get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1607 unsigned long pgoff, unsigned long flags)
1608{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001609 unsigned long (*get_area)(struct file *, unsigned long,
1610 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Al Viro9206de92009-12-03 15:23:11 -05001612 unsigned long error = arch_mmap_check(addr, len, flags);
1613 if (error)
1614 return error;
1615
1616 /* Careful about overflows.. */
1617 if (len > TASK_SIZE)
1618 return -ENOMEM;
1619
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001620 get_area = current->mm->get_unmapped_area;
1621 if (file && file->f_op && file->f_op->get_unmapped_area)
1622 get_area = file->f_op->get_unmapped_area;
1623 addr = get_area(file, addr, len, pgoff, flags);
1624 if (IS_ERR_VALUE(addr))
1625 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001627 if (addr > TASK_SIZE - len)
1628 return -ENOMEM;
1629 if (addr & ~PAGE_MASK)
1630 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001631
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001632 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633}
1634
1635EXPORT_SYMBOL(get_unmapped_area);
1636
1637/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001638struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639{
1640 struct vm_area_struct *vma = NULL;
1641
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001642 if (WARN_ON_ONCE(!mm)) /* Remove this in linux-3.6 */
1643 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001645 /* Check the cache first. */
1646 /* (Cache hit rate is typically around 35%.) */
1647 vma = mm->mmap_cache;
1648 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1649 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001651 rb_node = mm->mm_rb.rb_node;
1652 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001654 while (rb_node) {
1655 struct vm_area_struct *vma_tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001657 vma_tmp = rb_entry(rb_node,
1658 struct vm_area_struct, vm_rb);
1659
1660 if (vma_tmp->vm_end > addr) {
1661 vma = vma_tmp;
1662 if (vma_tmp->vm_start <= addr)
1663 break;
1664 rb_node = rb_node->rb_left;
1665 } else
1666 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 }
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001668 if (vma)
1669 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
1671 return vma;
1672}
1673
1674EXPORT_SYMBOL(find_vma);
1675
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001676/*
1677 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001678 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679struct vm_area_struct *
1680find_vma_prev(struct mm_struct *mm, unsigned long addr,
1681 struct vm_area_struct **pprev)
1682{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001683 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001685 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05001686 if (vma) {
1687 *pprev = vma->vm_prev;
1688 } else {
1689 struct rb_node *rb_node = mm->mm_rb.rb_node;
1690 *pprev = NULL;
1691 while (rb_node) {
1692 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1693 rb_node = rb_node->rb_right;
1694 }
1695 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001696 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697}
1698
1699/*
1700 * Verify that the stack growth is acceptable and
1701 * update accounting. This is shared with both the
1702 * grow-up and grow-down cases.
1703 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001704static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
1706 struct mm_struct *mm = vma->vm_mm;
1707 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001708 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
1710 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001711 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 return -ENOMEM;
1713
1714 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001715 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 return -ENOMEM;
1717
1718 /* mlock limit tests */
1719 if (vma->vm_flags & VM_LOCKED) {
1720 unsigned long locked;
1721 unsigned long limit;
1722 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001723 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1724 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 if (locked > limit && !capable(CAP_IPC_LOCK))
1726 return -ENOMEM;
1727 }
1728
Adam Litke0d59a012007-01-30 14:35:39 -08001729 /* Check to ensure the stack will not grow into a hugetlb-only region */
1730 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1731 vma->vm_end - size;
1732 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1733 return -EFAULT;
1734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 /*
1736 * Overcommit.. This must be the final test, as it will
1737 * update security statistics.
1738 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001739 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 return -ENOMEM;
1741
1742 /* Ok, everything looks good - let it rip */
1743 mm->total_vm += grow;
1744 if (vma->vm_flags & VM_LOCKED)
1745 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001746 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 return 0;
1748}
1749
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001750#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001752 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1753 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001755int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756{
1757 int error;
1758
1759 if (!(vma->vm_flags & VM_GROWSUP))
1760 return -EFAULT;
1761
1762 /*
1763 * We must make sure the anon_vma is allocated
1764 * so that the anon_vma locking is not a noop.
1765 */
1766 if (unlikely(anon_vma_prepare(vma)))
1767 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001768 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 /*
1771 * vma->vm_start/vm_end cannot change under us because the caller
1772 * is required to hold the mmap_sem in read mode. We need the
1773 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001774 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 */
Helge Deller06b32f32006-12-19 19:28:33 +01001776 if (address < PAGE_ALIGN(address+4))
1777 address = PAGE_ALIGN(address+4);
1778 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001779 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001780 return -ENOMEM;
1781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 error = 0;
1783
1784 /* Somebody else might have raced and expanded it already */
1785 if (address > vma->vm_end) {
1786 unsigned long size, grow;
1787
1788 size = address - vma->vm_start;
1789 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1790
Hugh Dickins42c36f62011-05-09 17:44:42 -07001791 error = -ENOMEM;
1792 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1793 error = acct_stack_growth(vma, size, grow);
1794 if (!error) {
1795 vma->vm_end = address;
1796 perf_event_mmap(vma);
1797 }
Eric B Munson3af9e852010-05-18 15:30:49 +01001798 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001800 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001801 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 return error;
1803}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001804#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1805
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806/*
1807 * vma is the first one with address < vma->vm_start. Have to extend vma.
1808 */
Michal Hockod05f3162011-05-24 17:11:44 -07001809int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001810 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811{
1812 int error;
1813
1814 /*
1815 * We must make sure the anon_vma is allocated
1816 * so that the anon_vma locking is not a noop.
1817 */
1818 if (unlikely(anon_vma_prepare(vma)))
1819 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001820
1821 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001822 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001823 if (error)
1824 return error;
1825
Rik van Rielbb4a3402010-08-09 17:18:37 -07001826 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
1828 /*
1829 * vma->vm_start/vm_end cannot change under us because the caller
1830 * is required to hold the mmap_sem in read mode. We need the
1831 * anon_vma lock to serialize against concurrent expand_stacks.
1832 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
1834 /* Somebody else might have raced and expanded it already */
1835 if (address < vma->vm_start) {
1836 unsigned long size, grow;
1837
1838 size = vma->vm_end - address;
1839 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1840
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001841 error = -ENOMEM;
1842 if (grow <= vma->vm_pgoff) {
1843 error = acct_stack_growth(vma, size, grow);
1844 if (!error) {
1845 vma->vm_start = address;
1846 vma->vm_pgoff -= grow;
1847 perf_event_mmap(vma);
1848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 }
1850 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001851 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001852 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 return error;
1854}
1855
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001856#ifdef CONFIG_STACK_GROWSUP
1857int expand_stack(struct vm_area_struct *vma, unsigned long address)
1858{
1859 return expand_upwards(vma, address);
1860}
1861
1862struct vm_area_struct *
1863find_extend_vma(struct mm_struct *mm, unsigned long addr)
1864{
1865 struct vm_area_struct *vma, *prev;
1866
1867 addr &= PAGE_MASK;
1868 vma = find_vma_prev(mm, addr, &prev);
1869 if (vma && (vma->vm_start <= addr))
1870 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001871 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001872 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001873 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001874 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001875 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001876 return prev;
1877}
1878#else
1879int expand_stack(struct vm_area_struct *vma, unsigned long address)
1880{
1881 return expand_downwards(vma, address);
1882}
1883
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884struct vm_area_struct *
1885find_extend_vma(struct mm_struct * mm, unsigned long addr)
1886{
1887 struct vm_area_struct * vma;
1888 unsigned long start;
1889
1890 addr &= PAGE_MASK;
1891 vma = find_vma(mm,addr);
1892 if (!vma)
1893 return NULL;
1894 if (vma->vm_start <= addr)
1895 return vma;
1896 if (!(vma->vm_flags & VM_GROWSDOWN))
1897 return NULL;
1898 start = vma->vm_start;
1899 if (expand_stack(vma, addr))
1900 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001901 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001902 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 return vma;
1905}
1906#endif
1907
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001909 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001911 *
1912 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001914static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001916 unsigned long nr_accounted = 0;
1917
Hugh Dickins365e9c872005-10-29 18:16:18 -07001918 /* Update high watermark before we lower total_vm */
1919 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001921 long nrpages = vma_pages(vma);
1922
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001923 if (vma->vm_flags & VM_ACCOUNT)
1924 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001925 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001926 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001927 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001928 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001929 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 validate_mm(mm);
1931}
1932
1933/*
1934 * Get rid of page table information in the indicated region.
1935 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001936 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 */
1938static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001939 struct vm_area_struct *vma, struct vm_area_struct *prev,
1940 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001942 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001943 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
1945 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001946 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001947 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001948 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001949 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1950 next ? next->vm_start : 0);
1951 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952}
1953
1954/*
1955 * Create a list of vma's touched by the unmap, removing them from the mm's
1956 * vma list as we go..
1957 */
1958static void
1959detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1960 struct vm_area_struct *prev, unsigned long end)
1961{
1962 struct vm_area_struct **insertion_point;
1963 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001964 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
1966 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001967 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 do {
1969 rb_erase(&vma->vm_rb, &mm->mm_rb);
1970 mm->map_count--;
1971 tail_vma = vma;
1972 vma = vma->vm_next;
1973 } while (vma && vma->vm_start < end);
1974 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001975 if (vma)
1976 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001978 if (mm->unmap_area == arch_unmap_area)
1979 addr = prev ? prev->vm_end : mm->mmap_base;
1980 else
1981 addr = vma ? vma->vm_start : mm->mmap_base;
1982 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 mm->mmap_cache = NULL; /* Kill the cache. */
1984}
1985
1986/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001987 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1988 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001990static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 unsigned long addr, int new_below)
1992{
1993 struct mempolicy *pol;
1994 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001995 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
Andi Kleena5516432008-07-23 21:27:41 -07001997 if (is_vm_hugetlb_page(vma) && (addr &
1998 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 return -EINVAL;
2000
Christoph Lametere94b1762006-12-06 20:33:17 -08002001 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002003 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
2005 /* most fields are the same, copy all, and then fixup */
2006 *new = *vma;
2007
Rik van Riel5beb4932010-03-05 13:42:07 -08002008 INIT_LIST_HEAD(&new->anon_vma_chain);
2009
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 if (new_below)
2011 new->vm_end = addr;
2012 else {
2013 new->vm_start = addr;
2014 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2015 }
2016
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002017 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08002019 err = PTR_ERR(pol);
2020 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 }
2022 vma_set_policy(new, pol);
2023
Rik van Riel5beb4932010-03-05 13:42:07 -08002024 if (anon_vma_clone(new, vma))
2025 goto out_free_mpol;
2026
Matt Helsley925d1c42008-04-29 01:01:36 -07002027 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002029 if (vma->vm_flags & VM_EXECUTABLE)
2030 added_exe_file_vma(mm);
2031 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
2033 if (new->vm_ops && new->vm_ops->open)
2034 new->vm_ops->open(new);
2035
2036 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002037 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2039 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002040 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Rik van Riel5beb4932010-03-05 13:42:07 -08002042 /* Success. */
2043 if (!err)
2044 return 0;
2045
2046 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002047 if (new->vm_ops && new->vm_ops->close)
2048 new->vm_ops->close(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002049 if (new->vm_file) {
2050 if (vma->vm_flags & VM_EXECUTABLE)
2051 removed_exe_file_vma(mm);
2052 fput(new->vm_file);
2053 }
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002054 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002055 out_free_mpol:
2056 mpol_put(pol);
2057 out_free_vma:
2058 kmem_cache_free(vm_area_cachep, new);
2059 out_err:
2060 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061}
2062
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002063/*
2064 * Split a vma into two pieces at address 'addr', a new vma is allocated
2065 * either for the first part or the tail.
2066 */
2067int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2068 unsigned long addr, int new_below)
2069{
2070 if (mm->map_count >= sysctl_max_map_count)
2071 return -ENOMEM;
2072
2073 return __split_vma(mm, vma, addr, new_below);
2074}
2075
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076/* Munmap is split into 2 main parts -- this part which finds
2077 * what needs doing, and the areas themselves, which do the
2078 * work. This now handles partial unmappings.
2079 * Jeremy Fitzhardinge <jeremy@goop.org>
2080 */
2081int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2082{
2083 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002084 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
2086 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2087 return -EINVAL;
2088
2089 if ((len = PAGE_ALIGN(len)) == 0)
2090 return -EINVAL;
2091
2092 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002093 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002094 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002096 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002097 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
2099 /* if it doesn't overlap, we have nothing.. */
2100 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002101 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 return 0;
2103
2104 /*
2105 * If we need to split any vma, do it now to save pain later.
2106 *
2107 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2108 * unmapped vm_area_struct will remain in use: so lower split_vma
2109 * places tmp vma above, and higher split_vma places tmp vma below.
2110 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002111 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002112 int error;
2113
2114 /*
2115 * Make sure that map_count on return from munmap() will
2116 * not exceed its limit; but let map_count go just above
2117 * its limit temporarily, to help free resources as expected.
2118 */
2119 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2120 return -ENOMEM;
2121
2122 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 if (error)
2124 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002125 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 }
2127
2128 /* Does it split the last one? */
2129 last = find_vma(mm, end);
2130 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002131 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 if (error)
2133 return error;
2134 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002135 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
2137 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002138 * unlock any mlock()ed ranges before detaching vmas
2139 */
2140 if (mm->locked_vm) {
2141 struct vm_area_struct *tmp = vma;
2142 while (tmp && tmp->vm_start < end) {
2143 if (tmp->vm_flags & VM_LOCKED) {
2144 mm->locked_vm -= vma_pages(tmp);
2145 munlock_vma_pages_all(tmp);
2146 }
2147 tmp = tmp->vm_next;
2148 }
2149 }
2150
2151 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 * Remove the vma's, and unmap the actual pages
2153 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002154 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2155 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
2157 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002158 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
2160 return 0;
2161}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162EXPORT_SYMBOL(do_munmap);
2163
Al Virobfce2812012-04-20 21:57:04 -04002164int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002165{
2166 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002167 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002168
2169 down_write(&mm->mmap_sem);
2170 ret = do_munmap(mm, start, len);
2171 up_write(&mm->mmap_sem);
2172 return ret;
2173}
2174EXPORT_SYMBOL(vm_munmap);
2175
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002176SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002179 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180}
2181
2182static inline void verify_mm_writelocked(struct mm_struct *mm)
2183{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002184#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2186 WARN_ON(1);
2187 up_read(&mm->mmap_sem);
2188 }
2189#endif
2190}
2191
2192/*
2193 * this is really a simplified "do_mmap". it only handles
2194 * anonymous maps. eventually we may be able to do some
2195 * brk-specific accounting here.
2196 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002197static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198{
2199 struct mm_struct * mm = current->mm;
2200 struct vm_area_struct * vma, * prev;
2201 unsigned long flags;
2202 struct rb_node ** rb_link, * rb_parent;
2203 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002204 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
2206 len = PAGE_ALIGN(len);
2207 if (!len)
2208 return addr;
2209
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002210 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002211 if (error)
2212 return error;
2213
Kirill Korotaev3a459752006-09-07 14:17:04 +04002214 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2215
Al Viro2c6a1012009-12-03 19:40:46 -05002216 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2217 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002218 return error;
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 /*
2221 * mlock MCL_FUTURE?
2222 */
2223 if (mm->def_flags & VM_LOCKED) {
2224 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002225 locked = len >> PAGE_SHIFT;
2226 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002227 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002228 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2230 return -EAGAIN;
2231 }
2232
2233 /*
2234 * mm->mmap_sem is required to protect against another thread
2235 * changing the mappings in case we sleep.
2236 */
2237 verify_mm_writelocked(mm);
2238
2239 /*
2240 * Clear old maps. this also does some error checking for us
2241 */
2242 munmap_back:
2243 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2244 if (vma && vma->vm_start < addr + len) {
2245 if (do_munmap(mm, addr, len))
2246 return -ENOMEM;
2247 goto munmap_back;
2248 }
2249
2250 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002251 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 return -ENOMEM;
2253
2254 if (mm->map_count > sysctl_max_map_count)
2255 return -ENOMEM;
2256
Al Viro191c5422012-02-13 03:58:52 +00002257 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 return -ENOMEM;
2259
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002261 vma = vma_merge(mm, prev, addr, addr + len, flags,
2262 NULL, NULL, pgoff, NULL);
2263 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 goto out;
2265
2266 /*
2267 * create a vma struct for an anonymous mapping
2268 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002269 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 if (!vma) {
2271 vm_unacct_memory(len >> PAGE_SHIFT);
2272 return -ENOMEM;
2273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
Rik van Riel5beb4932010-03-05 13:42:07 -08002275 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 vma->vm_mm = mm;
2277 vma->vm_start = addr;
2278 vma->vm_end = addr + len;
2279 vma->vm_pgoff = pgoff;
2280 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002281 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 vma_link(mm, vma, prev, rb_link, rb_parent);
2283out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002284 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 mm->total_vm += len >> PAGE_SHIFT;
2286 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002287 if (!mlock_vma_pages_range(vma, addr, addr + len))
2288 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 }
2290 return addr;
2291}
2292
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002293unsigned long vm_brk(unsigned long addr, unsigned long len)
2294{
2295 struct mm_struct *mm = current->mm;
2296 unsigned long ret;
2297
2298 down_write(&mm->mmap_sem);
2299 ret = do_brk(addr, len);
2300 up_write(&mm->mmap_sem);
2301 return ret;
2302}
2303EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305/* Release all mmaps. */
2306void exit_mmap(struct mm_struct *mm)
2307{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002308 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002309 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 unsigned long nr_accounted = 0;
2311
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002312 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002313 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002314
Rik van Rielba470de2008-10-18 20:26:50 -07002315 if (mm->locked_vm) {
2316 vma = mm->mmap;
2317 while (vma) {
2318 if (vma->vm_flags & VM_LOCKED)
2319 munlock_vma_pages_all(vma);
2320 vma = vma->vm_next;
2321 }
2322 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002323
2324 arch_exit_mmap(mm);
2325
Rik van Rielba470de2008-10-18 20:26:50 -07002326 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002327 if (!vma) /* Can happen if dup_mmap() received an OOM */
2328 return;
2329
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 flush_cache_mm(mm);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002332 tlb_gather_mmu(&tlb, mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002333 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002334 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002335 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002336
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002337 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Al Viro853f5e22012-03-05 14:03:47 -05002338 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002341 * Walk the list again, actually closing and freeing it,
2342 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002344 while (vma) {
2345 if (vma->vm_flags & VM_ACCOUNT)
2346 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002347 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002348 }
2349 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002350
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002351 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352}
2353
2354/* Insert vm structure into process list sorted by address
2355 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002356 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 */
2358int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2359{
2360 struct vm_area_struct * __vma, * prev;
2361 struct rb_node ** rb_link, * rb_parent;
2362
2363 /*
2364 * The vm_pgoff of a purely anonymous vma should be irrelevant
2365 * until its first write fault, when page's anon_vma and index
2366 * are set. But now set the vm_pgoff it will almost certainly
2367 * end up with (unless mremap moves it elsewhere before that
2368 * first wfault), so /proc/pid/maps tells a consistent story.
2369 *
2370 * By setting it to reflect the virtual start address of the
2371 * vma, merges and splits can happen in a seamless way, just
2372 * using the existing file pgoff checks and manipulations.
2373 * Similarly in do_mmap_pgoff and in do_brk.
2374 */
2375 if (!vma->vm_file) {
2376 BUG_ON(vma->anon_vma);
2377 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2378 }
2379 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2380 if (__vma && __vma->vm_start < vma->vm_end)
2381 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002382 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002383 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002384 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302385
Ingo Molnar7b2d81d2012-02-17 09:27:41 +01002386 if (vma->vm_file && uprobe_mmap(vma))
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302387 return -EINVAL;
2388
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 vma_link(mm, vma, prev, rb_link, rb_parent);
2390 return 0;
2391}
2392
2393/*
2394 * Copy the vma structure to a new location in the same mm,
2395 * prior to moving page table entries, to effect an mremap move.
2396 */
2397struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2398 unsigned long addr, unsigned long len, pgoff_t pgoff)
2399{
2400 struct vm_area_struct *vma = *vmap;
2401 unsigned long vma_start = vma->vm_start;
2402 struct mm_struct *mm = vma->vm_mm;
2403 struct vm_area_struct *new_vma, *prev;
2404 struct rb_node **rb_link, *rb_parent;
2405 struct mempolicy *pol;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002406 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
2408 /*
2409 * If anonymous vma has not yet been faulted, update new pgoff
2410 * to match new location, to increase its chance of merging.
2411 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002412 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002414 faulted_in_anon_vma = false;
2415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
2417 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2418 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2419 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2420 if (new_vma) {
2421 /*
2422 * Source vma may have been merged into new_vma
2423 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002424 if (unlikely(vma_start >= new_vma->vm_start &&
2425 vma_start < new_vma->vm_end)) {
2426 /*
2427 * The only way we can get a vma_merge with
2428 * self during an mremap is if the vma hasn't
2429 * been faulted in yet and we were allowed to
2430 * reset the dst vma->vm_pgoff to the
2431 * destination address of the mremap to allow
2432 * the merge to happen. mremap must change the
2433 * vm_pgoff linearity between src and dst vmas
2434 * (in turn preventing a vma_merge) to be
2435 * safe. It is only safe to keep the vm_pgoff
2436 * linear if there are no pages mapped yet.
2437 */
2438 VM_BUG_ON(faulted_in_anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 *vmap = new_vma;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002440 } else
2441 anon_vma_moveto_tail(new_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002443 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 if (new_vma) {
2445 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002446 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002447 if (IS_ERR(pol))
2448 goto out_free_vma;
2449 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2450 if (anon_vma_clone(new_vma, vma))
2451 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 vma_set_policy(new_vma, pol);
2453 new_vma->vm_start = addr;
2454 new_vma->vm_end = addr + len;
2455 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002456 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 get_file(new_vma->vm_file);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302458
Ingo Molnar7b2d81d2012-02-17 09:27:41 +01002459 if (uprobe_mmap(new_vma))
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302460 goto out_free_mempol;
2461
Matt Helsley925d1c42008-04-29 01:01:36 -07002462 if (vma->vm_flags & VM_EXECUTABLE)
2463 added_exe_file_vma(mm);
2464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (new_vma->vm_ops && new_vma->vm_ops->open)
2466 new_vma->vm_ops->open(new_vma);
2467 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2468 }
2469 }
2470 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002471
2472 out_free_mempol:
2473 mpol_put(pol);
2474 out_free_vma:
2475 kmem_cache_free(vm_area_cachep, new_vma);
2476 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002478
2479/*
2480 * Return true if the calling process may expand its vm space by the passed
2481 * number of pages
2482 */
2483int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2484{
2485 unsigned long cur = mm->total_vm; /* pages */
2486 unsigned long lim;
2487
Jiri Slaby59e99e52010-03-05 13:41:44 -08002488 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002489
2490 if (cur + npages > lim)
2491 return 0;
2492 return 1;
2493}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002494
2495
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002496static int special_mapping_fault(struct vm_area_struct *vma,
2497 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002498{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002499 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002500 struct page **pages;
2501
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002502 /*
2503 * special mappings have no vm_file, and in that case, the mm
2504 * uses vm_pgoff internally. So we have to subtract it from here.
2505 * We are allowed to do this because we are the mm; do not copy
2506 * this code into drivers!
2507 */
2508 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002509
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002510 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2511 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002512
2513 if (*pages) {
2514 struct page *page = *pages;
2515 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002516 vmf->page = page;
2517 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002518 }
2519
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002520 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002521}
2522
2523/*
2524 * Having a close hook prevents vma merging regardless of flags.
2525 */
2526static void special_mapping_close(struct vm_area_struct *vma)
2527{
2528}
2529
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002530static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002531 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002532 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002533};
2534
2535/*
2536 * Called with mm->mmap_sem held for writing.
2537 * Insert a new vma covering the given region, with the given flags.
2538 * Its pages are supplied by the given array of struct page *.
2539 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2540 * The region past the last page supplied will always produce SIGBUS.
2541 * The array pointer and the pages it points to are assumed to stay alive
2542 * for as long as this mapping might exist.
2543 */
2544int install_special_mapping(struct mm_struct *mm,
2545 unsigned long addr, unsigned long len,
2546 unsigned long vm_flags, struct page **pages)
2547{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002548 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002549 struct vm_area_struct *vma;
2550
2551 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2552 if (unlikely(vma == NULL))
2553 return -ENOMEM;
2554
Rik van Riel5beb4932010-03-05 13:42:07 -08002555 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002556 vma->vm_mm = mm;
2557 vma->vm_start = addr;
2558 vma->vm_end = addr + len;
2559
Nick Piggin2f987352008-02-02 03:08:53 +01002560 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002561 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002562
2563 vma->vm_ops = &special_mapping_vmops;
2564 vma->vm_private_data = pages;
2565
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002566 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2567 if (ret)
2568 goto out;
2569
2570 ret = insert_vm_struct(mm, vma);
2571 if (ret)
2572 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002573
2574 mm->total_vm += len >> PAGE_SHIFT;
2575
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002576 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002577
Roland McGrathfa5dc222007-02-08 14:20:41 -08002578 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002579
2580out:
2581 kmem_cache_free(vm_area_cachep, vma);
2582 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002583}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002584
2585static DEFINE_MUTEX(mm_all_locks_mutex);
2586
Peter Zijlstra454ed842008-08-11 09:30:25 +02002587static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002588{
Rik van Riel012f18002010-08-09 17:18:40 -07002589 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002590 /*
2591 * The LSB of head.next can't change from under us
2592 * because we hold the mm_all_locks_mutex.
2593 */
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002594 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002595 /*
2596 * We can safely modify head.next after taking the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002597 * anon_vma->root->mutex. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002598 * the same anon_vma we won't take it again.
2599 *
2600 * No need of atomic instructions here, head.next
2601 * can't change from under us thanks to the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002602 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002603 */
2604 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002605 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002606 BUG();
2607 }
2608}
2609
Peter Zijlstra454ed842008-08-11 09:30:25 +02002610static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002611{
2612 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2613 /*
2614 * AS_MM_ALL_LOCKS can't change from under us because
2615 * we hold the mm_all_locks_mutex.
2616 *
2617 * Operations on ->flags have to be atomic because
2618 * even if AS_MM_ALL_LOCKS is stable thanks to the
2619 * mm_all_locks_mutex, there may be other cpus
2620 * changing other bitflags in parallel to us.
2621 */
2622 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2623 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002624 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002625 }
2626}
2627
2628/*
2629 * This operation locks against the VM for all pte/vma/mm related
2630 * operations that could ever happen on a certain mm. This includes
2631 * vmtruncate, try_to_unmap, and all page faults.
2632 *
2633 * The caller must take the mmap_sem in write mode before calling
2634 * mm_take_all_locks(). The caller isn't allowed to release the
2635 * mmap_sem until mm_drop_all_locks() returns.
2636 *
2637 * mmap_sem in write mode is required in order to block all operations
2638 * that could modify pagetables and free pages without need of
2639 * altering the vma layout (for example populate_range() with
2640 * nonlinear vmas). It's also needed in write mode to avoid new
2641 * anon_vmas to be associated with existing vmas.
2642 *
2643 * A single task can't take more than one mm_take_all_locks() in a row
2644 * or it would deadlock.
2645 *
2646 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2647 * mapping->flags avoid to take the same lock twice, if more than one
2648 * vma in this mm is backed by the same anon_vma or address_space.
2649 *
2650 * We can take all the locks in random order because the VM code
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002651 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002652 * takes more than one of them in a row. Secondly we're protected
2653 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2654 *
2655 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2656 * that may have to take thousand of locks.
2657 *
2658 * mm_take_all_locks() can fail if it's interrupted by signals.
2659 */
2660int mm_take_all_locks(struct mm_struct *mm)
2661{
2662 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002663 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002664
2665 BUG_ON(down_read_trylock(&mm->mmap_sem));
2666
2667 mutex_lock(&mm_all_locks_mutex);
2668
2669 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2670 if (signal_pending(current))
2671 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002672 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002673 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002674 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002675
2676 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2677 if (signal_pending(current))
2678 goto out_unlock;
2679 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002680 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2681 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002682 }
2683
Kautuk Consul584cff52011-10-31 17:08:59 -07002684 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002685
2686out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07002687 mm_drop_all_locks(mm);
2688 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002689}
2690
2691static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2692{
Rik van Riel012f18002010-08-09 17:18:40 -07002693 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002694 /*
2695 * The LSB of head.next can't change to 0 from under
2696 * us because we hold the mm_all_locks_mutex.
2697 *
2698 * We must however clear the bitflag before unlocking
2699 * the vma so the users using the anon_vma->head will
2700 * never see our bitflag.
2701 *
2702 * No need of atomic instructions here, head.next
2703 * can't change from under us until we release the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002704 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002705 */
2706 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002707 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002708 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002709 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002710 }
2711}
2712
2713static void vm_unlock_mapping(struct address_space *mapping)
2714{
2715 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2716 /*
2717 * AS_MM_ALL_LOCKS can't change to 0 from under us
2718 * because we hold the mm_all_locks_mutex.
2719 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002720 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002721 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2722 &mapping->flags))
2723 BUG();
2724 }
2725}
2726
2727/*
2728 * The mmap_sem cannot be released by the caller until
2729 * mm_drop_all_locks() returns.
2730 */
2731void mm_drop_all_locks(struct mm_struct *mm)
2732{
2733 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002734 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002735
2736 BUG_ON(down_read_trylock(&mm->mmap_sem));
2737 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2738
2739 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2740 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002741 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2742 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002743 if (vma->vm_file && vma->vm_file->f_mapping)
2744 vm_unlock_mapping(vma->vm_file->f_mapping);
2745 }
2746
2747 mutex_unlock(&mm_all_locks_mutex);
2748}
David Howells8feae132009-01-08 12:04:47 +00002749
2750/*
2751 * initialise the VMA slab
2752 */
2753void __init mmap_init(void)
2754{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002755 int ret;
2756
2757 ret = percpu_counter_init(&vm_committed_as, 0);
2758 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002759}