blob: 2d942353d681a8b4f08155eebdcfb20b088093e7 [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/* description of effects of mapping type and prot in current implementation.
55 * this is due to the limited x86 page protection hardware. The expected
56 * behavior is in parens:
57 *
58 * map_type prot
59 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
60 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
61 * w: (no) no w: (no) no w: (yes) yes w: (no) no
62 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
63 *
64 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
65 * w: (no) no w: (no) no w: (copy) copy w: (no) no
66 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
67 *
68 */
69pgprot_t protection_map[16] = {
70 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
71 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
72};
73
Hugh Dickins804af2c2006-07-26 21:39:49 +010074pgprot_t vm_get_page_prot(unsigned long vm_flags)
75{
Dave Kleikampb845f312008-07-08 00:28:51 +100076 return __pgprot(pgprot_val(protection_map[vm_flags &
77 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
78 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010079}
80EXPORT_SYMBOL(vm_get_page_prot);
81
Shaohua Li34679d72011-05-24 17:11:18 -070082int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
83int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070084int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Shaohua Li34679d72011-05-24 17:11:18 -070085/*
86 * Make sure vm_committed_as in one cacheline and not cacheline shared with
87 * other variables. It can be updated by several CPUs frequently.
88 */
89struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91/*
92 * Check that a process has enough memory to allocate a new virtual
93 * mapping. 0 means there is enough memory for the allocation to
94 * succeed and -ENOMEM implies there is not.
95 *
96 * We currently support three overcommit policies, which are set via the
97 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
98 *
99 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
100 * Additional code 2002 Jul 20 by Robert Love.
101 *
102 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
103 *
104 * Note this is a helper function intended to be used by LSMs which
105 * wish to use this logic.
106 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700107int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
109 unsigned long free, allowed;
110
111 vm_acct_memory(pages);
112
113 /*
114 * Sometimes we want to use more memory than we have
115 */
116 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
117 return 0;
118
119 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700120 free = global_page_state(NR_FREE_PAGES);
121 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Dmitry Finkc15bef32011-07-25 17:12:19 -0700123 /*
124 * shmem pages shouldn't be counted as free in this
125 * case, they can't be purged, only swapped out, and
126 * that won't affect the overall amount of available
127 * memory in the system.
128 */
129 free -= global_page_state(NR_SHMEM);
130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 free += nr_swap_pages;
132
133 /*
134 * Any slabs which are created with the
135 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
136 * which are reclaimable, under pressure. The dentry
137 * cache and most inode caches should fall into this
138 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700139 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
141 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700142 * Leave reserved pages. The pages are not for anonymous pages.
143 */
144 if (free <= totalreserve_pages)
145 goto error;
146 else
147 free -= totalreserve_pages;
148
149 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 * Leave the last 3% for root
151 */
152 if (!cap_sys_admin)
153 free -= free / 32;
154
155 if (free > pages)
156 return 0;
157
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700158 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 }
160
161 allowed = (totalram_pages - hugetlb_total_pages())
162 * sysctl_overcommit_ratio / 100;
163 /*
164 * Leave the last 3% for root
165 */
166 if (!cap_sys_admin)
167 allowed -= allowed / 32;
168 allowed += total_swap_pages;
169
170 /* Don't let a single process grow too big:
171 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700172 if (mm)
173 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700175 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700177error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 vm_unacct_memory(pages);
179
180 return -ENOMEM;
181}
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700184 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 */
186static void __remove_shared_vm_struct(struct vm_area_struct *vma,
187 struct file *file, struct address_space *mapping)
188{
189 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800190 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 if (vma->vm_flags & VM_SHARED)
192 mapping->i_mmap_writable--;
193
194 flush_dcache_mmap_lock(mapping);
195 if (unlikely(vma->vm_flags & VM_NONLINEAR))
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700196 list_del_init(&vma->shared.nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700198 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 flush_dcache_mmap_unlock(mapping);
200}
201
202/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700203 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700204 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700206void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 struct file *file = vma->vm_file;
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 if (file) {
211 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700212 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700214 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700216}
217
218/*
219 * Close a vm structure and free it, returning the next.
220 */
221static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
222{
223 struct vm_area_struct *next = vma->vm_next;
224
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700225 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 if (vma->vm_ops && vma->vm_ops->close)
227 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700228 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700229 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700230 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700232 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700235static unsigned long do_brk(unsigned long addr, unsigned long len);
236
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100237SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
239 unsigned long rlim, retval;
240 unsigned long newbrk, oldbrk;
241 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700242 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244 down_write(&mm->mmap_sem);
245
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700246#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800247 /*
248 * CONFIG_COMPAT_BRK can still be overridden by setting
249 * randomize_va_space to 2, which will still cause mm->start_brk
250 * to be arbitrarily shifted
251 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700252 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800253 min_brk = mm->start_brk;
254 else
255 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700256#else
257 min_brk = mm->start_brk;
258#endif
259 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700261
262 /*
263 * Check against rlimit here. If this check is done later after the test
264 * of oldbrk with newbrk then it can escape the test and let the data
265 * segment grow beyond its set limit the in case where the limit is
266 * not page aligned -Ram Gupta
267 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800268 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100269 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
270 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700271 goto out;
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 newbrk = PAGE_ALIGN(brk);
274 oldbrk = PAGE_ALIGN(mm->brk);
275 if (oldbrk == newbrk)
276 goto set_brk;
277
278 /* Always allow shrinking brk. */
279 if (brk <= mm->brk) {
280 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
281 goto set_brk;
282 goto out;
283 }
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 /* Check against existing mmap mappings. */
286 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
287 goto out;
288
289 /* Ok, looks good - let it rip. */
290 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
291 goto out;
292set_brk:
293 mm->brk = brk;
294out:
295 retval = mm->brk;
296 up_write(&mm->mmap_sem);
297 return retval;
298}
299
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700300#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301static int browse_rb(struct rb_root *root)
302{
303 int i = 0, j;
304 struct rb_node *nd, *pn = NULL;
305 unsigned long prev = 0, pend = 0;
306
307 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
308 struct vm_area_struct *vma;
309 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
310 if (vma->vm_start < prev)
311 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
312 if (vma->vm_start < pend)
313 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
314 if (vma->vm_start > vma->vm_end)
315 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
316 i++;
317 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800318 prev = vma->vm_start;
319 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 }
321 j = 0;
322 for (nd = pn; nd; nd = rb_prev(nd)) {
323 j++;
324 }
325 if (i != j)
326 printk("backwards %d, forwards %d\n", j, i), i = 0;
327 return i;
328}
329
330void validate_mm(struct mm_struct *mm)
331{
332 int bug = 0;
333 int i = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700334 struct vm_area_struct *vma = mm->mmap;
335 while (vma) {
336 struct anon_vma_chain *avc;
337 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
338 anon_vma_interval_tree_verify(avc);
339 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 i++;
341 }
342 if (i != mm->map_count)
343 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
344 i = browse_rb(&mm->mm_rb);
345 if (i != mm->map_count)
346 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200347 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349#else
350#define validate_mm(mm) do { } while (0)
351#endif
352
Michel Lespinassebf181b92012-10-08 16:31:39 -0700353/*
354 * vma has some anon_vma assigned, and is already inserted on that
355 * anon_vma's interval trees.
356 *
357 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
358 * vma must be removed from the anon_vma's interval trees using
359 * anon_vma_interval_tree_pre_update_vma().
360 *
361 * After the update, the vma will be reinserted using
362 * anon_vma_interval_tree_post_update_vma().
363 *
364 * The entire update must be protected by exclusive mmap_sem and by
365 * the root anon_vma's mutex.
366 */
367static inline void
368anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
369{
370 struct anon_vma_chain *avc;
371
372 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
373 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
374}
375
376static inline void
377anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
378{
379 struct anon_vma_chain *avc;
380
381 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
382 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
383}
384
Hugh Dickins6597d782012-10-08 16:29:07 -0700385static int find_vma_links(struct mm_struct *mm, unsigned long addr,
386 unsigned long end, struct vm_area_struct **pprev,
387 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Hugh Dickins6597d782012-10-08 16:29:07 -0700389 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
391 __rb_link = &mm->mm_rb.rb_node;
392 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 while (*__rb_link) {
395 struct vm_area_struct *vma_tmp;
396
397 __rb_parent = *__rb_link;
398 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
399
400 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700401 /* Fail if an existing vma overlaps the area */
402 if (vma_tmp->vm_start < end)
403 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 __rb_link = &__rb_parent->rb_left;
405 } else {
406 rb_prev = __rb_parent;
407 __rb_link = &__rb_parent->rb_right;
408 }
409 }
410
411 *pprev = NULL;
412 if (rb_prev)
413 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
414 *rb_link = __rb_link;
415 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700416 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417}
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
420 struct rb_node **rb_link, struct rb_node *rb_parent)
421{
422 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
423 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
424}
425
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700426static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
ZhenwenXu48aae422009-01-06 14:40:21 -0800428 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430 file = vma->vm_file;
431 if (file) {
432 struct address_space *mapping = file->f_mapping;
433
434 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800435 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 if (vma->vm_flags & VM_SHARED)
437 mapping->i_mmap_writable++;
438
439 flush_dcache_mmap_lock(mapping);
440 if (unlikely(vma->vm_flags & VM_NONLINEAR))
441 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
442 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700443 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 flush_dcache_mmap_unlock(mapping);
445 }
446}
447
448static void
449__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
450 struct vm_area_struct *prev, struct rb_node **rb_link,
451 struct rb_node *rb_parent)
452{
453 __vma_link_list(mm, vma, prev, rb_parent);
454 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
457static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
458 struct vm_area_struct *prev, struct rb_node **rb_link,
459 struct rb_node *rb_parent)
460{
461 struct address_space *mapping = NULL;
462
463 if (vma->vm_file)
464 mapping = vma->vm_file->f_mapping;
465
Peter Zijlstra97a89412011-05-24 17:12:04 -0700466 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700467 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 __vma_link(mm, vma, prev, rb_link, rb_parent);
470 __vma_link_file(vma);
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700473 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475 mm->map_count++;
476 validate_mm(mm);
477}
478
479/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700480 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700481 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800483static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
Hugh Dickins6597d782012-10-08 16:29:07 -0700485 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800486 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Hugh Dickins6597d782012-10-08 16:29:07 -0700488 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
489 &prev, &rb_link, &rb_parent))
490 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 __vma_link(mm, vma, prev, rb_link, rb_parent);
492 mm->map_count++;
493}
494
495static inline void
496__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
497 struct vm_area_struct *prev)
498{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700499 struct vm_area_struct *next = vma->vm_next;
500
501 prev->vm_next = next;
502 if (next)
503 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 rb_erase(&vma->vm_rb, &mm->mm_rb);
505 if (mm->mmap_cache == vma)
506 mm->mmap_cache = prev;
507}
508
509/*
510 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
511 * is already present in an i_mmap tree without adjusting the tree.
512 * The following helper function should be used when such adjustments
513 * are necessary. The "insert" vma (if any) is to be inserted
514 * before we drop the necessary locks.
515 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800516int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
518{
519 struct mm_struct *mm = vma->vm_mm;
520 struct vm_area_struct *next = vma->vm_next;
521 struct vm_area_struct *importer = NULL;
522 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700523 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700524 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 long adjust_next = 0;
527 int remove_next = 0;
528
529 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700530 struct vm_area_struct *exporter = NULL;
531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 if (end >= next->vm_end) {
533 /*
534 * vma expands, overlapping all the next, and
535 * perhaps the one after too (mprotect case 6).
536 */
537again: remove_next = 1 + (end > next->vm_end);
538 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700539 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 importer = vma;
541 } else if (end > next->vm_start) {
542 /*
543 * vma expands, overlapping part of the next:
544 * mprotect case 5 shifting the boundary up.
545 */
546 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700547 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 importer = vma;
549 } else if (end < vma->vm_end) {
550 /*
551 * vma shrinks, and !insert tells it's not
552 * split_vma inserting another: so it must be
553 * mprotect case 4 shifting the boundary down.
554 */
555 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700556 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 importer = next;
558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Rik van Riel5beb4932010-03-05 13:42:07 -0800560 /*
561 * Easily overlooked: when mprotect shifts the boundary,
562 * make sure the expanding vma has anon_vma set if the
563 * shrinking vma had, to cover any anon pages imported.
564 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700565 if (exporter && exporter->anon_vma && !importer->anon_vma) {
566 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800567 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700568 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800569 }
570 }
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (file) {
573 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530574 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530576 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530577
578 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530579 uprobe_munmap(next, next->vm_start,
580 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530581 }
582
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700583 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700586 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 * are visible to arm/parisc __flush_dcache_page
588 * throughout; but we cannot insert into address
589 * space until vma start or end is updated.
590 */
591 __vma_link_file(insert);
592 }
593 }
594
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800595 vma_adjust_trans_huge(vma, start, end, adjust_next);
596
Michel Lespinassebf181b92012-10-08 16:31:39 -0700597 anon_vma = vma->anon_vma;
598 if (!anon_vma && adjust_next)
599 anon_vma = next->anon_vma;
600 if (anon_vma) {
Michel Lespinasseca42b262012-10-08 16:30:01 -0700601 VM_BUG_ON(adjust_next && next->anon_vma &&
602 anon_vma != next->anon_vma);
Rik van Riel012f18002010-08-09 17:18:40 -0700603 anon_vma_lock(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700604 anon_vma_interval_tree_pre_update_vma(vma);
605 if (adjust_next)
606 anon_vma_interval_tree_pre_update_vma(next);
607 }
Rik van Riel012f18002010-08-09 17:18:40 -0700608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 if (root) {
610 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700611 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700613 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 }
615
616 vma->vm_start = start;
617 vma->vm_end = end;
618 vma->vm_pgoff = pgoff;
619 if (adjust_next) {
620 next->vm_start += adjust_next << PAGE_SHIFT;
621 next->vm_pgoff += adjust_next;
622 }
623
624 if (root) {
625 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700626 vma_interval_tree_insert(next, root);
627 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 flush_dcache_mmap_unlock(mapping);
629 }
630
631 if (remove_next) {
632 /*
633 * vma_merge has merged next into vma, and needs
634 * us to remove next before dropping the locks.
635 */
636 __vma_unlink(mm, next, vma);
637 if (file)
638 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 } else if (insert) {
640 /*
641 * split_vma has split insert from vma, and needs
642 * us to insert it before dropping the locks
643 * (it may either follow vma or precede it).
644 */
645 __insert_vm_struct(mm, insert);
646 }
647
Michel Lespinassebf181b92012-10-08 16:31:39 -0700648 if (anon_vma) {
649 anon_vma_interval_tree_post_update_vma(vma);
650 if (adjust_next)
651 anon_vma_interval_tree_post_update_vma(next);
Rik van Riel012f18002010-08-09 17:18:40 -0700652 anon_vma_unlock(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700655 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530657 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100658 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530659
660 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100661 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530662 }
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700665 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530666 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700668 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800669 if (next->anon_vma)
670 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700672 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 kmem_cache_free(vm_area_cachep, next);
674 /*
675 * In mprotect's case 6 (see comments on vma_merge),
676 * we must remove another next too. It would clutter
677 * up the code too much to do both in one go.
678 */
679 if (remove_next == 2) {
680 next = vma->vm_next;
681 goto again;
682 }
683 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530684 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100685 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
687 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800688
689 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
692/*
693 * If the vma has a ->close operation then the driver probably needs to release
694 * per-vma resources, so we don't attempt to merge those.
695 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696static inline int is_mergeable_vma(struct vm_area_struct *vma,
697 struct file *file, unsigned long vm_flags)
698{
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700699 if (vma->vm_flags ^ vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return 0;
701 if (vma->vm_file != file)
702 return 0;
703 if (vma->vm_ops && vma->vm_ops->close)
704 return 0;
705 return 1;
706}
707
708static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700709 struct anon_vma *anon_vma2,
710 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
Shaohua Li965f55d2011-05-24 17:11:20 -0700712 /*
713 * The list_is_singular() test is to avoid merging VMA cloned from
714 * parents. This can improve scalability caused by anon_vma lock.
715 */
716 if ((!anon_vma1 || !anon_vma2) && (!vma ||
717 list_is_singular(&vma->anon_vma_chain)))
718 return 1;
719 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720}
721
722/*
723 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
724 * in front of (at a lower virtual address and file offset than) the vma.
725 *
726 * We cannot merge two vmas if they have differently assigned (non-NULL)
727 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
728 *
729 * We don't check here for the merged mmap wrapping around the end of pagecache
730 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
731 * wrap, nor mmaps which cover the final page at index -1UL.
732 */
733static int
734can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
735 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
736{
737 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700738 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (vma->vm_pgoff == vm_pgoff)
740 return 1;
741 }
742 return 0;
743}
744
745/*
746 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
747 * beyond (at a higher virtual address and file offset than) the vma.
748 *
749 * We cannot merge two vmas if they have differently assigned (non-NULL)
750 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
751 */
752static int
753can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
754 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
755{
756 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700757 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 pgoff_t vm_pglen;
759 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
760 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
761 return 1;
762 }
763 return 0;
764}
765
766/*
767 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
768 * whether that can be merged with its predecessor or its successor.
769 * Or both (it neatly fills a hole).
770 *
771 * In most cases - when called for mmap, brk or mremap - [addr,end) is
772 * certain not to be mapped by the time vma_merge is called; but when
773 * called for mprotect, it is certain to be already mapped (either at
774 * an offset within prev, or at the start of next), and the flags of
775 * this area are about to be changed to vm_flags - and the no-change
776 * case has already been eliminated.
777 *
778 * The following mprotect cases have to be considered, where AAAA is
779 * the area passed down from mprotect_fixup, never extending beyond one
780 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
781 *
782 * AAAA AAAA AAAA AAAA
783 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
784 * cannot merge might become might become might become
785 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
786 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
787 * mremap move: PPPPNNNNNNNN 8
788 * AAAA
789 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
790 * might become case 1 below case 2 below case 3 below
791 *
792 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
793 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
794 */
795struct vm_area_struct *vma_merge(struct mm_struct *mm,
796 struct vm_area_struct *prev, unsigned long addr,
797 unsigned long end, unsigned long vm_flags,
798 struct anon_vma *anon_vma, struct file *file,
799 pgoff_t pgoff, struct mempolicy *policy)
800{
801 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
802 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800803 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
805 /*
806 * We later require that vma->vm_flags == vm_flags,
807 * so this tests vma->vm_flags & VM_SPECIAL, too.
808 */
809 if (vm_flags & VM_SPECIAL)
810 return NULL;
811
812 if (prev)
813 next = prev->vm_next;
814 else
815 next = mm->mmap;
816 area = next;
817 if (next && next->vm_end == end) /* cases 6, 7, 8 */
818 next = next->vm_next;
819
820 /*
821 * Can it merge with the predecessor?
822 */
823 if (prev && prev->vm_end == addr &&
824 mpol_equal(vma_policy(prev), policy) &&
825 can_vma_merge_after(prev, vm_flags,
826 anon_vma, file, pgoff)) {
827 /*
828 * OK, it can. Can we now merge in the successor as well?
829 */
830 if (next && end == next->vm_start &&
831 mpol_equal(policy, vma_policy(next)) &&
832 can_vma_merge_before(next, vm_flags,
833 anon_vma, file, pgoff+pglen) &&
834 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700835 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800837 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 next->vm_end, prev->vm_pgoff, NULL);
839 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800840 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800842 if (err)
843 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800844 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 return prev;
846 }
847
848 /*
849 * Can this new request be merged in front of next?
850 */
851 if (next && end == next->vm_start &&
852 mpol_equal(policy, vma_policy(next)) &&
853 can_vma_merge_before(next, vm_flags,
854 anon_vma, file, pgoff+pglen)) {
855 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800856 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 addr, prev->vm_pgoff, NULL);
858 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800859 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800861 if (err)
862 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800863 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 return area;
865 }
866
867 return NULL;
868}
869
870/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700871 * Rough compatbility check to quickly see if it's even worth looking
872 * at sharing an anon_vma.
873 *
874 * They need to have the same vm_file, and the flags can only differ
875 * in things that mprotect may change.
876 *
877 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
878 * we can merge the two vma's. For example, we refuse to merge a vma if
879 * there is a vm_ops->close() function, because that indicates that the
880 * driver is doing some kind of reference counting. But that doesn't
881 * really matter for the anon_vma sharing case.
882 */
883static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
884{
885 return a->vm_end == b->vm_start &&
886 mpol_equal(vma_policy(a), vma_policy(b)) &&
887 a->vm_file == b->vm_file &&
888 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
889 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
890}
891
892/*
893 * Do some basic sanity checking to see if we can re-use the anon_vma
894 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
895 * the same as 'old', the other will be the new one that is trying
896 * to share the anon_vma.
897 *
898 * NOTE! This runs with mm_sem held for reading, so it is possible that
899 * the anon_vma of 'old' is concurrently in the process of being set up
900 * by another page fault trying to merge _that_. But that's ok: if it
901 * is being set up, that automatically means that it will be a singleton
902 * acceptable for merging, so we can do all of this optimistically. But
903 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
904 *
905 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
906 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
907 * is to return an anon_vma that is "complex" due to having gone through
908 * a fork).
909 *
910 * We also make sure that the two vma's are compatible (adjacent,
911 * and with the same memory policies). That's all stable, even with just
912 * a read lock on the mm_sem.
913 */
914static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
915{
916 if (anon_vma_compatible(a, b)) {
917 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
918
919 if (anon_vma && list_is_singular(&old->anon_vma_chain))
920 return anon_vma;
921 }
922 return NULL;
923}
924
925/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
927 * neighbouring vmas for a suitable anon_vma, before it goes off
928 * to allocate a new anon_vma. It checks because a repetitive
929 * sequence of mprotects and faults may otherwise lead to distinct
930 * anon_vmas being allocated, preventing vma merge in subsequent
931 * mprotect.
932 */
933struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
934{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700935 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
938 near = vma->vm_next;
939 if (!near)
940 goto try_prev;
941
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700942 anon_vma = reusable_anon_vma(near, vma, near);
943 if (anon_vma)
944 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -0700946 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (!near)
948 goto none;
949
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700950 anon_vma = reusable_anon_vma(near, near, vma);
951 if (anon_vma)
952 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953none:
954 /*
955 * There's no absolute need to look only at touching neighbours:
956 * we could search further afield for "compatible" anon_vmas.
957 * But it would probably just be a waste of time searching,
958 * or lead to too many vmas hanging off the same anon_vma.
959 * We're trying to allow mprotect remerging later on,
960 * not trying to minimize memory used for anon_vmas.
961 */
962 return NULL;
963}
964
965#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700966void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 struct file *file, long pages)
968{
969 const unsigned long stack_flags
970 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
971
Huang Shijie44de9d02012-07-31 16:41:49 -0700972 mm->total_vm += pages;
973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (file) {
975 mm->shared_vm += pages;
976 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
977 mm->exec_vm += pages;
978 } else if (flags & stack_flags)
979 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980}
981#endif /* CONFIG_PROC_FS */
982
983/*
Al Viro40401532012-02-13 03:58:52 +0000984 * If a hint addr is less than mmap_min_addr change hint to be as
985 * low as possible but still greater than mmap_min_addr
986 */
987static inline unsigned long round_hint_to_min(unsigned long hint)
988{
989 hint &= PAGE_MASK;
990 if (((void *)hint != NULL) &&
991 (hint < mmap_min_addr))
992 return PAGE_ALIGN(mmap_min_addr);
993 return hint;
994}
995
996/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700997 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 */
999
Al Viroe3fc6292012-05-30 20:08:42 -04001000unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 unsigned long len, unsigned long prot,
1002 unsigned long flags, unsigned long pgoff)
1003{
1004 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 struct inode *inode;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001006 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 /*
1009 * Does the application expect PROT_READ to imply PROT_EXEC?
1010 *
1011 * (the exception is when the underlying filesystem is noexec
1012 * mounted, in which case we dont add PROT_EXEC.)
1013 */
1014 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001015 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 prot |= PROT_EXEC;
1017
1018 if (!len)
1019 return -EINVAL;
1020
Eric Paris7cd94142007-11-26 18:47:40 -05001021 if (!(flags & MAP_FIXED))
1022 addr = round_hint_to_min(addr);
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 /* Careful about overflows.. */
1025 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001026 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 return -ENOMEM;
1028
1029 /* offset overflow? */
1030 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1031 return -EOVERFLOW;
1032
1033 /* Too many mappings? */
1034 if (mm->map_count > sysctl_max_map_count)
1035 return -ENOMEM;
1036
1037 /* Obtain the address to map to. we verify (or select) it and ensure
1038 * that it represents a valid section of the address space.
1039 */
1040 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1041 if (addr & ~PAGE_MASK)
1042 return addr;
1043
1044 /* Do simple checking here so the lower-level routines won't have
1045 * to. we assume access permissions have been handled by the open
1046 * of the memory object, so we don't do any here.
1047 */
1048 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1049 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1050
Huang Shijiecdf7b342009-09-21 17:03:36 -07001051 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 if (!can_do_mlock())
1053 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 /* mlock MCL_FUTURE? */
1056 if (vm_flags & VM_LOCKED) {
1057 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001058 locked = len >> PAGE_SHIFT;
1059 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001060 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001061 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1063 return -EAGAIN;
1064 }
1065
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001066 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 if (file) {
1069 switch (flags & MAP_TYPE) {
1070 case MAP_SHARED:
1071 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1072 return -EACCES;
1073
1074 /*
1075 * Make sure we don't allow writing to an append-only
1076 * file..
1077 */
1078 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1079 return -EACCES;
1080
1081 /*
1082 * Make sure there are no mandatory locks on the file.
1083 */
1084 if (locks_verify_locked(inode))
1085 return -EAGAIN;
1086
1087 vm_flags |= VM_SHARED | VM_MAYSHARE;
1088 if (!(file->f_mode & FMODE_WRITE))
1089 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1090
1091 /* fall through */
1092 case MAP_PRIVATE:
1093 if (!(file->f_mode & FMODE_READ))
1094 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001095 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001096 if (vm_flags & VM_EXEC)
1097 return -EPERM;
1098 vm_flags &= ~VM_MAYEXEC;
1099 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001100
1101 if (!file->f_op || !file->f_op->mmap)
1102 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 break;
1104
1105 default:
1106 return -EINVAL;
1107 }
1108 } else {
1109 switch (flags & MAP_TYPE) {
1110 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001111 /*
1112 * Ignore pgoff.
1113 */
1114 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 vm_flags |= VM_SHARED | VM_MAYSHARE;
1116 break;
1117 case MAP_PRIVATE:
1118 /*
1119 * Set pgoff according to addr for anon_vma.
1120 */
1121 pgoff = addr >> PAGE_SHIFT;
1122 break;
1123 default:
1124 return -EINVAL;
1125 }
1126 }
1127
Mel Gorman5a6fe122009-02-10 14:02:27 +00001128 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001129}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001130
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001131SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1132 unsigned long, prot, unsigned long, flags,
1133 unsigned long, fd, unsigned long, pgoff)
1134{
1135 struct file *file = NULL;
1136 unsigned long retval = -EBADF;
1137
1138 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001139 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001140 if (unlikely(flags & MAP_HUGETLB))
1141 return -EINVAL;
1142 file = fget(fd);
1143 if (!file)
1144 goto out;
1145 } else if (flags & MAP_HUGETLB) {
1146 struct user_struct *user = NULL;
1147 /*
1148 * VM_NORESERVE is used because the reservations will be
1149 * taken when vm_ops->mmap() is called
1150 * A dummy user value is used because we are not locking
1151 * memory so no accounting is necessary
1152 */
Steven Truelove40716e22012-03-21 16:34:14 -07001153 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1154 VM_NORESERVE, &user,
1155 HUGETLB_ANONHUGE_INODE);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001156 if (IS_ERR(file))
1157 return PTR_ERR(file);
1158 }
1159
1160 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1161
Al Viroeb36c582012-05-30 20:17:35 -04001162 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001163 if (file)
1164 fput(file);
1165out:
1166 return retval;
1167}
1168
Christoph Hellwiga4679372010-03-10 15:21:15 -08001169#ifdef __ARCH_WANT_SYS_OLD_MMAP
1170struct mmap_arg_struct {
1171 unsigned long addr;
1172 unsigned long len;
1173 unsigned long prot;
1174 unsigned long flags;
1175 unsigned long fd;
1176 unsigned long offset;
1177};
1178
1179SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1180{
1181 struct mmap_arg_struct a;
1182
1183 if (copy_from_user(&a, arg, sizeof(a)))
1184 return -EFAULT;
1185 if (a.offset & ~PAGE_MASK)
1186 return -EINVAL;
1187
1188 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1189 a.offset >> PAGE_SHIFT);
1190}
1191#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1192
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001193/*
1194 * Some shared mappigns will want the pages marked read-only
1195 * to track write events. If so, we'll downgrade vm_page_prot
1196 * to the private version (using protection_map[] without the
1197 * VM_SHARED bit).
1198 */
1199int vma_wants_writenotify(struct vm_area_struct *vma)
1200{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001201 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001202
1203 /* If it was private or non-writable, the write bit is already clear */
1204 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1205 return 0;
1206
1207 /* The backer wishes to know when pages are first written to? */
1208 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1209 return 1;
1210
1211 /* The open routine did something to the protections already? */
1212 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001213 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001214 return 0;
1215
1216 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001217 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001218 return 0;
1219
1220 /* Can the mapping track the dirty pages? */
1221 return vma->vm_file && vma->vm_file->f_mapping &&
1222 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1223}
1224
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001225/*
1226 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001227 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001228 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001229static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001230{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001231 /*
1232 * hugetlb has its own accounting separate from the core VM
1233 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1234 */
1235 if (file && is_file_hugepages(file))
1236 return 0;
1237
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001238 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1239}
1240
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001241unsigned long mmap_region(struct file *file, unsigned long addr,
1242 unsigned long len, unsigned long flags,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001243 vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001244{
1245 struct mm_struct *mm = current->mm;
1246 struct vm_area_struct *vma, *prev;
1247 int correct_wcount = 0;
1248 int error;
1249 struct rb_node **rb_link, *rb_parent;
1250 unsigned long charged = 0;
1251 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 /* Clear old maps */
1254 error = -ENOMEM;
1255munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001256 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 if (do_munmap(mm, addr, len))
1258 return -ENOMEM;
1259 goto munmap_back;
1260 }
1261
1262 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001263 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 return -ENOMEM;
1265
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001266 /*
1267 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001268 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001269 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001270 if ((flags & MAP_NORESERVE)) {
1271 /* We honor MAP_NORESERVE if allowed to overcommit */
1272 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1273 vm_flags |= VM_NORESERVE;
1274
1275 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1276 if (file && is_file_hugepages(file))
1277 vm_flags |= VM_NORESERVE;
1278 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001279
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001280 /*
1281 * Private writable mapping: check memory availability
1282 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001283 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001284 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001285 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001286 return -ENOMEM;
1287 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 }
1289
1290 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001291 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001293 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1294 if (vma)
1295 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297 /*
1298 * Determine the object being mapped and call the appropriate
1299 * specific mapper. the address has already been validated, but
1300 * not unmapped, but the maps are removed from the list.
1301 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001302 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 if (!vma) {
1304 error = -ENOMEM;
1305 goto unacct_error;
1306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 vma->vm_mm = mm;
1309 vma->vm_start = addr;
1310 vma->vm_end = addr + len;
1311 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001312 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001314 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
Hugh Dickinsce8fea72012-03-06 12:28:52 -08001316 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1320 goto free_vma;
1321 if (vm_flags & VM_DENYWRITE) {
1322 error = deny_write_access(file);
1323 if (error)
1324 goto free_vma;
1325 correct_wcount = 1;
1326 }
Al Virocb0942b2012-08-27 14:48:26 -04001327 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 error = file->f_op->mmap(file, vma);
1329 if (error)
1330 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001331
1332 /* Can addr have changed??
1333 *
1334 * Answer: Yes, several device drivers can do it in their
1335 * f_op->mmap method. -DaveM
1336 */
1337 addr = vma->vm_start;
1338 pgoff = vma->vm_pgoff;
1339 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 } else if (vm_flags & VM_SHARED) {
Al Viro835ee792012-03-05 06:39:47 +00001341 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1342 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 error = shmem_zero_setup(vma);
1344 if (error)
1345 goto free_vma;
1346 }
1347
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001348 if (vma_wants_writenotify(vma)) {
1349 pgprot_t pprot = vma->vm_page_prot;
1350
1351 /* Can vma->vm_page_prot have changed??
1352 *
1353 * Answer: Yes, drivers may have changed it in their
1354 * f_op->mmap method.
1355 *
1356 * Ensures that vmas marked as uncached stay that way.
1357 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001358 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001359 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1360 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1361 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001362
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001363 vma_link(mm, vma, prev, rb_link, rb_parent);
1364 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001365
1366 /* Once vma denies write, undo our temporary denial count */
1367 if (correct_wcount)
1368 atomic_inc(&inode->i_writecount);
1369out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001370 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001371
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001372 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001374 if (!mlock_vma_pages_range(vma, addr, addr + len))
1375 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001376 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001377 make_pages_present(addr, addr + len);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301378
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001379 if (file)
1380 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 return addr;
1383
1384unmap_and_free_vma:
1385 if (correct_wcount)
1386 atomic_inc(&inode->i_writecount);
1387 vma->vm_file = NULL;
1388 fput(file);
1389
1390 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001391 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1392 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393free_vma:
1394 kmem_cache_free(vm_area_cachep, vma);
1395unacct_error:
1396 if (charged)
1397 vm_unacct_memory(charged);
1398 return error;
1399}
1400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401/* Get an address range which is currently unmapped.
1402 * For shmat() with addr=0.
1403 *
1404 * Ugly calling convention alert:
1405 * Return value with the low bits set means error value,
1406 * ie
1407 * if (ret & ~PAGE_MASK)
1408 * error = ret;
1409 *
1410 * This function "knows" that -ENOMEM has the bits set.
1411 */
1412#ifndef HAVE_ARCH_UNMAPPED_AREA
1413unsigned long
1414arch_get_unmapped_area(struct file *filp, unsigned long addr,
1415 unsigned long len, unsigned long pgoff, unsigned long flags)
1416{
1417 struct mm_struct *mm = current->mm;
1418 struct vm_area_struct *vma;
1419 unsigned long start_addr;
1420
1421 if (len > TASK_SIZE)
1422 return -ENOMEM;
1423
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001424 if (flags & MAP_FIXED)
1425 return addr;
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (addr) {
1428 addr = PAGE_ALIGN(addr);
1429 vma = find_vma(mm, addr);
1430 if (TASK_SIZE - len >= addr &&
1431 (!vma || addr + len <= vma->vm_start))
1432 return addr;
1433 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001434 if (len > mm->cached_hole_size) {
1435 start_addr = addr = mm->free_area_cache;
1436 } else {
1437 start_addr = addr = TASK_UNMAPPED_BASE;
1438 mm->cached_hole_size = 0;
1439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
1441full_search:
1442 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1443 /* At this point: (!vma || addr < vma->vm_end). */
1444 if (TASK_SIZE - len < addr) {
1445 /*
1446 * Start a new search - just in case we missed
1447 * some holes.
1448 */
1449 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001450 addr = TASK_UNMAPPED_BASE;
1451 start_addr = addr;
1452 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 goto full_search;
1454 }
1455 return -ENOMEM;
1456 }
1457 if (!vma || addr + len <= vma->vm_start) {
1458 /*
1459 * Remember the place where we stopped the search:
1460 */
1461 mm->free_area_cache = addr + len;
1462 return addr;
1463 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001464 if (addr + mm->cached_hole_size < vma->vm_start)
1465 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 addr = vma->vm_end;
1467 }
1468}
1469#endif
1470
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001471void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
1473 /*
1474 * Is this a new hole at the lowest possible address?
1475 */
Xiao Guangrongf44d2192012-03-21 16:33:56 -07001476 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001477 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
1479
1480/*
1481 * This mmap-allocator allocates new areas top-down from below the
1482 * stack's low limit (the base):
1483 */
1484#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1485unsigned long
1486arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1487 const unsigned long len, const unsigned long pgoff,
1488 const unsigned long flags)
1489{
1490 struct vm_area_struct *vma;
1491 struct mm_struct *mm = current->mm;
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001492 unsigned long addr = addr0, start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494 /* requested length too big for entire address space */
1495 if (len > TASK_SIZE)
1496 return -ENOMEM;
1497
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001498 if (flags & MAP_FIXED)
1499 return addr;
1500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 /* requesting a specific address */
1502 if (addr) {
1503 addr = PAGE_ALIGN(addr);
1504 vma = find_vma(mm, addr);
1505 if (TASK_SIZE - len >= addr &&
1506 (!vma || addr + len <= vma->vm_start))
1507 return addr;
1508 }
1509
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001510 /* check if free_area_cache is useful for us */
1511 if (len <= mm->cached_hole_size) {
1512 mm->cached_hole_size = 0;
1513 mm->free_area_cache = mm->mmap_base;
1514 }
1515
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001516try_again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 /* either no address requested or can't fit in requested address hole */
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001518 start_addr = addr = mm->free_area_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001520 if (addr < len)
1521 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001523 addr -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 do {
1525 /*
1526 * Lookup failure means no vma is above this address,
1527 * else if new region fits below vma->vm_start,
1528 * return with success:
1529 */
1530 vma = find_vma(mm, addr);
1531 if (!vma || addr+len <= vma->vm_start)
1532 /* remember the address as a hint for next time */
1533 return (mm->free_area_cache = addr);
1534
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001535 /* remember the largest hole we saw so far */
1536 if (addr + mm->cached_hole_size < vma->vm_start)
1537 mm->cached_hole_size = vma->vm_start - addr;
1538
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 /* try just below the current vma->vm_start */
1540 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001541 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001543fail:
1544 /*
1545 * if hint left us with no space for the requested
1546 * mapping then try again:
1547 *
1548 * Note: this is different with the case of bottomup
1549 * which does the fully line-search, but we use find_vma
1550 * here that causes some holes skipped.
1551 */
1552 if (start_addr != mm->mmap_base) {
1553 mm->free_area_cache = mm->mmap_base;
1554 mm->cached_hole_size = 0;
1555 goto try_again;
1556 }
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 /*
1559 * A failed mmap() very likely causes application failure,
1560 * so fall back to the bottom-up function here. This scenario
1561 * can happen with large stack limits and large mmap()
1562 * allocations.
1563 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001564 mm->cached_hole_size = ~0UL;
1565 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1567 /*
1568 * Restore the topdown base:
1569 */
1570 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001571 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
1573 return addr;
1574}
1575#endif
1576
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001577void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578{
1579 /*
1580 * Is this a new hole at the highest possible address?
1581 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001582 if (addr > mm->free_area_cache)
1583 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001586 if (mm->free_area_cache > mm->mmap_base)
1587 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588}
1589
1590unsigned long
1591get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1592 unsigned long pgoff, unsigned long flags)
1593{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001594 unsigned long (*get_area)(struct file *, unsigned long,
1595 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
Al Viro9206de92009-12-03 15:23:11 -05001597 unsigned long error = arch_mmap_check(addr, len, flags);
1598 if (error)
1599 return error;
1600
1601 /* Careful about overflows.. */
1602 if (len > TASK_SIZE)
1603 return -ENOMEM;
1604
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001605 get_area = current->mm->get_unmapped_area;
1606 if (file && file->f_op && file->f_op->get_unmapped_area)
1607 get_area = file->f_op->get_unmapped_area;
1608 addr = get_area(file, addr, len, pgoff, flags);
1609 if (IS_ERR_VALUE(addr))
1610 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001612 if (addr > TASK_SIZE - len)
1613 return -ENOMEM;
1614 if (addr & ~PAGE_MASK)
1615 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001616
Al Viro9ac4ed42012-05-30 17:13:15 -04001617 addr = arch_rebalance_pgtables(addr, len);
1618 error = security_mmap_addr(addr);
1619 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620}
1621
1622EXPORT_SYMBOL(get_unmapped_area);
1623
1624/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001625struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
1627 struct vm_area_struct *vma = NULL;
1628
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001629 if (WARN_ON_ONCE(!mm)) /* Remove this in linux-3.6 */
1630 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001632 /* Check the cache first. */
1633 /* (Cache hit rate is typically around 35%.) */
1634 vma = mm->mmap_cache;
1635 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1636 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001638 rb_node = mm->mm_rb.rb_node;
1639 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001641 while (rb_node) {
1642 struct vm_area_struct *vma_tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001644 vma_tmp = rb_entry(rb_node,
1645 struct vm_area_struct, vm_rb);
1646
1647 if (vma_tmp->vm_end > addr) {
1648 vma = vma_tmp;
1649 if (vma_tmp->vm_start <= addr)
1650 break;
1651 rb_node = rb_node->rb_left;
1652 } else
1653 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 }
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001655 if (vma)
1656 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 }
1658 return vma;
1659}
1660
1661EXPORT_SYMBOL(find_vma);
1662
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001663/*
1664 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001665 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666struct vm_area_struct *
1667find_vma_prev(struct mm_struct *mm, unsigned long addr,
1668 struct vm_area_struct **pprev)
1669{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001670 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001672 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05001673 if (vma) {
1674 *pprev = vma->vm_prev;
1675 } else {
1676 struct rb_node *rb_node = mm->mm_rb.rb_node;
1677 *pprev = NULL;
1678 while (rb_node) {
1679 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1680 rb_node = rb_node->rb_right;
1681 }
1682 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001683 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684}
1685
1686/*
1687 * Verify that the stack growth is acceptable and
1688 * update accounting. This is shared with both the
1689 * grow-up and grow-down cases.
1690 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001691static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
1693 struct mm_struct *mm = vma->vm_mm;
1694 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001695 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
1697 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001698 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 return -ENOMEM;
1700
1701 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001702 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 return -ENOMEM;
1704
1705 /* mlock limit tests */
1706 if (vma->vm_flags & VM_LOCKED) {
1707 unsigned long locked;
1708 unsigned long limit;
1709 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001710 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1711 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 if (locked > limit && !capable(CAP_IPC_LOCK))
1713 return -ENOMEM;
1714 }
1715
Adam Litke0d59a012007-01-30 14:35:39 -08001716 /* Check to ensure the stack will not grow into a hugetlb-only region */
1717 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1718 vma->vm_end - size;
1719 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1720 return -EFAULT;
1721
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 /*
1723 * Overcommit.. This must be the final test, as it will
1724 * update security statistics.
1725 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001726 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 return -ENOMEM;
1728
1729 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 if (vma->vm_flags & VM_LOCKED)
1731 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001732 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 return 0;
1734}
1735
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001736#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001738 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1739 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001741int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742{
1743 int error;
1744
1745 if (!(vma->vm_flags & VM_GROWSUP))
1746 return -EFAULT;
1747
1748 /*
1749 * We must make sure the anon_vma is allocated
1750 * so that the anon_vma locking is not a noop.
1751 */
1752 if (unlikely(anon_vma_prepare(vma)))
1753 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001754 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
1756 /*
1757 * vma->vm_start/vm_end cannot change under us because the caller
1758 * is required to hold the mmap_sem in read mode. We need the
1759 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001760 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 */
Helge Deller06b32f32006-12-19 19:28:33 +01001762 if (address < PAGE_ALIGN(address+4))
1763 address = PAGE_ALIGN(address+4);
1764 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001765 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001766 return -ENOMEM;
1767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 error = 0;
1769
1770 /* Somebody else might have raced and expanded it already */
1771 if (address > vma->vm_end) {
1772 unsigned long size, grow;
1773
1774 size = address - vma->vm_start;
1775 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1776
Hugh Dickins42c36f62011-05-09 17:44:42 -07001777 error = -ENOMEM;
1778 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1779 error = acct_stack_growth(vma, size, grow);
1780 if (!error) {
Michel Lespinassebf181b92012-10-08 16:31:39 -07001781 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07001782 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001783 anon_vma_interval_tree_post_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07001784 perf_event_mmap(vma);
1785 }
Eric B Munson3af9e852010-05-18 15:30:49 +01001786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001788 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001789 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07001790 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 return error;
1792}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001793#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1794
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795/*
1796 * vma is the first one with address < vma->vm_start. Have to extend vma.
1797 */
Michal Hockod05f3162011-05-24 17:11:44 -07001798int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001799 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800{
1801 int error;
1802
1803 /*
1804 * We must make sure the anon_vma is allocated
1805 * so that the anon_vma locking is not a noop.
1806 */
1807 if (unlikely(anon_vma_prepare(vma)))
1808 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001809
1810 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04001811 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05001812 if (error)
1813 return error;
1814
Rik van Rielbb4a3402010-08-09 17:18:37 -07001815 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
1817 /*
1818 * vma->vm_start/vm_end cannot change under us because the caller
1819 * is required to hold the mmap_sem in read mode. We need the
1820 * anon_vma lock to serialize against concurrent expand_stacks.
1821 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
1823 /* Somebody else might have raced and expanded it already */
1824 if (address < vma->vm_start) {
1825 unsigned long size, grow;
1826
1827 size = vma->vm_end - address;
1828 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1829
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001830 error = -ENOMEM;
1831 if (grow <= vma->vm_pgoff) {
1832 error = acct_stack_growth(vma, size, grow);
1833 if (!error) {
Michel Lespinassebf181b92012-10-08 16:31:39 -07001834 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001835 vma->vm_start = address;
1836 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001837 anon_vma_interval_tree_post_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001838 perf_event_mmap(vma);
1839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 }
1841 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001842 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001843 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07001844 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 return error;
1846}
1847
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001848#ifdef CONFIG_STACK_GROWSUP
1849int expand_stack(struct vm_area_struct *vma, unsigned long address)
1850{
1851 return expand_upwards(vma, address);
1852}
1853
1854struct vm_area_struct *
1855find_extend_vma(struct mm_struct *mm, unsigned long addr)
1856{
1857 struct vm_area_struct *vma, *prev;
1858
1859 addr &= PAGE_MASK;
1860 vma = find_vma_prev(mm, addr, &prev);
1861 if (vma && (vma->vm_start <= addr))
1862 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001863 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001864 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001865 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001866 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001867 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001868 return prev;
1869}
1870#else
1871int expand_stack(struct vm_area_struct *vma, unsigned long address)
1872{
1873 return expand_downwards(vma, address);
1874}
1875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876struct vm_area_struct *
1877find_extend_vma(struct mm_struct * mm, unsigned long addr)
1878{
1879 struct vm_area_struct * vma;
1880 unsigned long start;
1881
1882 addr &= PAGE_MASK;
1883 vma = find_vma(mm,addr);
1884 if (!vma)
1885 return NULL;
1886 if (vma->vm_start <= addr)
1887 return vma;
1888 if (!(vma->vm_flags & VM_GROWSDOWN))
1889 return NULL;
1890 start = vma->vm_start;
1891 if (expand_stack(vma, addr))
1892 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001893 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001894 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 return vma;
1897}
1898#endif
1899
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001901 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001903 *
1904 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001906static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001908 unsigned long nr_accounted = 0;
1909
Hugh Dickins365e9c872005-10-29 18:16:18 -07001910 /* Update high watermark before we lower total_vm */
1911 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001913 long nrpages = vma_pages(vma);
1914
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001915 if (vma->vm_flags & VM_ACCOUNT)
1916 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001917 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001918 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001919 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001920 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 validate_mm(mm);
1922}
1923
1924/*
1925 * Get rid of page table information in the indicated region.
1926 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001927 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 */
1929static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001930 struct vm_area_struct *vma, struct vm_area_struct *prev,
1931 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001933 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001934 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
1936 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001937 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001938 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001939 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001940 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1941 next ? next->vm_start : 0);
1942 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943}
1944
1945/*
1946 * Create a list of vma's touched by the unmap, removing them from the mm's
1947 * vma list as we go..
1948 */
1949static void
1950detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1951 struct vm_area_struct *prev, unsigned long end)
1952{
1953 struct vm_area_struct **insertion_point;
1954 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001955 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
1957 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001958 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 do {
1960 rb_erase(&vma->vm_rb, &mm->mm_rb);
1961 mm->map_count--;
1962 tail_vma = vma;
1963 vma = vma->vm_next;
1964 } while (vma && vma->vm_start < end);
1965 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001966 if (vma)
1967 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001969 if (mm->unmap_area == arch_unmap_area)
1970 addr = prev ? prev->vm_end : mm->mmap_base;
1971 else
1972 addr = vma ? vma->vm_start : mm->mmap_base;
1973 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 mm->mmap_cache = NULL; /* Kill the cache. */
1975}
1976
1977/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001978 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1979 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001981static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 unsigned long addr, int new_below)
1983{
1984 struct mempolicy *pol;
1985 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001986 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Andi Kleena5516432008-07-23 21:27:41 -07001988 if (is_vm_hugetlb_page(vma) && (addr &
1989 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 return -EINVAL;
1991
Christoph Lametere94b1762006-12-06 20:33:17 -08001992 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001994 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
1996 /* most fields are the same, copy all, and then fixup */
1997 *new = *vma;
1998
Rik van Riel5beb4932010-03-05 13:42:07 -08001999 INIT_LIST_HEAD(&new->anon_vma_chain);
2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (new_below)
2002 new->vm_end = addr;
2003 else {
2004 new->vm_start = addr;
2005 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2006 }
2007
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002008 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08002010 err = PTR_ERR(pol);
2011 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 }
2013 vma_set_policy(new, pol);
2014
Rik van Riel5beb4932010-03-05 13:42:07 -08002015 if (anon_vma_clone(new, vma))
2016 goto out_free_mpol;
2017
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002018 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 get_file(new->vm_file);
2020
2021 if (new->vm_ops && new->vm_ops->open)
2022 new->vm_ops->open(new);
2023
2024 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002025 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2027 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002028 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Rik van Riel5beb4932010-03-05 13:42:07 -08002030 /* Success. */
2031 if (!err)
2032 return 0;
2033
2034 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002035 if (new->vm_ops && new->vm_ops->close)
2036 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002037 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002038 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002039 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002040 out_free_mpol:
2041 mpol_put(pol);
2042 out_free_vma:
2043 kmem_cache_free(vm_area_cachep, new);
2044 out_err:
2045 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046}
2047
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002048/*
2049 * Split a vma into two pieces at address 'addr', a new vma is allocated
2050 * either for the first part or the tail.
2051 */
2052int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2053 unsigned long addr, int new_below)
2054{
2055 if (mm->map_count >= sysctl_max_map_count)
2056 return -ENOMEM;
2057
2058 return __split_vma(mm, vma, addr, new_below);
2059}
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061/* Munmap is split into 2 main parts -- this part which finds
2062 * what needs doing, and the areas themselves, which do the
2063 * work. This now handles partial unmappings.
2064 * Jeremy Fitzhardinge <jeremy@goop.org>
2065 */
2066int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2067{
2068 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002069 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2072 return -EINVAL;
2073
2074 if ((len = PAGE_ALIGN(len)) == 0)
2075 return -EINVAL;
2076
2077 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002078 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002079 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002081 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002082 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
2084 /* if it doesn't overlap, we have nothing.. */
2085 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002086 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 return 0;
2088
2089 /*
2090 * If we need to split any vma, do it now to save pain later.
2091 *
2092 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2093 * unmapped vm_area_struct will remain in use: so lower split_vma
2094 * places tmp vma above, and higher split_vma places tmp vma below.
2095 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002096 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002097 int error;
2098
2099 /*
2100 * Make sure that map_count on return from munmap() will
2101 * not exceed its limit; but let map_count go just above
2102 * its limit temporarily, to help free resources as expected.
2103 */
2104 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2105 return -ENOMEM;
2106
2107 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 if (error)
2109 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002110 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
2112
2113 /* Does it split the last one? */
2114 last = find_vma(mm, end);
2115 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002116 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 if (error)
2118 return error;
2119 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002120 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
2122 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002123 * unlock any mlock()ed ranges before detaching vmas
2124 */
2125 if (mm->locked_vm) {
2126 struct vm_area_struct *tmp = vma;
2127 while (tmp && tmp->vm_start < end) {
2128 if (tmp->vm_flags & VM_LOCKED) {
2129 mm->locked_vm -= vma_pages(tmp);
2130 munlock_vma_pages_all(tmp);
2131 }
2132 tmp = tmp->vm_next;
2133 }
2134 }
2135
2136 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 * Remove the vma's, and unmap the actual pages
2138 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002139 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2140 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
2142 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002143 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
2145 return 0;
2146}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
Al Virobfce2812012-04-20 21:57:04 -04002148int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002149{
2150 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002151 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002152
2153 down_write(&mm->mmap_sem);
2154 ret = do_munmap(mm, start, len);
2155 up_write(&mm->mmap_sem);
2156 return ret;
2157}
2158EXPORT_SYMBOL(vm_munmap);
2159
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002160SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002163 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164}
2165
2166static inline void verify_mm_writelocked(struct mm_struct *mm)
2167{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002168#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2170 WARN_ON(1);
2171 up_read(&mm->mmap_sem);
2172 }
2173#endif
2174}
2175
2176/*
2177 * this is really a simplified "do_mmap". it only handles
2178 * anonymous maps. eventually we may be able to do some
2179 * brk-specific accounting here.
2180 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002181static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182{
2183 struct mm_struct * mm = current->mm;
2184 struct vm_area_struct * vma, * prev;
2185 unsigned long flags;
2186 struct rb_node ** rb_link, * rb_parent;
2187 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002188 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
2190 len = PAGE_ALIGN(len);
2191 if (!len)
2192 return addr;
2193
Kirill Korotaev3a459752006-09-07 14:17:04 +04002194 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2195
Al Viro2c6a1012009-12-03 19:40:46 -05002196 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2197 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002198 return error;
2199
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 /*
2201 * mlock MCL_FUTURE?
2202 */
2203 if (mm->def_flags & VM_LOCKED) {
2204 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002205 locked = len >> PAGE_SHIFT;
2206 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002207 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002208 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2210 return -EAGAIN;
2211 }
2212
2213 /*
2214 * mm->mmap_sem is required to protect against another thread
2215 * changing the mappings in case we sleep.
2216 */
2217 verify_mm_writelocked(mm);
2218
2219 /*
2220 * Clear old maps. this also does some error checking for us
2221 */
2222 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002223 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 if (do_munmap(mm, addr, len))
2225 return -ENOMEM;
2226 goto munmap_back;
2227 }
2228
2229 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002230 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 return -ENOMEM;
2232
2233 if (mm->map_count > sysctl_max_map_count)
2234 return -ENOMEM;
2235
Al Viro191c5422012-02-13 03:58:52 +00002236 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 return -ENOMEM;
2238
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002240 vma = vma_merge(mm, prev, addr, addr + len, flags,
2241 NULL, NULL, pgoff, NULL);
2242 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 goto out;
2244
2245 /*
2246 * create a vma struct for an anonymous mapping
2247 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002248 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 if (!vma) {
2250 vm_unacct_memory(len >> PAGE_SHIFT);
2251 return -ENOMEM;
2252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
Rik van Riel5beb4932010-03-05 13:42:07 -08002254 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 vma->vm_mm = mm;
2256 vma->vm_start = addr;
2257 vma->vm_end = addr + len;
2258 vma->vm_pgoff = pgoff;
2259 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002260 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 vma_link(mm, vma, prev, rb_link, rb_parent);
2262out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002263 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 mm->total_vm += len >> PAGE_SHIFT;
2265 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002266 if (!mlock_vma_pages_range(vma, addr, addr + len))
2267 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 }
2269 return addr;
2270}
2271
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002272unsigned long vm_brk(unsigned long addr, unsigned long len)
2273{
2274 struct mm_struct *mm = current->mm;
2275 unsigned long ret;
2276
2277 down_write(&mm->mmap_sem);
2278 ret = do_brk(addr, len);
2279 up_write(&mm->mmap_sem);
2280 return ret;
2281}
2282EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
2284/* Release all mmaps. */
2285void exit_mmap(struct mm_struct *mm)
2286{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002287 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002288 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 unsigned long nr_accounted = 0;
2290
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002291 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002292 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002293
Rik van Rielba470de2008-10-18 20:26:50 -07002294 if (mm->locked_vm) {
2295 vma = mm->mmap;
2296 while (vma) {
2297 if (vma->vm_flags & VM_LOCKED)
2298 munlock_vma_pages_all(vma);
2299 vma = vma->vm_next;
2300 }
2301 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002302
2303 arch_exit_mmap(mm);
2304
Rik van Rielba470de2008-10-18 20:26:50 -07002305 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002306 if (!vma) /* Can happen if dup_mmap() received an OOM */
2307 return;
2308
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 flush_cache_mm(mm);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002311 tlb_gather_mmu(&tlb, mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002312 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002313 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002314 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002315
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002316 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Al Viro853f5e22012-03-05 14:03:47 -05002317 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002320 * Walk the list again, actually closing and freeing it,
2321 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002323 while (vma) {
2324 if (vma->vm_flags & VM_ACCOUNT)
2325 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002326 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002327 }
2328 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002329
Hugh Dickinsf9aed622012-08-21 16:15:45 -07002330 WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331}
2332
2333/* Insert vm structure into process list sorted by address
2334 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002335 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002337int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338{
Hugh Dickins6597d782012-10-08 16:29:07 -07002339 struct vm_area_struct *prev;
2340 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
2342 /*
2343 * The vm_pgoff of a purely anonymous vma should be irrelevant
2344 * until its first write fault, when page's anon_vma and index
2345 * are set. But now set the vm_pgoff it will almost certainly
2346 * end up with (unless mremap moves it elsewhere before that
2347 * first wfault), so /proc/pid/maps tells a consistent story.
2348 *
2349 * By setting it to reflect the virtual start address of the
2350 * vma, merges and splits can happen in a seamless way, just
2351 * using the existing file pgoff checks and manipulations.
2352 * Similarly in do_mmap_pgoff and in do_brk.
2353 */
2354 if (!vma->vm_file) {
2355 BUG_ON(vma->anon_vma);
2356 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2357 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002358 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2359 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002361 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002362 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002363 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302364
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 vma_link(mm, vma, prev, rb_link, rb_parent);
2366 return 0;
2367}
2368
2369/*
2370 * Copy the vma structure to a new location in the same mm,
2371 * prior to moving page table entries, to effect an mremap move.
2372 */
2373struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002374 unsigned long addr, unsigned long len, pgoff_t pgoff,
2375 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376{
2377 struct vm_area_struct *vma = *vmap;
2378 unsigned long vma_start = vma->vm_start;
2379 struct mm_struct *mm = vma->vm_mm;
2380 struct vm_area_struct *new_vma, *prev;
2381 struct rb_node **rb_link, *rb_parent;
2382 struct mempolicy *pol;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002383 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
2385 /*
2386 * If anonymous vma has not yet been faulted, update new pgoff
2387 * to match new location, to increase its chance of merging.
2388 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002389 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002391 faulted_in_anon_vma = false;
2392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
Hugh Dickins6597d782012-10-08 16:29:07 -07002394 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2395 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2397 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2398 if (new_vma) {
2399 /*
2400 * Source vma may have been merged into new_vma
2401 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002402 if (unlikely(vma_start >= new_vma->vm_start &&
2403 vma_start < new_vma->vm_end)) {
2404 /*
2405 * The only way we can get a vma_merge with
2406 * self during an mremap is if the vma hasn't
2407 * been faulted in yet and we were allowed to
2408 * reset the dst vma->vm_pgoff to the
2409 * destination address of the mremap to allow
2410 * the merge to happen. mremap must change the
2411 * vm_pgoff linearity between src and dst vmas
2412 * (in turn preventing a vma_merge) to be
2413 * safe. It is only safe to keep the vm_pgoff
2414 * linear if there are no pages mapped yet.
2415 */
2416 VM_BUG_ON(faulted_in_anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002417 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002418 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002419 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002421 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 if (new_vma) {
2423 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 new_vma->vm_start = addr;
2425 new_vma->vm_end = addr + len;
2426 new_vma->vm_pgoff = pgoff;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002427 pol = mpol_dup(vma_policy(vma));
2428 if (IS_ERR(pol))
2429 goto out_free_vma;
2430 vma_set_policy(new_vma, pol);
2431 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2432 if (anon_vma_clone(new_vma, vma))
2433 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002434 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 get_file(new_vma->vm_file);
2436 if (new_vma->vm_ops && new_vma->vm_ops->open)
2437 new_vma->vm_ops->open(new_vma);
2438 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002439 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 }
2441 }
2442 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002443
2444 out_free_mempol:
2445 mpol_put(pol);
2446 out_free_vma:
2447 kmem_cache_free(vm_area_cachep, new_vma);
2448 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002450
2451/*
2452 * Return true if the calling process may expand its vm space by the passed
2453 * number of pages
2454 */
2455int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2456{
2457 unsigned long cur = mm->total_vm; /* pages */
2458 unsigned long lim;
2459
Jiri Slaby59e99e52010-03-05 13:41:44 -08002460 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002461
2462 if (cur + npages > lim)
2463 return 0;
2464 return 1;
2465}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002466
2467
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002468static int special_mapping_fault(struct vm_area_struct *vma,
2469 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002470{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002471 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002472 struct page **pages;
2473
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002474 /*
2475 * special mappings have no vm_file, and in that case, the mm
2476 * uses vm_pgoff internally. So we have to subtract it from here.
2477 * We are allowed to do this because we are the mm; do not copy
2478 * this code into drivers!
2479 */
2480 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002481
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002482 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2483 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002484
2485 if (*pages) {
2486 struct page *page = *pages;
2487 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002488 vmf->page = page;
2489 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002490 }
2491
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002492 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002493}
2494
2495/*
2496 * Having a close hook prevents vma merging regardless of flags.
2497 */
2498static void special_mapping_close(struct vm_area_struct *vma)
2499{
2500}
2501
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002502static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002503 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002504 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002505};
2506
2507/*
2508 * Called with mm->mmap_sem held for writing.
2509 * Insert a new vma covering the given region, with the given flags.
2510 * Its pages are supplied by the given array of struct page *.
2511 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2512 * The region past the last page supplied will always produce SIGBUS.
2513 * The array pointer and the pages it points to are assumed to stay alive
2514 * for as long as this mapping might exist.
2515 */
2516int install_special_mapping(struct mm_struct *mm,
2517 unsigned long addr, unsigned long len,
2518 unsigned long vm_flags, struct page **pages)
2519{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002520 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002521 struct vm_area_struct *vma;
2522
2523 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2524 if (unlikely(vma == NULL))
2525 return -ENOMEM;
2526
Rik van Riel5beb4932010-03-05 13:42:07 -08002527 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002528 vma->vm_mm = mm;
2529 vma->vm_start = addr;
2530 vma->vm_end = addr + len;
2531
Nick Piggin2f987352008-02-02 03:08:53 +01002532 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002533 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002534
2535 vma->vm_ops = &special_mapping_vmops;
2536 vma->vm_private_data = pages;
2537
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002538 ret = insert_vm_struct(mm, vma);
2539 if (ret)
2540 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002541
2542 mm->total_vm += len >> PAGE_SHIFT;
2543
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002544 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002545
Roland McGrathfa5dc222007-02-08 14:20:41 -08002546 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002547
2548out:
2549 kmem_cache_free(vm_area_cachep, vma);
2550 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002551}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002552
2553static DEFINE_MUTEX(mm_all_locks_mutex);
2554
Peter Zijlstra454ed842008-08-11 09:30:25 +02002555static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002556{
Michel Lespinassebf181b92012-10-08 16:31:39 -07002557 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002558 /*
2559 * The LSB of head.next can't change from under us
2560 * because we hold the mm_all_locks_mutex.
2561 */
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002562 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002563 /*
2564 * We can safely modify head.next after taking the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002565 * anon_vma->root->mutex. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002566 * the same anon_vma we won't take it again.
2567 *
2568 * No need of atomic instructions here, head.next
2569 * can't change from under us thanks to the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002570 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002571 */
2572 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07002573 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002574 BUG();
2575 }
2576}
2577
Peter Zijlstra454ed842008-08-11 09:30:25 +02002578static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002579{
2580 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2581 /*
2582 * AS_MM_ALL_LOCKS can't change from under us because
2583 * we hold the mm_all_locks_mutex.
2584 *
2585 * Operations on ->flags have to be atomic because
2586 * even if AS_MM_ALL_LOCKS is stable thanks to the
2587 * mm_all_locks_mutex, there may be other cpus
2588 * changing other bitflags in parallel to us.
2589 */
2590 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2591 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002592 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002593 }
2594}
2595
2596/*
2597 * This operation locks against the VM for all pte/vma/mm related
2598 * operations that could ever happen on a certain mm. This includes
2599 * vmtruncate, try_to_unmap, and all page faults.
2600 *
2601 * The caller must take the mmap_sem in write mode before calling
2602 * mm_take_all_locks(). The caller isn't allowed to release the
2603 * mmap_sem until mm_drop_all_locks() returns.
2604 *
2605 * mmap_sem in write mode is required in order to block all operations
2606 * that could modify pagetables and free pages without need of
2607 * altering the vma layout (for example populate_range() with
2608 * nonlinear vmas). It's also needed in write mode to avoid new
2609 * anon_vmas to be associated with existing vmas.
2610 *
2611 * A single task can't take more than one mm_take_all_locks() in a row
2612 * or it would deadlock.
2613 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07002614 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002615 * mapping->flags avoid to take the same lock twice, if more than one
2616 * vma in this mm is backed by the same anon_vma or address_space.
2617 *
2618 * We can take all the locks in random order because the VM code
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002619 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002620 * takes more than one of them in a row. Secondly we're protected
2621 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2622 *
2623 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2624 * that may have to take thousand of locks.
2625 *
2626 * mm_take_all_locks() can fail if it's interrupted by signals.
2627 */
2628int mm_take_all_locks(struct mm_struct *mm)
2629{
2630 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002631 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002632
2633 BUG_ON(down_read_trylock(&mm->mmap_sem));
2634
2635 mutex_lock(&mm_all_locks_mutex);
2636
2637 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2638 if (signal_pending(current))
2639 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002640 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002641 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002642 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002643
2644 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2645 if (signal_pending(current))
2646 goto out_unlock;
2647 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002648 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2649 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002650 }
2651
Kautuk Consul584cff52011-10-31 17:08:59 -07002652 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002653
2654out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07002655 mm_drop_all_locks(mm);
2656 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002657}
2658
2659static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2660{
Michel Lespinassebf181b92012-10-08 16:31:39 -07002661 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002662 /*
2663 * The LSB of head.next can't change to 0 from under
2664 * us because we hold the mm_all_locks_mutex.
2665 *
2666 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07002667 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002668 * never see our bitflag.
2669 *
2670 * No need of atomic instructions here, head.next
2671 * can't change from under us until we release the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002672 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002673 */
2674 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07002675 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002676 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002677 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002678 }
2679}
2680
2681static void vm_unlock_mapping(struct address_space *mapping)
2682{
2683 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2684 /*
2685 * AS_MM_ALL_LOCKS can't change to 0 from under us
2686 * because we hold the mm_all_locks_mutex.
2687 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002688 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002689 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2690 &mapping->flags))
2691 BUG();
2692 }
2693}
2694
2695/*
2696 * The mmap_sem cannot be released by the caller until
2697 * mm_drop_all_locks() returns.
2698 */
2699void mm_drop_all_locks(struct mm_struct *mm)
2700{
2701 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002702 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002703
2704 BUG_ON(down_read_trylock(&mm->mmap_sem));
2705 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2706
2707 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2708 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002709 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2710 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002711 if (vma->vm_file && vma->vm_file->f_mapping)
2712 vm_unlock_mapping(vma->vm_file->f_mapping);
2713 }
2714
2715 mutex_unlock(&mm_all_locks_mutex);
2716}
David Howells8feae132009-01-08 12:04:47 +00002717
2718/*
2719 * initialise the VMA slab
2720 */
2721void __init mmap_init(void)
2722{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002723 int ret;
2724
2725 ret = percpu_counter_init(&vm_committed_as, 0);
2726 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002727}