blob: 5ac533f88e99ccf1ef2b13a146b8c076e7949402 [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))
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700202 list_del_init(&vma->shared.nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700204 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 flush_dcache_mmap_unlock(mapping);
206}
207
208/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700209 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700210 * 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);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700234 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700235 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700236 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700238 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700241static unsigned long do_brk(unsigned long addr, unsigned long len);
242
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100243SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 unsigned long rlim, retval;
246 unsigned long newbrk, oldbrk;
247 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700248 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 down_write(&mm->mmap_sem);
251
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700252#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800253 /*
254 * CONFIG_COMPAT_BRK can still be overridden by setting
255 * randomize_va_space to 2, which will still cause mm->start_brk
256 * to be arbitrarily shifted
257 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700258 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800259 min_brk = mm->start_brk;
260 else
261 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700262#else
263 min_brk = mm->start_brk;
264#endif
265 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700267
268 /*
269 * Check against rlimit here. If this check is done later after the test
270 * of oldbrk with newbrk then it can escape the test and let the data
271 * segment grow beyond its set limit the in case where the limit is
272 * not page aligned -Ram Gupta
273 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800274 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100275 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
276 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700277 goto out;
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 newbrk = PAGE_ALIGN(brk);
280 oldbrk = PAGE_ALIGN(mm->brk);
281 if (oldbrk == newbrk)
282 goto set_brk;
283
284 /* Always allow shrinking brk. */
285 if (brk <= mm->brk) {
286 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
287 goto set_brk;
288 goto out;
289 }
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 /* Check against existing mmap mappings. */
292 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
293 goto out;
294
295 /* Ok, looks good - let it rip. */
296 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
297 goto out;
298set_brk:
299 mm->brk = brk;
300out:
301 retval = mm->brk;
302 up_write(&mm->mmap_sem);
303 return retval;
304}
305
306#ifdef DEBUG_MM_RB
307static int browse_rb(struct rb_root *root)
308{
309 int i = 0, j;
310 struct rb_node *nd, *pn = NULL;
311 unsigned long prev = 0, pend = 0;
312
313 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
314 struct vm_area_struct *vma;
315 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
316 if (vma->vm_start < prev)
317 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
318 if (vma->vm_start < pend)
319 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
320 if (vma->vm_start > vma->vm_end)
321 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
322 i++;
323 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800324 prev = vma->vm_start;
325 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 }
327 j = 0;
328 for (nd = pn; nd; nd = rb_prev(nd)) {
329 j++;
330 }
331 if (i != j)
332 printk("backwards %d, forwards %d\n", j, i), i = 0;
333 return i;
334}
335
336void validate_mm(struct mm_struct *mm)
337{
338 int bug = 0;
339 int i = 0;
340 struct vm_area_struct *tmp = mm->mmap;
341 while (tmp) {
342 tmp = tmp->vm_next;
343 i++;
344 }
345 if (i != mm->map_count)
346 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
347 i = browse_rb(&mm->mm_rb);
348 if (i != mm->map_count)
349 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200350 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351}
352#else
353#define validate_mm(mm) do { } while (0)
354#endif
355
Hugh Dickins6597d782012-10-08 16:29:07 -0700356static int find_vma_links(struct mm_struct *mm, unsigned long addr,
357 unsigned long end, struct vm_area_struct **pprev,
358 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Hugh Dickins6597d782012-10-08 16:29:07 -0700360 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 __rb_link = &mm->mm_rb.rb_node;
363 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 while (*__rb_link) {
366 struct vm_area_struct *vma_tmp;
367
368 __rb_parent = *__rb_link;
369 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
370
371 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700372 /* Fail if an existing vma overlaps the area */
373 if (vma_tmp->vm_start < end)
374 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 __rb_link = &__rb_parent->rb_left;
376 } else {
377 rb_prev = __rb_parent;
378 __rb_link = &__rb_parent->rb_right;
379 }
380 }
381
382 *pprev = NULL;
383 if (rb_prev)
384 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
385 *rb_link = __rb_link;
386 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700387 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
391 struct rb_node **rb_link, struct rb_node *rb_parent)
392{
393 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
394 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
395}
396
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700397static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
ZhenwenXu48aae422009-01-06 14:40:21 -0800399 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
401 file = vma->vm_file;
402 if (file) {
403 struct address_space *mapping = file->f_mapping;
404
405 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800406 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 if (vma->vm_flags & VM_SHARED)
408 mapping->i_mmap_writable++;
409
410 flush_dcache_mmap_lock(mapping);
411 if (unlikely(vma->vm_flags & VM_NONLINEAR))
412 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
413 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700414 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 flush_dcache_mmap_unlock(mapping);
416 }
417}
418
419static void
420__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
421 struct vm_area_struct *prev, struct rb_node **rb_link,
422 struct rb_node *rb_parent)
423{
424 __vma_link_list(mm, vma, prev, rb_parent);
425 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
428static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
429 struct vm_area_struct *prev, struct rb_node **rb_link,
430 struct rb_node *rb_parent)
431{
432 struct address_space *mapping = NULL;
433
434 if (vma->vm_file)
435 mapping = vma->vm_file->f_mapping;
436
Peter Zijlstra97a89412011-05-24 17:12:04 -0700437 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700438 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440 __vma_link(mm, vma, prev, rb_link, rb_parent);
441 __vma_link_file(vma);
442
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700444 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 mm->map_count++;
447 validate_mm(mm);
448}
449
450/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700451 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700452 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800454static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Hugh Dickins6597d782012-10-08 16:29:07 -0700456 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800457 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Hugh Dickins6597d782012-10-08 16:29:07 -0700459 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
460 &prev, &rb_link, &rb_parent))
461 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 __vma_link(mm, vma, prev, rb_link, rb_parent);
463 mm->map_count++;
464}
465
466static inline void
467__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
468 struct vm_area_struct *prev)
469{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700470 struct vm_area_struct *next = vma->vm_next;
471
472 prev->vm_next = next;
473 if (next)
474 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 rb_erase(&vma->vm_rb, &mm->mm_rb);
476 if (mm->mmap_cache == vma)
477 mm->mmap_cache = prev;
478}
479
480/*
481 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
482 * is already present in an i_mmap tree without adjusting the tree.
483 * The following helper function should be used when such adjustments
484 * are necessary. The "insert" vma (if any) is to be inserted
485 * before we drop the necessary locks.
486 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800487int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
489{
490 struct mm_struct *mm = vma->vm_mm;
491 struct vm_area_struct *next = vma->vm_next;
492 struct vm_area_struct *importer = NULL;
493 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700494 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700495 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 long adjust_next = 0;
498 int remove_next = 0;
499
500 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700501 struct vm_area_struct *exporter = NULL;
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 if (end >= next->vm_end) {
504 /*
505 * vma expands, overlapping all the next, and
506 * perhaps the one after too (mprotect case 6).
507 */
508again: remove_next = 1 + (end > next->vm_end);
509 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700510 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 importer = vma;
512 } else if (end > next->vm_start) {
513 /*
514 * vma expands, overlapping part of the next:
515 * mprotect case 5 shifting the boundary up.
516 */
517 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700518 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 importer = vma;
520 } else if (end < vma->vm_end) {
521 /*
522 * vma shrinks, and !insert tells it's not
523 * split_vma inserting another: so it must be
524 * mprotect case 4 shifting the boundary down.
525 */
526 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700527 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 importer = next;
529 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Rik van Riel5beb4932010-03-05 13:42:07 -0800531 /*
532 * Easily overlooked: when mprotect shifts the boundary,
533 * make sure the expanding vma has anon_vma set if the
534 * shrinking vma had, to cover any anon pages imported.
535 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700536 if (exporter && exporter->anon_vma && !importer->anon_vma) {
537 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800538 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700539 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800540 }
541 }
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 if (file) {
544 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530545 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530547 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530548
549 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530550 uprobe_munmap(next, next->vm_start,
551 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530552 }
553
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700554 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700557 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 * are visible to arm/parisc __flush_dcache_page
559 * throughout; but we cannot insert into address
560 * space until vma start or end is updated.
561 */
562 __vma_link_file(insert);
563 }
564 }
565
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800566 vma_adjust_trans_huge(vma, start, end, adjust_next);
567
Rik van Riel012f18002010-08-09 17:18:40 -0700568 /*
569 * When changing only vma->vm_end, we don't really need anon_vma
570 * lock. This is a fairly rare case by itself, but the anon_vma
571 * lock may be shared between many sibling processes. Skipping
572 * the lock for brk adjustments makes a difference sometimes.
573 */
Shaohua Li5f70b962011-05-24 17:11:19 -0700574 if (vma->anon_vma && (importer || start != vma->vm_start)) {
Rik van Riel012f18002010-08-09 17:18:40 -0700575 anon_vma = vma->anon_vma;
Michel Lespinasseca42b262012-10-08 16:30:01 -0700576 VM_BUG_ON(adjust_next && next->anon_vma &&
577 anon_vma != next->anon_vma);
578 } else if (adjust_next && next->anon_vma)
579 anon_vma = next->anon_vma;
580 if (anon_vma)
Rik van Riel012f18002010-08-09 17:18:40 -0700581 anon_vma_lock(anon_vma);
Rik van Riel012f18002010-08-09 17:18:40 -0700582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 if (root) {
584 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700585 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700587 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
589
590 vma->vm_start = start;
591 vma->vm_end = end;
592 vma->vm_pgoff = pgoff;
593 if (adjust_next) {
594 next->vm_start += adjust_next << PAGE_SHIFT;
595 next->vm_pgoff += adjust_next;
596 }
597
598 if (root) {
599 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700600 vma_interval_tree_insert(next, root);
601 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 flush_dcache_mmap_unlock(mapping);
603 }
604
605 if (remove_next) {
606 /*
607 * vma_merge has merged next into vma, and needs
608 * us to remove next before dropping the locks.
609 */
610 __vma_unlink(mm, next, vma);
611 if (file)
612 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 } else if (insert) {
614 /*
615 * split_vma has split insert from vma, and needs
616 * us to insert it before dropping the locks
617 * (it may either follow vma or precede it).
618 */
619 __insert_vm_struct(mm, insert);
620 }
621
Rik van Riel012f18002010-08-09 17:18:40 -0700622 if (anon_vma)
623 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700625 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530627 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100628 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530629
630 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100631 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530632 }
633
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700635 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530636 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700638 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800639 if (next->anon_vma)
640 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700642 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 kmem_cache_free(vm_area_cachep, next);
644 /*
645 * In mprotect's case 6 (see comments on vma_merge),
646 * we must remove another next too. It would clutter
647 * up the code too much to do both in one go.
648 */
649 if (remove_next == 2) {
650 next = vma->vm_next;
651 goto again;
652 }
653 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530654 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100655 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800658
659 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
662/*
663 * If the vma has a ->close operation then the driver probably needs to release
664 * per-vma resources, so we don't attempt to merge those.
665 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666static inline int is_mergeable_vma(struct vm_area_struct *vma,
667 struct file *file, unsigned long vm_flags)
668{
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700669 if (vma->vm_flags ^ vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 return 0;
671 if (vma->vm_file != file)
672 return 0;
673 if (vma->vm_ops && vma->vm_ops->close)
674 return 0;
675 return 1;
676}
677
678static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700679 struct anon_vma *anon_vma2,
680 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Shaohua Li965f55d2011-05-24 17:11:20 -0700682 /*
683 * The list_is_singular() test is to avoid merging VMA cloned from
684 * parents. This can improve scalability caused by anon_vma lock.
685 */
686 if ((!anon_vma1 || !anon_vma2) && (!vma ||
687 list_is_singular(&vma->anon_vma_chain)))
688 return 1;
689 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
692/*
693 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
694 * in front of (at a lower virtual address and file offset than) the vma.
695 *
696 * We cannot merge two vmas if they have differently assigned (non-NULL)
697 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
698 *
699 * We don't check here for the merged mmap wrapping around the end of pagecache
700 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
701 * wrap, nor mmaps which cover the final page at index -1UL.
702 */
703static int
704can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
705 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
706{
707 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700708 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 if (vma->vm_pgoff == vm_pgoff)
710 return 1;
711 }
712 return 0;
713}
714
715/*
716 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
717 * beyond (at a higher virtual address and file offset than) the vma.
718 *
719 * We cannot merge two vmas if they have differently assigned (non-NULL)
720 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
721 */
722static int
723can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
724 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
725{
726 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700727 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 pgoff_t vm_pglen;
729 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
730 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
731 return 1;
732 }
733 return 0;
734}
735
736/*
737 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
738 * whether that can be merged with its predecessor or its successor.
739 * Or both (it neatly fills a hole).
740 *
741 * In most cases - when called for mmap, brk or mremap - [addr,end) is
742 * certain not to be mapped by the time vma_merge is called; but when
743 * called for mprotect, it is certain to be already mapped (either at
744 * an offset within prev, or at the start of next), and the flags of
745 * this area are about to be changed to vm_flags - and the no-change
746 * case has already been eliminated.
747 *
748 * The following mprotect cases have to be considered, where AAAA is
749 * the area passed down from mprotect_fixup, never extending beyond one
750 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
751 *
752 * AAAA AAAA AAAA AAAA
753 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
754 * cannot merge might become might become might become
755 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
756 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
757 * mremap move: PPPPNNNNNNNN 8
758 * AAAA
759 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
760 * might become case 1 below case 2 below case 3 below
761 *
762 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
763 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
764 */
765struct vm_area_struct *vma_merge(struct mm_struct *mm,
766 struct vm_area_struct *prev, unsigned long addr,
767 unsigned long end, unsigned long vm_flags,
768 struct anon_vma *anon_vma, struct file *file,
769 pgoff_t pgoff, struct mempolicy *policy)
770{
771 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
772 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800773 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
775 /*
776 * We later require that vma->vm_flags == vm_flags,
777 * so this tests vma->vm_flags & VM_SPECIAL, too.
778 */
779 if (vm_flags & VM_SPECIAL)
780 return NULL;
781
782 if (prev)
783 next = prev->vm_next;
784 else
785 next = mm->mmap;
786 area = next;
787 if (next && next->vm_end == end) /* cases 6, 7, 8 */
788 next = next->vm_next;
789
790 /*
791 * Can it merge with the predecessor?
792 */
793 if (prev && prev->vm_end == addr &&
794 mpol_equal(vma_policy(prev), policy) &&
795 can_vma_merge_after(prev, vm_flags,
796 anon_vma, file, pgoff)) {
797 /*
798 * OK, it can. Can we now merge in the successor as well?
799 */
800 if (next && end == next->vm_start &&
801 mpol_equal(policy, vma_policy(next)) &&
802 can_vma_merge_before(next, vm_flags,
803 anon_vma, file, pgoff+pglen) &&
804 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700805 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800807 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 next->vm_end, prev->vm_pgoff, NULL);
809 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800810 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800812 if (err)
813 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800814 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 return prev;
816 }
817
818 /*
819 * Can this new request be merged in front of next?
820 */
821 if (next && end == next->vm_start &&
822 mpol_equal(policy, vma_policy(next)) &&
823 can_vma_merge_before(next, vm_flags,
824 anon_vma, file, pgoff+pglen)) {
825 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800826 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 addr, prev->vm_pgoff, NULL);
828 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800829 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800831 if (err)
832 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800833 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 return area;
835 }
836
837 return NULL;
838}
839
840/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700841 * Rough compatbility check to quickly see if it's even worth looking
842 * at sharing an anon_vma.
843 *
844 * They need to have the same vm_file, and the flags can only differ
845 * in things that mprotect may change.
846 *
847 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
848 * we can merge the two vma's. For example, we refuse to merge a vma if
849 * there is a vm_ops->close() function, because that indicates that the
850 * driver is doing some kind of reference counting. But that doesn't
851 * really matter for the anon_vma sharing case.
852 */
853static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
854{
855 return a->vm_end == b->vm_start &&
856 mpol_equal(vma_policy(a), vma_policy(b)) &&
857 a->vm_file == b->vm_file &&
858 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
859 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
860}
861
862/*
863 * Do some basic sanity checking to see if we can re-use the anon_vma
864 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
865 * the same as 'old', the other will be the new one that is trying
866 * to share the anon_vma.
867 *
868 * NOTE! This runs with mm_sem held for reading, so it is possible that
869 * the anon_vma of 'old' is concurrently in the process of being set up
870 * by another page fault trying to merge _that_. But that's ok: if it
871 * is being set up, that automatically means that it will be a singleton
872 * acceptable for merging, so we can do all of this optimistically. But
873 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
874 *
875 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
876 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
877 * is to return an anon_vma that is "complex" due to having gone through
878 * a fork).
879 *
880 * We also make sure that the two vma's are compatible (adjacent,
881 * and with the same memory policies). That's all stable, even with just
882 * a read lock on the mm_sem.
883 */
884static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
885{
886 if (anon_vma_compatible(a, b)) {
887 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
888
889 if (anon_vma && list_is_singular(&old->anon_vma_chain))
890 return anon_vma;
891 }
892 return NULL;
893}
894
895/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
897 * neighbouring vmas for a suitable anon_vma, before it goes off
898 * to allocate a new anon_vma. It checks because a repetitive
899 * sequence of mprotects and faults may otherwise lead to distinct
900 * anon_vmas being allocated, preventing vma merge in subsequent
901 * mprotect.
902 */
903struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
904{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700905 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 near = vma->vm_next;
909 if (!near)
910 goto try_prev;
911
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700912 anon_vma = reusable_anon_vma(near, vma, near);
913 if (anon_vma)
914 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -0700916 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 if (!near)
918 goto none;
919
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700920 anon_vma = reusable_anon_vma(near, near, vma);
921 if (anon_vma)
922 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923none:
924 /*
925 * There's no absolute need to look only at touching neighbours:
926 * we could search further afield for "compatible" anon_vmas.
927 * But it would probably just be a waste of time searching,
928 * or lead to too many vmas hanging off the same anon_vma.
929 * We're trying to allow mprotect remerging later on,
930 * not trying to minimize memory used for anon_vmas.
931 */
932 return NULL;
933}
934
935#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700936void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 struct file *file, long pages)
938{
939 const unsigned long stack_flags
940 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
941
Huang Shijie44de9d02012-07-31 16:41:49 -0700942 mm->total_vm += pages;
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (file) {
945 mm->shared_vm += pages;
946 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
947 mm->exec_vm += pages;
948 } else if (flags & stack_flags)
949 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950}
951#endif /* CONFIG_PROC_FS */
952
953/*
Al Viro40401532012-02-13 03:58:52 +0000954 * If a hint addr is less than mmap_min_addr change hint to be as
955 * low as possible but still greater than mmap_min_addr
956 */
957static inline unsigned long round_hint_to_min(unsigned long hint)
958{
959 hint &= PAGE_MASK;
960 if (((void *)hint != NULL) &&
961 (hint < mmap_min_addr))
962 return PAGE_ALIGN(mmap_min_addr);
963 return hint;
964}
965
966/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700967 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 */
969
Al Viroe3fc6292012-05-30 20:08:42 -0400970unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 unsigned long len, unsigned long prot,
972 unsigned long flags, unsigned long pgoff)
973{
974 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 struct inode *inode;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900976 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /*
979 * Does the application expect PROT_READ to imply PROT_EXEC?
980 *
981 * (the exception is when the underlying filesystem is noexec
982 * mounted, in which case we dont add PROT_EXEC.)
983 */
984 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800985 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 prot |= PROT_EXEC;
987
988 if (!len)
989 return -EINVAL;
990
Eric Paris7cd94142007-11-26 18:47:40 -0500991 if (!(flags & MAP_FIXED))
992 addr = round_hint_to_min(addr);
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 /* Careful about overflows.. */
995 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500996 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 return -ENOMEM;
998
999 /* offset overflow? */
1000 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1001 return -EOVERFLOW;
1002
1003 /* Too many mappings? */
1004 if (mm->map_count > sysctl_max_map_count)
1005 return -ENOMEM;
1006
1007 /* Obtain the address to map to. we verify (or select) it and ensure
1008 * that it represents a valid section of the address space.
1009 */
1010 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1011 if (addr & ~PAGE_MASK)
1012 return addr;
1013
1014 /* Do simple checking here so the lower-level routines won't have
1015 * to. we assume access permissions have been handled by the open
1016 * of the memory object, so we don't do any here.
1017 */
1018 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1019 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1020
Huang Shijiecdf7b342009-09-21 17:03:36 -07001021 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 if (!can_do_mlock())
1023 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 /* mlock MCL_FUTURE? */
1026 if (vm_flags & VM_LOCKED) {
1027 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001028 locked = len >> PAGE_SHIFT;
1029 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001030 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001031 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1033 return -EAGAIN;
1034 }
1035
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001036 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 if (file) {
1039 switch (flags & MAP_TYPE) {
1040 case MAP_SHARED:
1041 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1042 return -EACCES;
1043
1044 /*
1045 * Make sure we don't allow writing to an append-only
1046 * file..
1047 */
1048 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1049 return -EACCES;
1050
1051 /*
1052 * Make sure there are no mandatory locks on the file.
1053 */
1054 if (locks_verify_locked(inode))
1055 return -EAGAIN;
1056
1057 vm_flags |= VM_SHARED | VM_MAYSHARE;
1058 if (!(file->f_mode & FMODE_WRITE))
1059 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1060
1061 /* fall through */
1062 case MAP_PRIVATE:
1063 if (!(file->f_mode & FMODE_READ))
1064 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001065 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001066 if (vm_flags & VM_EXEC)
1067 return -EPERM;
1068 vm_flags &= ~VM_MAYEXEC;
1069 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001070
1071 if (!file->f_op || !file->f_op->mmap)
1072 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 break;
1074
1075 default:
1076 return -EINVAL;
1077 }
1078 } else {
1079 switch (flags & MAP_TYPE) {
1080 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001081 /*
1082 * Ignore pgoff.
1083 */
1084 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 vm_flags |= VM_SHARED | VM_MAYSHARE;
1086 break;
1087 case MAP_PRIVATE:
1088 /*
1089 * Set pgoff according to addr for anon_vma.
1090 */
1091 pgoff = addr >> PAGE_SHIFT;
1092 break;
1093 default:
1094 return -EINVAL;
1095 }
1096 }
1097
Mel Gorman5a6fe122009-02-10 14:02:27 +00001098 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001099}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001100
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001101SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1102 unsigned long, prot, unsigned long, flags,
1103 unsigned long, fd, unsigned long, pgoff)
1104{
1105 struct file *file = NULL;
1106 unsigned long retval = -EBADF;
1107
1108 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001109 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001110 if (unlikely(flags & MAP_HUGETLB))
1111 return -EINVAL;
1112 file = fget(fd);
1113 if (!file)
1114 goto out;
1115 } else if (flags & MAP_HUGETLB) {
1116 struct user_struct *user = NULL;
1117 /*
1118 * VM_NORESERVE is used because the reservations will be
1119 * taken when vm_ops->mmap() is called
1120 * A dummy user value is used because we are not locking
1121 * memory so no accounting is necessary
1122 */
Steven Truelove40716e22012-03-21 16:34:14 -07001123 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1124 VM_NORESERVE, &user,
1125 HUGETLB_ANONHUGE_INODE);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001126 if (IS_ERR(file))
1127 return PTR_ERR(file);
1128 }
1129
1130 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1131
Al Viroeb36c582012-05-30 20:17:35 -04001132 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001133 if (file)
1134 fput(file);
1135out:
1136 return retval;
1137}
1138
Christoph Hellwiga4679372010-03-10 15:21:15 -08001139#ifdef __ARCH_WANT_SYS_OLD_MMAP
1140struct mmap_arg_struct {
1141 unsigned long addr;
1142 unsigned long len;
1143 unsigned long prot;
1144 unsigned long flags;
1145 unsigned long fd;
1146 unsigned long offset;
1147};
1148
1149SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1150{
1151 struct mmap_arg_struct a;
1152
1153 if (copy_from_user(&a, arg, sizeof(a)))
1154 return -EFAULT;
1155 if (a.offset & ~PAGE_MASK)
1156 return -EINVAL;
1157
1158 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1159 a.offset >> PAGE_SHIFT);
1160}
1161#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1162
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001163/*
1164 * Some shared mappigns will want the pages marked read-only
1165 * to track write events. If so, we'll downgrade vm_page_prot
1166 * to the private version (using protection_map[] without the
1167 * VM_SHARED bit).
1168 */
1169int vma_wants_writenotify(struct vm_area_struct *vma)
1170{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001171 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001172
1173 /* If it was private or non-writable, the write bit is already clear */
1174 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1175 return 0;
1176
1177 /* The backer wishes to know when pages are first written to? */
1178 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1179 return 1;
1180
1181 /* The open routine did something to the protections already? */
1182 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001183 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001184 return 0;
1185
1186 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001187 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001188 return 0;
1189
1190 /* Can the mapping track the dirty pages? */
1191 return vma->vm_file && vma->vm_file->f_mapping &&
1192 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1193}
1194
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001195/*
1196 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001197 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001198 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001199static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001200{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001201 /*
1202 * hugetlb has its own accounting separate from the core VM
1203 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1204 */
1205 if (file && is_file_hugepages(file))
1206 return 0;
1207
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001208 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1209}
1210
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001211unsigned long mmap_region(struct file *file, unsigned long addr,
1212 unsigned long len, unsigned long flags,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001213 vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001214{
1215 struct mm_struct *mm = current->mm;
1216 struct vm_area_struct *vma, *prev;
1217 int correct_wcount = 0;
1218 int error;
1219 struct rb_node **rb_link, *rb_parent;
1220 unsigned long charged = 0;
1221 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 /* Clear old maps */
1224 error = -ENOMEM;
1225munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001226 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (do_munmap(mm, addr, len))
1228 return -ENOMEM;
1229 goto munmap_back;
1230 }
1231
1232 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001233 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 return -ENOMEM;
1235
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001236 /*
1237 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001238 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001239 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001240 if ((flags & MAP_NORESERVE)) {
1241 /* We honor MAP_NORESERVE if allowed to overcommit */
1242 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1243 vm_flags |= VM_NORESERVE;
1244
1245 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1246 if (file && is_file_hugepages(file))
1247 vm_flags |= VM_NORESERVE;
1248 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001249
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001250 /*
1251 * Private writable mapping: check memory availability
1252 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001253 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001254 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001255 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001256 return -ENOMEM;
1257 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 }
1259
1260 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001261 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001263 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1264 if (vma)
1265 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
1267 /*
1268 * Determine the object being mapped and call the appropriate
1269 * specific mapper. the address has already been validated, but
1270 * not unmapped, but the maps are removed from the list.
1271 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001272 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 if (!vma) {
1274 error = -ENOMEM;
1275 goto unacct_error;
1276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278 vma->vm_mm = mm;
1279 vma->vm_start = addr;
1280 vma->vm_end = addr + len;
1281 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001282 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001284 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Hugh Dickinsce8fea72012-03-06 12:28:52 -08001286 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1290 goto free_vma;
1291 if (vm_flags & VM_DENYWRITE) {
1292 error = deny_write_access(file);
1293 if (error)
1294 goto free_vma;
1295 correct_wcount = 1;
1296 }
Al Virocb0942b2012-08-27 14:48:26 -04001297 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 error = file->f_op->mmap(file, vma);
1299 if (error)
1300 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001301
1302 /* Can addr have changed??
1303 *
1304 * Answer: Yes, several device drivers can do it in their
1305 * f_op->mmap method. -DaveM
1306 */
1307 addr = vma->vm_start;
1308 pgoff = vma->vm_pgoff;
1309 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 } else if (vm_flags & VM_SHARED) {
Al Viro835ee792012-03-05 06:39:47 +00001311 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1312 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 error = shmem_zero_setup(vma);
1314 if (error)
1315 goto free_vma;
1316 }
1317
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001318 if (vma_wants_writenotify(vma)) {
1319 pgprot_t pprot = vma->vm_page_prot;
1320
1321 /* Can vma->vm_page_prot have changed??
1322 *
1323 * Answer: Yes, drivers may have changed it in their
1324 * f_op->mmap method.
1325 *
1326 * Ensures that vmas marked as uncached stay that way.
1327 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001328 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001329 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1330 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1331 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001332
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001333 vma_link(mm, vma, prev, rb_link, rb_parent);
1334 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001335
1336 /* Once vma denies write, undo our temporary denial count */
1337 if (correct_wcount)
1338 atomic_inc(&inode->i_writecount);
1339out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001340 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001341
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001342 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001344 if (!mlock_vma_pages_range(vma, addr, addr + len))
1345 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001346 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001347 make_pages_present(addr, addr + len);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301348
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001349 if (file)
1350 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 return addr;
1353
1354unmap_and_free_vma:
1355 if (correct_wcount)
1356 atomic_inc(&inode->i_writecount);
1357 vma->vm_file = NULL;
1358 fput(file);
1359
1360 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001361 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1362 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363free_vma:
1364 kmem_cache_free(vm_area_cachep, vma);
1365unacct_error:
1366 if (charged)
1367 vm_unacct_memory(charged);
1368 return error;
1369}
1370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371/* Get an address range which is currently unmapped.
1372 * For shmat() with addr=0.
1373 *
1374 * Ugly calling convention alert:
1375 * Return value with the low bits set means error value,
1376 * ie
1377 * if (ret & ~PAGE_MASK)
1378 * error = ret;
1379 *
1380 * This function "knows" that -ENOMEM has the bits set.
1381 */
1382#ifndef HAVE_ARCH_UNMAPPED_AREA
1383unsigned long
1384arch_get_unmapped_area(struct file *filp, unsigned long addr,
1385 unsigned long len, unsigned long pgoff, unsigned long flags)
1386{
1387 struct mm_struct *mm = current->mm;
1388 struct vm_area_struct *vma;
1389 unsigned long start_addr;
1390
1391 if (len > TASK_SIZE)
1392 return -ENOMEM;
1393
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001394 if (flags & MAP_FIXED)
1395 return addr;
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 if (addr) {
1398 addr = PAGE_ALIGN(addr);
1399 vma = find_vma(mm, addr);
1400 if (TASK_SIZE - len >= addr &&
1401 (!vma || addr + len <= vma->vm_start))
1402 return addr;
1403 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001404 if (len > mm->cached_hole_size) {
1405 start_addr = addr = mm->free_area_cache;
1406 } else {
1407 start_addr = addr = TASK_UNMAPPED_BASE;
1408 mm->cached_hole_size = 0;
1409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
1411full_search:
1412 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1413 /* At this point: (!vma || addr < vma->vm_end). */
1414 if (TASK_SIZE - len < addr) {
1415 /*
1416 * Start a new search - just in case we missed
1417 * some holes.
1418 */
1419 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001420 addr = TASK_UNMAPPED_BASE;
1421 start_addr = addr;
1422 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 goto full_search;
1424 }
1425 return -ENOMEM;
1426 }
1427 if (!vma || addr + len <= vma->vm_start) {
1428 /*
1429 * Remember the place where we stopped the search:
1430 */
1431 mm->free_area_cache = addr + len;
1432 return addr;
1433 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001434 if (addr + mm->cached_hole_size < vma->vm_start)
1435 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 addr = vma->vm_end;
1437 }
1438}
1439#endif
1440
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001441void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442{
1443 /*
1444 * Is this a new hole at the lowest possible address?
1445 */
Xiao Guangrongf44d2192012-03-21 16:33:56 -07001446 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001447 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448}
1449
1450/*
1451 * This mmap-allocator allocates new areas top-down from below the
1452 * stack's low limit (the base):
1453 */
1454#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1455unsigned long
1456arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1457 const unsigned long len, const unsigned long pgoff,
1458 const unsigned long flags)
1459{
1460 struct vm_area_struct *vma;
1461 struct mm_struct *mm = current->mm;
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001462 unsigned long addr = addr0, start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
1464 /* requested length too big for entire address space */
1465 if (len > TASK_SIZE)
1466 return -ENOMEM;
1467
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001468 if (flags & MAP_FIXED)
1469 return addr;
1470
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 /* requesting a specific address */
1472 if (addr) {
1473 addr = PAGE_ALIGN(addr);
1474 vma = find_vma(mm, addr);
1475 if (TASK_SIZE - len >= addr &&
1476 (!vma || addr + len <= vma->vm_start))
1477 return addr;
1478 }
1479
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001480 /* check if free_area_cache is useful for us */
1481 if (len <= mm->cached_hole_size) {
1482 mm->cached_hole_size = 0;
1483 mm->free_area_cache = mm->mmap_base;
1484 }
1485
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001486try_again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 /* either no address requested or can't fit in requested address hole */
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001488 start_addr = addr = mm->free_area_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001490 if (addr < len)
1491 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001493 addr -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 do {
1495 /*
1496 * Lookup failure means no vma is above this address,
1497 * else if new region fits below vma->vm_start,
1498 * return with success:
1499 */
1500 vma = find_vma(mm, addr);
1501 if (!vma || addr+len <= vma->vm_start)
1502 /* remember the address as a hint for next time */
1503 return (mm->free_area_cache = addr);
1504
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001505 /* remember the largest hole we saw so far */
1506 if (addr + mm->cached_hole_size < vma->vm_start)
1507 mm->cached_hole_size = vma->vm_start - addr;
1508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 /* try just below the current vma->vm_start */
1510 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001511 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001513fail:
1514 /*
1515 * if hint left us with no space for the requested
1516 * mapping then try again:
1517 *
1518 * Note: this is different with the case of bottomup
1519 * which does the fully line-search, but we use find_vma
1520 * here that causes some holes skipped.
1521 */
1522 if (start_addr != mm->mmap_base) {
1523 mm->free_area_cache = mm->mmap_base;
1524 mm->cached_hole_size = 0;
1525 goto try_again;
1526 }
1527
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 /*
1529 * A failed mmap() very likely causes application failure,
1530 * so fall back to the bottom-up function here. This scenario
1531 * can happen with large stack limits and large mmap()
1532 * allocations.
1533 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001534 mm->cached_hole_size = ~0UL;
1535 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1537 /*
1538 * Restore the topdown base:
1539 */
1540 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001541 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
1543 return addr;
1544}
1545#endif
1546
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001547void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548{
1549 /*
1550 * Is this a new hole at the highest possible address?
1551 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001552 if (addr > mm->free_area_cache)
1553 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
1555 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001556 if (mm->free_area_cache > mm->mmap_base)
1557 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558}
1559
1560unsigned long
1561get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1562 unsigned long pgoff, unsigned long flags)
1563{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001564 unsigned long (*get_area)(struct file *, unsigned long,
1565 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Al Viro9206de92009-12-03 15:23:11 -05001567 unsigned long error = arch_mmap_check(addr, len, flags);
1568 if (error)
1569 return error;
1570
1571 /* Careful about overflows.. */
1572 if (len > TASK_SIZE)
1573 return -ENOMEM;
1574
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001575 get_area = current->mm->get_unmapped_area;
1576 if (file && file->f_op && file->f_op->get_unmapped_area)
1577 get_area = file->f_op->get_unmapped_area;
1578 addr = get_area(file, addr, len, pgoff, flags);
1579 if (IS_ERR_VALUE(addr))
1580 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001582 if (addr > TASK_SIZE - len)
1583 return -ENOMEM;
1584 if (addr & ~PAGE_MASK)
1585 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001586
Al Viro9ac4ed42012-05-30 17:13:15 -04001587 addr = arch_rebalance_pgtables(addr, len);
1588 error = security_mmap_addr(addr);
1589 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590}
1591
1592EXPORT_SYMBOL(get_unmapped_area);
1593
1594/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001595struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
1597 struct vm_area_struct *vma = NULL;
1598
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001599 if (WARN_ON_ONCE(!mm)) /* Remove this in linux-3.6 */
1600 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001602 /* Check the cache first. */
1603 /* (Cache hit rate is typically around 35%.) */
1604 vma = mm->mmap_cache;
1605 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1606 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001608 rb_node = mm->mm_rb.rb_node;
1609 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001611 while (rb_node) {
1612 struct vm_area_struct *vma_tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001614 vma_tmp = rb_entry(rb_node,
1615 struct vm_area_struct, vm_rb);
1616
1617 if (vma_tmp->vm_end > addr) {
1618 vma = vma_tmp;
1619 if (vma_tmp->vm_start <= addr)
1620 break;
1621 rb_node = rb_node->rb_left;
1622 } else
1623 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 }
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001625 if (vma)
1626 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 }
1628 return vma;
1629}
1630
1631EXPORT_SYMBOL(find_vma);
1632
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001633/*
1634 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001635 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636struct vm_area_struct *
1637find_vma_prev(struct mm_struct *mm, unsigned long addr,
1638 struct vm_area_struct **pprev)
1639{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001640 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001642 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05001643 if (vma) {
1644 *pprev = vma->vm_prev;
1645 } else {
1646 struct rb_node *rb_node = mm->mm_rb.rb_node;
1647 *pprev = NULL;
1648 while (rb_node) {
1649 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1650 rb_node = rb_node->rb_right;
1651 }
1652 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001653 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654}
1655
1656/*
1657 * Verify that the stack growth is acceptable and
1658 * update accounting. This is shared with both the
1659 * grow-up and grow-down cases.
1660 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001661static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662{
1663 struct mm_struct *mm = vma->vm_mm;
1664 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001665 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001668 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 return -ENOMEM;
1670
1671 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001672 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return -ENOMEM;
1674
1675 /* mlock limit tests */
1676 if (vma->vm_flags & VM_LOCKED) {
1677 unsigned long locked;
1678 unsigned long limit;
1679 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001680 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1681 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 if (locked > limit && !capable(CAP_IPC_LOCK))
1683 return -ENOMEM;
1684 }
1685
Adam Litke0d59a012007-01-30 14:35:39 -08001686 /* Check to ensure the stack will not grow into a hugetlb-only region */
1687 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1688 vma->vm_end - size;
1689 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1690 return -EFAULT;
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 /*
1693 * Overcommit.. This must be the final test, as it will
1694 * update security statistics.
1695 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001696 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 return -ENOMEM;
1698
1699 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 if (vma->vm_flags & VM_LOCKED)
1701 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001702 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 return 0;
1704}
1705
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001706#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001708 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1709 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001711int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712{
1713 int error;
1714
1715 if (!(vma->vm_flags & VM_GROWSUP))
1716 return -EFAULT;
1717
1718 /*
1719 * We must make sure the anon_vma is allocated
1720 * so that the anon_vma locking is not a noop.
1721 */
1722 if (unlikely(anon_vma_prepare(vma)))
1723 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001724 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725
1726 /*
1727 * vma->vm_start/vm_end cannot change under us because the caller
1728 * is required to hold the mmap_sem in read mode. We need the
1729 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001730 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 */
Helge Deller06b32f32006-12-19 19:28:33 +01001732 if (address < PAGE_ALIGN(address+4))
1733 address = PAGE_ALIGN(address+4);
1734 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001735 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001736 return -ENOMEM;
1737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 error = 0;
1739
1740 /* Somebody else might have raced and expanded it already */
1741 if (address > vma->vm_end) {
1742 unsigned long size, grow;
1743
1744 size = address - vma->vm_start;
1745 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1746
Hugh Dickins42c36f62011-05-09 17:44:42 -07001747 error = -ENOMEM;
1748 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1749 error = acct_stack_growth(vma, size, grow);
1750 if (!error) {
1751 vma->vm_end = address;
1752 perf_event_mmap(vma);
1753 }
Eric B Munson3af9e852010-05-18 15:30:49 +01001754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001756 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001757 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 return error;
1759}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001760#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1761
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762/*
1763 * vma is the first one with address < vma->vm_start. Have to extend vma.
1764 */
Michal Hockod05f3162011-05-24 17:11:44 -07001765int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001766 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767{
1768 int error;
1769
1770 /*
1771 * We must make sure the anon_vma is allocated
1772 * so that the anon_vma locking is not a noop.
1773 */
1774 if (unlikely(anon_vma_prepare(vma)))
1775 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001776
1777 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04001778 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05001779 if (error)
1780 return error;
1781
Rik van Rielbb4a3402010-08-09 17:18:37 -07001782 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
1784 /*
1785 * vma->vm_start/vm_end cannot change under us because the caller
1786 * is required to hold the mmap_sem in read mode. We need the
1787 * anon_vma lock to serialize against concurrent expand_stacks.
1788 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790 /* Somebody else might have raced and expanded it already */
1791 if (address < vma->vm_start) {
1792 unsigned long size, grow;
1793
1794 size = vma->vm_end - address;
1795 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1796
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001797 error = -ENOMEM;
1798 if (grow <= vma->vm_pgoff) {
1799 error = acct_stack_growth(vma, size, grow);
1800 if (!error) {
1801 vma->vm_start = address;
1802 vma->vm_pgoff -= grow;
1803 perf_event_mmap(vma);
1804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
1806 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001807 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001808 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 return error;
1810}
1811
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001812#ifdef CONFIG_STACK_GROWSUP
1813int expand_stack(struct vm_area_struct *vma, unsigned long address)
1814{
1815 return expand_upwards(vma, address);
1816}
1817
1818struct vm_area_struct *
1819find_extend_vma(struct mm_struct *mm, unsigned long addr)
1820{
1821 struct vm_area_struct *vma, *prev;
1822
1823 addr &= PAGE_MASK;
1824 vma = find_vma_prev(mm, addr, &prev);
1825 if (vma && (vma->vm_start <= addr))
1826 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001827 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001828 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001829 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001830 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001831 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001832 return prev;
1833}
1834#else
1835int expand_stack(struct vm_area_struct *vma, unsigned long address)
1836{
1837 return expand_downwards(vma, address);
1838}
1839
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840struct vm_area_struct *
1841find_extend_vma(struct mm_struct * mm, unsigned long addr)
1842{
1843 struct vm_area_struct * vma;
1844 unsigned long start;
1845
1846 addr &= PAGE_MASK;
1847 vma = find_vma(mm,addr);
1848 if (!vma)
1849 return NULL;
1850 if (vma->vm_start <= addr)
1851 return vma;
1852 if (!(vma->vm_flags & VM_GROWSDOWN))
1853 return NULL;
1854 start = vma->vm_start;
1855 if (expand_stack(vma, addr))
1856 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001857 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001858 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 return vma;
1861}
1862#endif
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001865 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001867 *
1868 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001870static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001872 unsigned long nr_accounted = 0;
1873
Hugh Dickins365e9c872005-10-29 18:16:18 -07001874 /* Update high watermark before we lower total_vm */
1875 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001877 long nrpages = vma_pages(vma);
1878
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001879 if (vma->vm_flags & VM_ACCOUNT)
1880 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001881 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001882 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001883 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001884 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 validate_mm(mm);
1886}
1887
1888/*
1889 * Get rid of page table information in the indicated region.
1890 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001891 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 */
1893static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001894 struct vm_area_struct *vma, struct vm_area_struct *prev,
1895 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001897 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001898 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
1900 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001901 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001902 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001903 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001904 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1905 next ? next->vm_start : 0);
1906 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907}
1908
1909/*
1910 * Create a list of vma's touched by the unmap, removing them from the mm's
1911 * vma list as we go..
1912 */
1913static void
1914detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1915 struct vm_area_struct *prev, unsigned long end)
1916{
1917 struct vm_area_struct **insertion_point;
1918 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001919 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001922 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 do {
1924 rb_erase(&vma->vm_rb, &mm->mm_rb);
1925 mm->map_count--;
1926 tail_vma = vma;
1927 vma = vma->vm_next;
1928 } while (vma && vma->vm_start < end);
1929 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001930 if (vma)
1931 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001933 if (mm->unmap_area == arch_unmap_area)
1934 addr = prev ? prev->vm_end : mm->mmap_base;
1935 else
1936 addr = vma ? vma->vm_start : mm->mmap_base;
1937 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 mm->mmap_cache = NULL; /* Kill the cache. */
1939}
1940
1941/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001942 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1943 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001945static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 unsigned long addr, int new_below)
1947{
1948 struct mempolicy *pol;
1949 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001950 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
Andi Kleena5516432008-07-23 21:27:41 -07001952 if (is_vm_hugetlb_page(vma) && (addr &
1953 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return -EINVAL;
1955
Christoph Lametere94b1762006-12-06 20:33:17 -08001956 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001958 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 /* most fields are the same, copy all, and then fixup */
1961 *new = *vma;
1962
Rik van Riel5beb4932010-03-05 13:42:07 -08001963 INIT_LIST_HEAD(&new->anon_vma_chain);
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 if (new_below)
1966 new->vm_end = addr;
1967 else {
1968 new->vm_start = addr;
1969 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1970 }
1971
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001972 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001974 err = PTR_ERR(pol);
1975 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 }
1977 vma_set_policy(new, pol);
1978
Rik van Riel5beb4932010-03-05 13:42:07 -08001979 if (anon_vma_clone(new, vma))
1980 goto out_free_mpol;
1981
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07001982 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 get_file(new->vm_file);
1984
1985 if (new->vm_ops && new->vm_ops->open)
1986 new->vm_ops->open(new);
1987
1988 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08001989 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1991 else
Rik van Riel5beb4932010-03-05 13:42:07 -08001992 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Rik van Riel5beb4932010-03-05 13:42:07 -08001994 /* Success. */
1995 if (!err)
1996 return 0;
1997
1998 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04001999 if (new->vm_ops && new->vm_ops->close)
2000 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002001 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002002 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002003 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002004 out_free_mpol:
2005 mpol_put(pol);
2006 out_free_vma:
2007 kmem_cache_free(vm_area_cachep, new);
2008 out_err:
2009 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010}
2011
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002012/*
2013 * Split a vma into two pieces at address 'addr', a new vma is allocated
2014 * either for the first part or the tail.
2015 */
2016int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2017 unsigned long addr, int new_below)
2018{
2019 if (mm->map_count >= sysctl_max_map_count)
2020 return -ENOMEM;
2021
2022 return __split_vma(mm, vma, addr, new_below);
2023}
2024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025/* Munmap is split into 2 main parts -- this part which finds
2026 * what needs doing, and the areas themselves, which do the
2027 * work. This now handles partial unmappings.
2028 * Jeremy Fitzhardinge <jeremy@goop.org>
2029 */
2030int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2031{
2032 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002033 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
2035 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2036 return -EINVAL;
2037
2038 if ((len = PAGE_ALIGN(len)) == 0)
2039 return -EINVAL;
2040
2041 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002042 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002043 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002045 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002046 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
2048 /* if it doesn't overlap, we have nothing.. */
2049 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002050 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 return 0;
2052
2053 /*
2054 * If we need to split any vma, do it now to save pain later.
2055 *
2056 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2057 * unmapped vm_area_struct will remain in use: so lower split_vma
2058 * places tmp vma above, and higher split_vma places tmp vma below.
2059 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002060 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002061 int error;
2062
2063 /*
2064 * Make sure that map_count on return from munmap() will
2065 * not exceed its limit; but let map_count go just above
2066 * its limit temporarily, to help free resources as expected.
2067 */
2068 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2069 return -ENOMEM;
2070
2071 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 if (error)
2073 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002074 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 }
2076
2077 /* Does it split the last one? */
2078 last = find_vma(mm, end);
2079 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002080 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 if (error)
2082 return error;
2083 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002084 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
2086 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002087 * unlock any mlock()ed ranges before detaching vmas
2088 */
2089 if (mm->locked_vm) {
2090 struct vm_area_struct *tmp = vma;
2091 while (tmp && tmp->vm_start < end) {
2092 if (tmp->vm_flags & VM_LOCKED) {
2093 mm->locked_vm -= vma_pages(tmp);
2094 munlock_vma_pages_all(tmp);
2095 }
2096 tmp = tmp->vm_next;
2097 }
2098 }
2099
2100 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 * Remove the vma's, and unmap the actual pages
2102 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002103 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2104 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002107 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
2109 return 0;
2110}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Al Virobfce2812012-04-20 21:57:04 -04002112int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002113{
2114 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002115 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002116
2117 down_write(&mm->mmap_sem);
2118 ret = do_munmap(mm, start, len);
2119 up_write(&mm->mmap_sem);
2120 return ret;
2121}
2122EXPORT_SYMBOL(vm_munmap);
2123
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002124SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002127 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128}
2129
2130static inline void verify_mm_writelocked(struct mm_struct *mm)
2131{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002132#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2134 WARN_ON(1);
2135 up_read(&mm->mmap_sem);
2136 }
2137#endif
2138}
2139
2140/*
2141 * this is really a simplified "do_mmap". it only handles
2142 * anonymous maps. eventually we may be able to do some
2143 * brk-specific accounting here.
2144 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002145static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
2147 struct mm_struct * mm = current->mm;
2148 struct vm_area_struct * vma, * prev;
2149 unsigned long flags;
2150 struct rb_node ** rb_link, * rb_parent;
2151 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002152 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
2154 len = PAGE_ALIGN(len);
2155 if (!len)
2156 return addr;
2157
Kirill Korotaev3a459752006-09-07 14:17:04 +04002158 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2159
Al Viro2c6a1012009-12-03 19:40:46 -05002160 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2161 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002162 return error;
2163
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 /*
2165 * mlock MCL_FUTURE?
2166 */
2167 if (mm->def_flags & VM_LOCKED) {
2168 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002169 locked = len >> PAGE_SHIFT;
2170 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002171 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002172 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2174 return -EAGAIN;
2175 }
2176
2177 /*
2178 * mm->mmap_sem is required to protect against another thread
2179 * changing the mappings in case we sleep.
2180 */
2181 verify_mm_writelocked(mm);
2182
2183 /*
2184 * Clear old maps. this also does some error checking for us
2185 */
2186 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002187 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 if (do_munmap(mm, addr, len))
2189 return -ENOMEM;
2190 goto munmap_back;
2191 }
2192
2193 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002194 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 return -ENOMEM;
2196
2197 if (mm->map_count > sysctl_max_map_count)
2198 return -ENOMEM;
2199
Al Viro191c5422012-02-13 03:58:52 +00002200 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 return -ENOMEM;
2202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002204 vma = vma_merge(mm, prev, addr, addr + len, flags,
2205 NULL, NULL, pgoff, NULL);
2206 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 goto out;
2208
2209 /*
2210 * create a vma struct for an anonymous mapping
2211 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002212 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 if (!vma) {
2214 vm_unacct_memory(len >> PAGE_SHIFT);
2215 return -ENOMEM;
2216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
Rik van Riel5beb4932010-03-05 13:42:07 -08002218 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 vma->vm_mm = mm;
2220 vma->vm_start = addr;
2221 vma->vm_end = addr + len;
2222 vma->vm_pgoff = pgoff;
2223 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002224 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 vma_link(mm, vma, prev, rb_link, rb_parent);
2226out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002227 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 mm->total_vm += len >> PAGE_SHIFT;
2229 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002230 if (!mlock_vma_pages_range(vma, addr, addr + len))
2231 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 }
2233 return addr;
2234}
2235
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002236unsigned long vm_brk(unsigned long addr, unsigned long len)
2237{
2238 struct mm_struct *mm = current->mm;
2239 unsigned long ret;
2240
2241 down_write(&mm->mmap_sem);
2242 ret = do_brk(addr, len);
2243 up_write(&mm->mmap_sem);
2244 return ret;
2245}
2246EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248/* Release all mmaps. */
2249void exit_mmap(struct mm_struct *mm)
2250{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002251 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002252 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 unsigned long nr_accounted = 0;
2254
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002255 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002256 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002257
Rik van Rielba470de2008-10-18 20:26:50 -07002258 if (mm->locked_vm) {
2259 vma = mm->mmap;
2260 while (vma) {
2261 if (vma->vm_flags & VM_LOCKED)
2262 munlock_vma_pages_all(vma);
2263 vma = vma->vm_next;
2264 }
2265 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002266
2267 arch_exit_mmap(mm);
2268
Rik van Rielba470de2008-10-18 20:26:50 -07002269 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002270 if (!vma) /* Can happen if dup_mmap() received an OOM */
2271 return;
2272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 flush_cache_mm(mm);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002275 tlb_gather_mmu(&tlb, mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002276 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002277 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002278 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002279
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002280 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Al Viro853f5e22012-03-05 14:03:47 -05002281 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002284 * Walk the list again, actually closing and freeing it,
2285 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002287 while (vma) {
2288 if (vma->vm_flags & VM_ACCOUNT)
2289 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002290 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002291 }
2292 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002293
Hugh Dickinsf9aed622012-08-21 16:15:45 -07002294 WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295}
2296
2297/* Insert vm structure into process list sorted by address
2298 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002299 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002301int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302{
Hugh Dickins6597d782012-10-08 16:29:07 -07002303 struct vm_area_struct *prev;
2304 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305
2306 /*
2307 * The vm_pgoff of a purely anonymous vma should be irrelevant
2308 * until its first write fault, when page's anon_vma and index
2309 * are set. But now set the vm_pgoff it will almost certainly
2310 * end up with (unless mremap moves it elsewhere before that
2311 * first wfault), so /proc/pid/maps tells a consistent story.
2312 *
2313 * By setting it to reflect the virtual start address of the
2314 * vma, merges and splits can happen in a seamless way, just
2315 * using the existing file pgoff checks and manipulations.
2316 * Similarly in do_mmap_pgoff and in do_brk.
2317 */
2318 if (!vma->vm_file) {
2319 BUG_ON(vma->anon_vma);
2320 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2321 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002322 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2323 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002325 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002326 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002327 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302328
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 vma_link(mm, vma, prev, rb_link, rb_parent);
2330 return 0;
2331}
2332
2333/*
2334 * Copy the vma structure to a new location in the same mm,
2335 * prior to moving page table entries, to effect an mremap move.
2336 */
2337struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2338 unsigned long addr, unsigned long len, pgoff_t pgoff)
2339{
2340 struct vm_area_struct *vma = *vmap;
2341 unsigned long vma_start = vma->vm_start;
2342 struct mm_struct *mm = vma->vm_mm;
2343 struct vm_area_struct *new_vma, *prev;
2344 struct rb_node **rb_link, *rb_parent;
2345 struct mempolicy *pol;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002346 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
2348 /*
2349 * If anonymous vma has not yet been faulted, update new pgoff
2350 * to match new location, to increase its chance of merging.
2351 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002352 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002354 faulted_in_anon_vma = false;
2355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
Hugh Dickins6597d782012-10-08 16:29:07 -07002357 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2358 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2360 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2361 if (new_vma) {
2362 /*
2363 * Source vma may have been merged into new_vma
2364 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002365 if (unlikely(vma_start >= new_vma->vm_start &&
2366 vma_start < new_vma->vm_end)) {
2367 /*
2368 * The only way we can get a vma_merge with
2369 * self during an mremap is if the vma hasn't
2370 * been faulted in yet and we were allowed to
2371 * reset the dst vma->vm_pgoff to the
2372 * destination address of the mremap to allow
2373 * the merge to happen. mremap must change the
2374 * vm_pgoff linearity between src and dst vmas
2375 * (in turn preventing a vma_merge) to be
2376 * safe. It is only safe to keep the vm_pgoff
2377 * linear if there are no pages mapped yet.
2378 */
2379 VM_BUG_ON(faulted_in_anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 *vmap = new_vma;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002381 } else
2382 anon_vma_moveto_tail(new_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002384 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 if (new_vma) {
2386 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002387 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002388 if (IS_ERR(pol))
2389 goto out_free_vma;
2390 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2391 if (anon_vma_clone(new_vma, vma))
2392 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 vma_set_policy(new_vma, pol);
2394 new_vma->vm_start = addr;
2395 new_vma->vm_end = addr + len;
2396 new_vma->vm_pgoff = pgoff;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002397 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 get_file(new_vma->vm_file);
2399 if (new_vma->vm_ops && new_vma->vm_ops->open)
2400 new_vma->vm_ops->open(new_vma);
2401 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2402 }
2403 }
2404 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002405
2406 out_free_mempol:
2407 mpol_put(pol);
2408 out_free_vma:
2409 kmem_cache_free(vm_area_cachep, new_vma);
2410 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002412
2413/*
2414 * Return true if the calling process may expand its vm space by the passed
2415 * number of pages
2416 */
2417int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2418{
2419 unsigned long cur = mm->total_vm; /* pages */
2420 unsigned long lim;
2421
Jiri Slaby59e99e52010-03-05 13:41:44 -08002422 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002423
2424 if (cur + npages > lim)
2425 return 0;
2426 return 1;
2427}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002428
2429
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002430static int special_mapping_fault(struct vm_area_struct *vma,
2431 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002432{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002433 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002434 struct page **pages;
2435
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002436 /*
2437 * special mappings have no vm_file, and in that case, the mm
2438 * uses vm_pgoff internally. So we have to subtract it from here.
2439 * We are allowed to do this because we are the mm; do not copy
2440 * this code into drivers!
2441 */
2442 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002443
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002444 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2445 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002446
2447 if (*pages) {
2448 struct page *page = *pages;
2449 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002450 vmf->page = page;
2451 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002452 }
2453
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002454 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002455}
2456
2457/*
2458 * Having a close hook prevents vma merging regardless of flags.
2459 */
2460static void special_mapping_close(struct vm_area_struct *vma)
2461{
2462}
2463
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002464static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002465 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002466 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002467};
2468
2469/*
2470 * Called with mm->mmap_sem held for writing.
2471 * Insert a new vma covering the given region, with the given flags.
2472 * Its pages are supplied by the given array of struct page *.
2473 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2474 * The region past the last page supplied will always produce SIGBUS.
2475 * The array pointer and the pages it points to are assumed to stay alive
2476 * for as long as this mapping might exist.
2477 */
2478int install_special_mapping(struct mm_struct *mm,
2479 unsigned long addr, unsigned long len,
2480 unsigned long vm_flags, struct page **pages)
2481{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002482 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002483 struct vm_area_struct *vma;
2484
2485 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2486 if (unlikely(vma == NULL))
2487 return -ENOMEM;
2488
Rik van Riel5beb4932010-03-05 13:42:07 -08002489 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002490 vma->vm_mm = mm;
2491 vma->vm_start = addr;
2492 vma->vm_end = addr + len;
2493
Nick Piggin2f987352008-02-02 03:08:53 +01002494 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002495 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002496
2497 vma->vm_ops = &special_mapping_vmops;
2498 vma->vm_private_data = pages;
2499
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002500 ret = insert_vm_struct(mm, vma);
2501 if (ret)
2502 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002503
2504 mm->total_vm += len >> PAGE_SHIFT;
2505
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002506 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002507
Roland McGrathfa5dc222007-02-08 14:20:41 -08002508 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002509
2510out:
2511 kmem_cache_free(vm_area_cachep, vma);
2512 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002513}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002514
2515static DEFINE_MUTEX(mm_all_locks_mutex);
2516
Peter Zijlstra454ed842008-08-11 09:30:25 +02002517static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002518{
Rik van Riel012f18002010-08-09 17:18:40 -07002519 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002520 /*
2521 * The LSB of head.next can't change from under us
2522 * because we hold the mm_all_locks_mutex.
2523 */
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002524 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002525 /*
2526 * We can safely modify head.next after taking the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002527 * anon_vma->root->mutex. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002528 * the same anon_vma we won't take it again.
2529 *
2530 * No need of atomic instructions here, head.next
2531 * can't change from under us thanks to the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002532 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002533 */
2534 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002535 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002536 BUG();
2537 }
2538}
2539
Peter Zijlstra454ed842008-08-11 09:30:25 +02002540static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002541{
2542 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2543 /*
2544 * AS_MM_ALL_LOCKS can't change from under us because
2545 * we hold the mm_all_locks_mutex.
2546 *
2547 * Operations on ->flags have to be atomic because
2548 * even if AS_MM_ALL_LOCKS is stable thanks to the
2549 * mm_all_locks_mutex, there may be other cpus
2550 * changing other bitflags in parallel to us.
2551 */
2552 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2553 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002554 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002555 }
2556}
2557
2558/*
2559 * This operation locks against the VM for all pte/vma/mm related
2560 * operations that could ever happen on a certain mm. This includes
2561 * vmtruncate, try_to_unmap, and all page faults.
2562 *
2563 * The caller must take the mmap_sem in write mode before calling
2564 * mm_take_all_locks(). The caller isn't allowed to release the
2565 * mmap_sem until mm_drop_all_locks() returns.
2566 *
2567 * mmap_sem in write mode is required in order to block all operations
2568 * that could modify pagetables and free pages without need of
2569 * altering the vma layout (for example populate_range() with
2570 * nonlinear vmas). It's also needed in write mode to avoid new
2571 * anon_vmas to be associated with existing vmas.
2572 *
2573 * A single task can't take more than one mm_take_all_locks() in a row
2574 * or it would deadlock.
2575 *
2576 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2577 * mapping->flags avoid to take the same lock twice, if more than one
2578 * vma in this mm is backed by the same anon_vma or address_space.
2579 *
2580 * We can take all the locks in random order because the VM code
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002581 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002582 * takes more than one of them in a row. Secondly we're protected
2583 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2584 *
2585 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2586 * that may have to take thousand of locks.
2587 *
2588 * mm_take_all_locks() can fail if it's interrupted by signals.
2589 */
2590int mm_take_all_locks(struct mm_struct *mm)
2591{
2592 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002593 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002594
2595 BUG_ON(down_read_trylock(&mm->mmap_sem));
2596
2597 mutex_lock(&mm_all_locks_mutex);
2598
2599 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2600 if (signal_pending(current))
2601 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002602 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002603 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002604 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002605
2606 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2607 if (signal_pending(current))
2608 goto out_unlock;
2609 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002610 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2611 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002612 }
2613
Kautuk Consul584cff52011-10-31 17:08:59 -07002614 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002615
2616out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07002617 mm_drop_all_locks(mm);
2618 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002619}
2620
2621static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2622{
Rik van Riel012f18002010-08-09 17:18:40 -07002623 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002624 /*
2625 * The LSB of head.next can't change to 0 from under
2626 * us because we hold the mm_all_locks_mutex.
2627 *
2628 * We must however clear the bitflag before unlocking
2629 * the vma so the users using the anon_vma->head will
2630 * never see our bitflag.
2631 *
2632 * No need of atomic instructions here, head.next
2633 * can't change from under us until we release the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002634 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002635 */
2636 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f18002010-08-09 17:18:40 -07002637 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002638 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002639 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002640 }
2641}
2642
2643static void vm_unlock_mapping(struct address_space *mapping)
2644{
2645 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2646 /*
2647 * AS_MM_ALL_LOCKS can't change to 0 from under us
2648 * because we hold the mm_all_locks_mutex.
2649 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002650 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002651 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2652 &mapping->flags))
2653 BUG();
2654 }
2655}
2656
2657/*
2658 * The mmap_sem cannot be released by the caller until
2659 * mm_drop_all_locks() returns.
2660 */
2661void mm_drop_all_locks(struct mm_struct *mm)
2662{
2663 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002664 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002665
2666 BUG_ON(down_read_trylock(&mm->mmap_sem));
2667 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2668
2669 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2670 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002671 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2672 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002673 if (vma->vm_file && vma->vm_file->f_mapping)
2674 vm_unlock_mapping(vma->vm_file->f_mapping);
2675 }
2676
2677 mutex_unlock(&mm_all_locks_mutex);
2678}
David Howells8feae132009-01-08 12:04:47 +00002679
2680/*
2681 * initialise the VMA slab
2682 */
2683void __init mmap_init(void)
2684{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002685 int ret;
2686
2687 ret = percpu_counter_init(&vm_committed_as, 0);
2688 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002689}