blob: 848ef52d96031f5061bc86582aa0120ccd0394e8 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34#include <asm/uaccess.h>
35#include <asm/cacheflush.h>
36#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020037#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Jan Beulich42b77722008-07-23 21:27:10 -070039#include "internal.h"
40
Kirill Korotaev3a459752006-09-07 14:17:04 +040041#ifndef arch_mmap_check
42#define arch_mmap_check(addr, len, flags) (0)
43#endif
44
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080045#ifndef arch_rebalance_pgtables
46#define arch_rebalance_pgtables(addr, len) (addr)
47#endif
48
Hugh Dickinse0da3822005-04-19 13:29:15 -070049static void unmap_region(struct mm_struct *mm,
50 struct vm_area_struct *vma, struct vm_area_struct *prev,
51 unsigned long start, unsigned long end);
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053/*
54 * WARNING: the debugging will use recursive algorithms so never enable this
55 * unless you know what you are doing.
56 */
57#undef DEBUG_MM_RB
58
59/* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
62 *
63 * map_type prot
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
68 *
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 *
73 */
74pgprot_t protection_map[16] = {
75 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
76 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
77};
78
Hugh Dickins804af2c2006-07-26 21:39:49 +010079pgprot_t vm_get_page_prot(unsigned long vm_flags)
80{
Dave Kleikampb845f312008-07-08 00:28:51 +100081 return __pgprot(pgprot_val(protection_map[vm_flags &
82 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
83 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010084}
85EXPORT_SYMBOL(vm_get_page_prot);
86
Shaohua Li34679d72011-05-24 17:11:18 -070087int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
88int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070089int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Shaohua Li34679d72011-05-24 17:11:18 -070090/*
91 * Make sure vm_committed_as in one cacheline and not cacheline shared with
92 * other variables. It can be updated by several CPUs frequently.
93 */
94struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96/*
97 * Check that a process has enough memory to allocate a new virtual
98 * mapping. 0 means there is enough memory for the allocation to
99 * succeed and -ENOMEM implies there is not.
100 *
101 * We currently support three overcommit policies, which are set via the
102 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
103 *
104 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
105 * Additional code 2002 Jul 20 by Robert Love.
106 *
107 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
108 *
109 * Note this is a helper function intended to be used by LSMs which
110 * wish to use this logic.
111 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700112int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
114 unsigned long free, allowed;
115
116 vm_acct_memory(pages);
117
118 /*
119 * Sometimes we want to use more memory than we have
120 */
121 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
122 return 0;
123
124 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700125 free = global_page_state(NR_FREE_PAGES);
126 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Dmitry Finkc15bef32011-07-25 17:12:19 -0700128 /*
129 * shmem pages shouldn't be counted as free in this
130 * case, they can't be purged, only swapped out, and
131 * that won't affect the overall amount of available
132 * memory in the system.
133 */
134 free -= global_page_state(NR_SHMEM);
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 free += nr_swap_pages;
137
138 /*
139 * Any slabs which are created with the
140 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
141 * which are reclaimable, under pressure. The dentry
142 * cache and most inode caches should fall into this
143 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700144 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700147 * Leave reserved pages. The pages are not for anonymous pages.
148 */
149 if (free <= totalreserve_pages)
150 goto error;
151 else
152 free -= totalreserve_pages;
153
154 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 * Leave the last 3% for root
156 */
157 if (!cap_sys_admin)
158 free -= free / 32;
159
160 if (free > pages)
161 return 0;
162
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700163 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 }
165
166 allowed = (totalram_pages - hugetlb_total_pages())
167 * sysctl_overcommit_ratio / 100;
168 /*
169 * Leave the last 3% for root
170 */
171 if (!cap_sys_admin)
172 allowed -= allowed / 32;
173 allowed += total_swap_pages;
174
175 /* Don't let a single process grow too big:
176 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -0700177 if (mm)
178 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700180 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700182error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 vm_unacct_memory(pages);
184
185 return -ENOMEM;
186}
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700189 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 */
191static void __remove_shared_vm_struct(struct vm_area_struct *vma,
192 struct file *file, struct address_space *mapping)
193{
194 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800195 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 if (vma->vm_flags & VM_SHARED)
197 mapping->i_mmap_writable--;
198
199 flush_dcache_mmap_lock(mapping);
200 if (unlikely(vma->vm_flags & VM_NONLINEAR))
201 list_del_init(&vma->shared.vm_set.list);
202 else
203 vma_prio_tree_remove(vma, &mapping->i_mmap);
204 flush_dcache_mmap_unlock(mapping);
205}
206
207/*
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700208 * Unlink a file-based vm structure from its prio_tree, to hide
209 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700211void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
213 struct file *file = vma->vm_file;
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (file) {
216 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700217 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700219 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700221}
222
223/*
224 * Close a vm structure and free it, returning the next.
225 */
226static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
227{
228 struct vm_area_struct *next = vma->vm_next;
229
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700230 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 if (vma->vm_ops && vma->vm_ops->close)
232 vma->vm_ops->close(vma);
Matt Helsley925d1c42008-04-29 01:01:36 -0700233 if (vma->vm_file) {
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700234 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700235 if (vma->vm_flags & VM_EXECUTABLE)
236 removed_exe_file_vma(vma->vm_mm);
237 }
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700238 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700240 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700243static unsigned long do_brk(unsigned long addr, unsigned long len);
244
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100245SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 unsigned long rlim, retval;
248 unsigned long newbrk, oldbrk;
249 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700250 unsigned long min_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 down_write(&mm->mmap_sem);
253
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700254#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800255 /*
256 * CONFIG_COMPAT_BRK can still be overridden by setting
257 * randomize_va_space to 2, which will still cause mm->start_brk
258 * to be arbitrarily shifted
259 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700260 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800261 min_brk = mm->start_brk;
262 else
263 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700264#else
265 min_brk = mm->start_brk;
266#endif
267 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700269
270 /*
271 * Check against rlimit here. If this check is done later after the test
272 * of oldbrk with newbrk then it can escape the test and let the data
273 * segment grow beyond its set limit the in case where the limit is
274 * not page aligned -Ram Gupta
275 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800276 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100277 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
278 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700279 goto out;
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 newbrk = PAGE_ALIGN(brk);
282 oldbrk = PAGE_ALIGN(mm->brk);
283 if (oldbrk == newbrk)
284 goto set_brk;
285
286 /* Always allow shrinking brk. */
287 if (brk <= mm->brk) {
288 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
289 goto set_brk;
290 goto out;
291 }
292
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 /* Check against existing mmap mappings. */
294 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
295 goto out;
296
297 /* Ok, looks good - let it rip. */
298 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
299 goto out;
300set_brk:
301 mm->brk = brk;
302out:
303 retval = mm->brk;
304 up_write(&mm->mmap_sem);
305 return retval;
306}
307
308#ifdef DEBUG_MM_RB
309static int browse_rb(struct rb_root *root)
310{
311 int i = 0, j;
312 struct rb_node *nd, *pn = NULL;
313 unsigned long prev = 0, pend = 0;
314
315 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
316 struct vm_area_struct *vma;
317 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
318 if (vma->vm_start < prev)
319 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
320 if (vma->vm_start < pend)
321 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
322 if (vma->vm_start > vma->vm_end)
323 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
324 i++;
325 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800326 prev = vma->vm_start;
327 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 }
329 j = 0;
330 for (nd = pn; nd; nd = rb_prev(nd)) {
331 j++;
332 }
333 if (i != j)
334 printk("backwards %d, forwards %d\n", j, i), i = 0;
335 return i;
336}
337
338void validate_mm(struct mm_struct *mm)
339{
340 int bug = 0;
341 int i = 0;
342 struct vm_area_struct *tmp = mm->mmap;
343 while (tmp) {
344 tmp = tmp->vm_next;
345 i++;
346 }
347 if (i != mm->map_count)
348 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
349 i = browse_rb(&mm->mm_rb);
350 if (i != mm->map_count)
351 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200352 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353}
354#else
355#define validate_mm(mm) do { } while (0)
356#endif
357
358static struct vm_area_struct *
359find_vma_prepare(struct mm_struct *mm, unsigned long addr,
360 struct vm_area_struct **pprev, struct rb_node ***rb_link,
361 struct rb_node ** rb_parent)
362{
363 struct vm_area_struct * vma;
364 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
365
366 __rb_link = &mm->mm_rb.rb_node;
367 rb_prev = __rb_parent = NULL;
368 vma = NULL;
369
370 while (*__rb_link) {
371 struct vm_area_struct *vma_tmp;
372
373 __rb_parent = *__rb_link;
374 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
375
376 if (vma_tmp->vm_end > addr) {
377 vma = vma_tmp;
378 if (vma_tmp->vm_start <= addr)
Benny Halevydfe195f2008-08-05 13:01:41 -0700379 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 __rb_link = &__rb_parent->rb_left;
381 } else {
382 rb_prev = __rb_parent;
383 __rb_link = &__rb_parent->rb_right;
384 }
385 }
386
387 *pprev = NULL;
388 if (rb_prev)
389 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
390 *rb_link = __rb_link;
391 *rb_parent = __rb_parent;
392 return vma;
393}
394
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
396 struct rb_node **rb_link, struct rb_node *rb_parent)
397{
398 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
399 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
400}
401
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700402static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
ZhenwenXu48aae422009-01-06 14:40:21 -0800404 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
406 file = vma->vm_file;
407 if (file) {
408 struct address_space *mapping = file->f_mapping;
409
410 if (vma->vm_flags & VM_DENYWRITE)
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800411 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 if (vma->vm_flags & VM_SHARED)
413 mapping->i_mmap_writable++;
414
415 flush_dcache_mmap_lock(mapping);
416 if (unlikely(vma->vm_flags & VM_NONLINEAR))
417 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
418 else
419 vma_prio_tree_insert(vma, &mapping->i_mmap);
420 flush_dcache_mmap_unlock(mapping);
421 }
422}
423
424static void
425__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
426 struct vm_area_struct *prev, struct rb_node **rb_link,
427 struct rb_node *rb_parent)
428{
429 __vma_link_list(mm, vma, prev, rb_parent);
430 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431}
432
433static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
434 struct vm_area_struct *prev, struct rb_node **rb_link,
435 struct rb_node *rb_parent)
436{
437 struct address_space *mapping = NULL;
438
439 if (vma->vm_file)
440 mapping = vma->vm_file->f_mapping;
441
Peter Zijlstra97a89412011-05-24 17:12:04 -0700442 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700443 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
445 __vma_link(mm, vma, prev, rb_link, rb_parent);
446 __vma_link_file(vma);
447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700449 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 mm->map_count++;
452 validate_mm(mm);
453}
454
455/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700456 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
457 * mm's list and rbtree. It has already been inserted into the prio_tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800459static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
ZhenwenXu48aae422009-01-06 14:40:21 -0800461 struct vm_area_struct *__vma, *prev;
462 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200465 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 __vma_link(mm, vma, prev, rb_link, rb_parent);
467 mm->map_count++;
468}
469
470static inline void
471__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
472 struct vm_area_struct *prev)
473{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700474 struct vm_area_struct *next = vma->vm_next;
475
476 prev->vm_next = next;
477 if (next)
478 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 rb_erase(&vma->vm_rb, &mm->mm_rb);
480 if (mm->mmap_cache == vma)
481 mm->mmap_cache = prev;
482}
483
484/*
485 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
486 * is already present in an i_mmap tree without adjusting the tree.
487 * The following helper function should be used when such adjustments
488 * are necessary. The "insert" vma (if any) is to be inserted
489 * before we drop the necessary locks.
490 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800491int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
493{
494 struct mm_struct *mm = vma->vm_mm;
495 struct vm_area_struct *next = vma->vm_next;
496 struct vm_area_struct *importer = NULL;
497 struct address_space *mapping = NULL;
498 struct prio_tree_root *root = NULL;
Rik van Riel012f1802010-08-09 17:18:40 -0700499 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 long adjust_next = 0;
502 int remove_next = 0;
503
504 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700505 struct vm_area_struct *exporter = NULL;
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (end >= next->vm_end) {
508 /*
509 * vma expands, overlapping all the next, and
510 * perhaps the one after too (mprotect case 6).
511 */
512again: remove_next = 1 + (end > next->vm_end);
513 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700514 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 importer = vma;
516 } else if (end > next->vm_start) {
517 /*
518 * vma expands, overlapping part of the next:
519 * mprotect case 5 shifting the boundary up.
520 */
521 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700522 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 importer = vma;
524 } else if (end < vma->vm_end) {
525 /*
526 * vma shrinks, and !insert tells it's not
527 * split_vma inserting another: so it must be
528 * mprotect case 4 shifting the boundary down.
529 */
530 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700531 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 importer = next;
533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Rik van Riel5beb4932010-03-05 13:42:07 -0800535 /*
536 * Easily overlooked: when mprotect shifts the boundary,
537 * make sure the expanding vma has anon_vma set if the
538 * shrinking vma had, to cover any anon pages imported.
539 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700540 if (exporter && exporter->anon_vma && !importer->anon_vma) {
541 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800542 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700543 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800544 }
545 }
546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 if (file) {
548 mapping = file->f_mapping;
549 if (!(vma->vm_flags & VM_NONLINEAR))
550 root = &mapping->i_mmap;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700551 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 /*
554 * Put into prio_tree now, so instantiated pages
555 * are visible to arm/parisc __flush_dcache_page
556 * throughout; but we cannot insert into address
557 * space until vma start or end is updated.
558 */
559 __vma_link_file(insert);
560 }
561 }
562
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800563 vma_adjust_trans_huge(vma, start, end, adjust_next);
564
Rik van Riel012f1802010-08-09 17:18:40 -0700565 /*
566 * When changing only vma->vm_end, we don't really need anon_vma
567 * lock. This is a fairly rare case by itself, but the anon_vma
568 * lock may be shared between many sibling processes. Skipping
569 * the lock for brk adjustments makes a difference sometimes.
570 */
Shaohua Li5f70b962011-05-24 17:11:19 -0700571 if (vma->anon_vma && (importer || start != vma->vm_start)) {
Rik van Riel012f1802010-08-09 17:18:40 -0700572 anon_vma = vma->anon_vma;
573 anon_vma_lock(anon_vma);
574 }
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 if (root) {
577 flush_dcache_mmap_lock(mapping);
578 vma_prio_tree_remove(vma, root);
579 if (adjust_next)
580 vma_prio_tree_remove(next, root);
581 }
582
583 vma->vm_start = start;
584 vma->vm_end = end;
585 vma->vm_pgoff = pgoff;
586 if (adjust_next) {
587 next->vm_start += adjust_next << PAGE_SHIFT;
588 next->vm_pgoff += adjust_next;
589 }
590
591 if (root) {
592 if (adjust_next)
593 vma_prio_tree_insert(next, root);
594 vma_prio_tree_insert(vma, root);
595 flush_dcache_mmap_unlock(mapping);
596 }
597
598 if (remove_next) {
599 /*
600 * vma_merge has merged next into vma, and needs
601 * us to remove next before dropping the locks.
602 */
603 __vma_unlink(mm, next, vma);
604 if (file)
605 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 } else if (insert) {
607 /*
608 * split_vma has split insert from vma, and needs
609 * us to insert it before dropping the locks
610 * (it may either follow vma or precede it).
611 */
612 __insert_vm_struct(mm, insert);
613 }
614
Rik van Riel012f1802010-08-09 17:18:40 -0700615 if (anon_vma)
616 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700618 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
620 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700621 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700623 if (next->vm_flags & VM_EXECUTABLE)
624 removed_exe_file_vma(mm);
625 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800626 if (next->anon_vma)
627 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 mm->map_count--;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700629 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 kmem_cache_free(vm_area_cachep, next);
631 /*
632 * In mprotect's case 6 (see comments on vma_merge),
633 * we must remove another next too. It would clutter
634 * up the code too much to do both in one go.
635 */
636 if (remove_next == 2) {
637 next = vma->vm_next;
638 goto again;
639 }
640 }
641
642 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800643
644 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647/*
648 * If the vma has a ->close operation then the driver probably needs to release
649 * per-vma resources, so we don't attempt to merge those.
650 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651static inline int is_mergeable_vma(struct vm_area_struct *vma,
652 struct file *file, unsigned long vm_flags)
653{
Hugh Dickins8314c4f2009-09-21 17:02:25 -0700654 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
655 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 return 0;
657 if (vma->vm_file != file)
658 return 0;
659 if (vma->vm_ops && vma->vm_ops->close)
660 return 0;
661 return 1;
662}
663
664static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700665 struct anon_vma *anon_vma2,
666 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Shaohua Li965f55d2011-05-24 17:11:20 -0700668 /*
669 * The list_is_singular() test is to avoid merging VMA cloned from
670 * parents. This can improve scalability caused by anon_vma lock.
671 */
672 if ((!anon_vma1 || !anon_vma2) && (!vma ||
673 list_is_singular(&vma->anon_vma_chain)))
674 return 1;
675 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
678/*
679 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
680 * in front of (at a lower virtual address and file offset than) the vma.
681 *
682 * We cannot merge two vmas if they have differently assigned (non-NULL)
683 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
684 *
685 * We don't check here for the merged mmap wrapping around the end of pagecache
686 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
687 * wrap, nor mmaps which cover the final page at index -1UL.
688 */
689static int
690can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
691 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
692{
693 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700694 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 if (vma->vm_pgoff == vm_pgoff)
696 return 1;
697 }
698 return 0;
699}
700
701/*
702 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
703 * beyond (at a higher virtual address and file offset than) the vma.
704 *
705 * We cannot merge two vmas if they have differently assigned (non-NULL)
706 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
707 */
708static int
709can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
710 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
711{
712 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700713 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 pgoff_t vm_pglen;
715 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
716 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
717 return 1;
718 }
719 return 0;
720}
721
722/*
723 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
724 * whether that can be merged with its predecessor or its successor.
725 * Or both (it neatly fills a hole).
726 *
727 * In most cases - when called for mmap, brk or mremap - [addr,end) is
728 * certain not to be mapped by the time vma_merge is called; but when
729 * called for mprotect, it is certain to be already mapped (either at
730 * an offset within prev, or at the start of next), and the flags of
731 * this area are about to be changed to vm_flags - and the no-change
732 * case has already been eliminated.
733 *
734 * The following mprotect cases have to be considered, where AAAA is
735 * the area passed down from mprotect_fixup, never extending beyond one
736 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
737 *
738 * AAAA AAAA AAAA AAAA
739 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
740 * cannot merge might become might become might become
741 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
742 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
743 * mremap move: PPPPNNNNNNNN 8
744 * AAAA
745 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
746 * might become case 1 below case 2 below case 3 below
747 *
748 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
749 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
750 */
751struct vm_area_struct *vma_merge(struct mm_struct *mm,
752 struct vm_area_struct *prev, unsigned long addr,
753 unsigned long end, unsigned long vm_flags,
754 struct anon_vma *anon_vma, struct file *file,
755 pgoff_t pgoff, struct mempolicy *policy)
756{
757 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
758 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800759 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761 /*
762 * We later require that vma->vm_flags == vm_flags,
763 * so this tests vma->vm_flags & VM_SPECIAL, too.
764 */
765 if (vm_flags & VM_SPECIAL)
766 return NULL;
767
768 if (prev)
769 next = prev->vm_next;
770 else
771 next = mm->mmap;
772 area = next;
773 if (next && next->vm_end == end) /* cases 6, 7, 8 */
774 next = next->vm_next;
775
776 /*
777 * Can it merge with the predecessor?
778 */
779 if (prev && prev->vm_end == addr &&
780 mpol_equal(vma_policy(prev), policy) &&
781 can_vma_merge_after(prev, vm_flags,
782 anon_vma, file, pgoff)) {
783 /*
784 * OK, it can. Can we now merge in the successor as well?
785 */
786 if (next && end == next->vm_start &&
787 mpol_equal(policy, vma_policy(next)) &&
788 can_vma_merge_before(next, vm_flags,
789 anon_vma, file, pgoff+pglen) &&
790 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700791 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800793 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 next->vm_end, prev->vm_pgoff, NULL);
795 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800796 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800798 if (err)
799 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800800 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 return prev;
802 }
803
804 /*
805 * Can this new request be merged in front of next?
806 */
807 if (next && end == next->vm_start &&
808 mpol_equal(policy, vma_policy(next)) &&
809 can_vma_merge_before(next, vm_flags,
810 anon_vma, file, pgoff+pglen)) {
811 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800812 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 addr, prev->vm_pgoff, NULL);
814 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800815 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800817 if (err)
818 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -0800819 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 return area;
821 }
822
823 return NULL;
824}
825
826/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700827 * Rough compatbility check to quickly see if it's even worth looking
828 * at sharing an anon_vma.
829 *
830 * They need to have the same vm_file, and the flags can only differ
831 * in things that mprotect may change.
832 *
833 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
834 * we can merge the two vma's. For example, we refuse to merge a vma if
835 * there is a vm_ops->close() function, because that indicates that the
836 * driver is doing some kind of reference counting. But that doesn't
837 * really matter for the anon_vma sharing case.
838 */
839static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
840{
841 return a->vm_end == b->vm_start &&
842 mpol_equal(vma_policy(a), vma_policy(b)) &&
843 a->vm_file == b->vm_file &&
844 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
845 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
846}
847
848/*
849 * Do some basic sanity checking to see if we can re-use the anon_vma
850 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
851 * the same as 'old', the other will be the new one that is trying
852 * to share the anon_vma.
853 *
854 * NOTE! This runs with mm_sem held for reading, so it is possible that
855 * the anon_vma of 'old' is concurrently in the process of being set up
856 * by another page fault trying to merge _that_. But that's ok: if it
857 * is being set up, that automatically means that it will be a singleton
858 * acceptable for merging, so we can do all of this optimistically. But
859 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
860 *
861 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
862 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
863 * is to return an anon_vma that is "complex" due to having gone through
864 * a fork).
865 *
866 * We also make sure that the two vma's are compatible (adjacent,
867 * and with the same memory policies). That's all stable, even with just
868 * a read lock on the mm_sem.
869 */
870static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
871{
872 if (anon_vma_compatible(a, b)) {
873 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
874
875 if (anon_vma && list_is_singular(&old->anon_vma_chain))
876 return anon_vma;
877 }
878 return NULL;
879}
880
881/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
883 * neighbouring vmas for a suitable anon_vma, before it goes off
884 * to allocate a new anon_vma. It checks because a repetitive
885 * sequence of mprotects and faults may otherwise lead to distinct
886 * anon_vmas being allocated, preventing vma merge in subsequent
887 * mprotect.
888 */
889struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
890{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700891 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
894 near = vma->vm_next;
895 if (!near)
896 goto try_prev;
897
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700898 anon_vma = reusable_anon_vma(near, vma, near);
899 if (anon_vma)
900 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -0700902 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (!near)
904 goto none;
905
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -0700906 anon_vma = reusable_anon_vma(near, near, vma);
907 if (anon_vma)
908 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909none:
910 /*
911 * There's no absolute need to look only at touching neighbours:
912 * we could search further afield for "compatible" anon_vmas.
913 * But it would probably just be a waste of time searching,
914 * or lead to too many vmas hanging off the same anon_vma.
915 * We're trying to allow mprotect remerging later on,
916 * not trying to minimize memory used for anon_vmas.
917 */
918 return NULL;
919}
920
921#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700922void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 struct file *file, long pages)
924{
925 const unsigned long stack_flags
926 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (file) {
929 mm->shared_vm += pages;
930 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
931 mm->exec_vm += pages;
932 } else if (flags & stack_flags)
933 mm->stack_vm += pages;
934 if (flags & (VM_RESERVED|VM_IO))
935 mm->reserved_vm += pages;
936}
937#endif /* CONFIG_PROC_FS */
938
939/*
Al Viro40401532012-02-13 03:58:52 +0000940 * If a hint addr is less than mmap_min_addr change hint to be as
941 * low as possible but still greater than mmap_min_addr
942 */
943static inline unsigned long round_hint_to_min(unsigned long hint)
944{
945 hint &= PAGE_MASK;
946 if (((void *)hint != NULL) &&
947 (hint < mmap_min_addr))
948 return PAGE_ALIGN(mmap_min_addr);
949 return hint;
950}
951
952/*
Jianjun Kong27f5de72009-09-17 19:26:26 -0700953 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 */
955
Linus Torvalds6be5ceb2012-04-20 17:13:58 -0700956static unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 unsigned long len, unsigned long prot,
958 unsigned long flags, unsigned long pgoff)
959{
960 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 struct inode *inode;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900962 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 int error;
Miklos Szeredi0165ab42007-07-15 23:38:26 -0700964 unsigned long reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 /*
967 * Does the application expect PROT_READ to imply PROT_EXEC?
968 *
969 * (the exception is when the underlying filesystem is noexec
970 * mounted, in which case we dont add PROT_EXEC.)
971 */
972 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -0800973 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 prot |= PROT_EXEC;
975
976 if (!len)
977 return -EINVAL;
978
Eric Paris7cd94142007-11-26 18:47:40 -0500979 if (!(flags & MAP_FIXED))
980 addr = round_hint_to_min(addr);
981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 /* Careful about overflows.. */
983 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -0500984 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 return -ENOMEM;
986
987 /* offset overflow? */
988 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
989 return -EOVERFLOW;
990
991 /* Too many mappings? */
992 if (mm->map_count > sysctl_max_map_count)
993 return -ENOMEM;
994
995 /* Obtain the address to map to. we verify (or select) it and ensure
996 * that it represents a valid section of the address space.
997 */
998 addr = get_unmapped_area(file, addr, len, pgoff, flags);
999 if (addr & ~PAGE_MASK)
1000 return addr;
1001
1002 /* Do simple checking here so the lower-level routines won't have
1003 * to. we assume access permissions have been handled by the open
1004 * of the memory object, so we don't do any here.
1005 */
1006 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1007 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1008
Huang Shijiecdf7b342009-09-21 17:03:36 -07001009 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 if (!can_do_mlock())
1011 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 /* mlock MCL_FUTURE? */
1014 if (vm_flags & VM_LOCKED) {
1015 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001016 locked = len >> PAGE_SHIFT;
1017 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001018 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001019 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1021 return -EAGAIN;
1022 }
1023
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001024 inode = file ? file->f_path.dentry->d_inode : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026 if (file) {
1027 switch (flags & MAP_TYPE) {
1028 case MAP_SHARED:
1029 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1030 return -EACCES;
1031
1032 /*
1033 * Make sure we don't allow writing to an append-only
1034 * file..
1035 */
1036 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1037 return -EACCES;
1038
1039 /*
1040 * Make sure there are no mandatory locks on the file.
1041 */
1042 if (locks_verify_locked(inode))
1043 return -EAGAIN;
1044
1045 vm_flags |= VM_SHARED | VM_MAYSHARE;
1046 if (!(file->f_mode & FMODE_WRITE))
1047 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1048
1049 /* fall through */
1050 case MAP_PRIVATE:
1051 if (!(file->f_mode & FMODE_READ))
1052 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001053 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001054 if (vm_flags & VM_EXEC)
1055 return -EPERM;
1056 vm_flags &= ~VM_MAYEXEC;
1057 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001058
1059 if (!file->f_op || !file->f_op->mmap)
1060 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 break;
1062
1063 default:
1064 return -EINVAL;
1065 }
1066 } else {
1067 switch (flags & MAP_TYPE) {
1068 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001069 /*
1070 * Ignore pgoff.
1071 */
1072 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 vm_flags |= VM_SHARED | VM_MAYSHARE;
1074 break;
1075 case MAP_PRIVATE:
1076 /*
1077 * Set pgoff according to addr for anon_vma.
1078 */
1079 pgoff = addr >> PAGE_SHIFT;
1080 break;
1081 default:
1082 return -EINVAL;
1083 }
1084 }
1085
Eric Parised032182007-06-28 15:55:21 -04001086 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 if (error)
1088 return error;
Eric Parised032182007-06-28 15:55:21 -04001089
Mel Gorman5a6fe122009-02-10 14:02:27 +00001090 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001091}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001092
1093unsigned long do_mmap(struct file *file, unsigned long addr,
1094 unsigned long len, unsigned long prot,
1095 unsigned long flag, unsigned long offset)
1096{
1097 if (unlikely(offset + PAGE_ALIGN(len) < offset))
1098 return -EINVAL;
1099 if (unlikely(offset & ~PAGE_MASK))
1100 return -EINVAL;
1101 return do_mmap_pgoff(file, addr, len, prot, flag, offset >> PAGE_SHIFT);
1102}
1103EXPORT_SYMBOL(do_mmap);
1104
1105unsigned long vm_mmap(struct file *file, unsigned long addr,
1106 unsigned long len, unsigned long prot,
1107 unsigned long flag, unsigned long offset)
1108{
1109 unsigned long ret;
1110 struct mm_struct *mm = current->mm;
1111
1112 down_write(&mm->mmap_sem);
1113 ret = do_mmap(file, addr, len, prot, flag, offset);
1114 up_write(&mm->mmap_sem);
1115 return ret;
1116}
1117EXPORT_SYMBOL(vm_mmap);
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001118
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001119SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1120 unsigned long, prot, unsigned long, flags,
1121 unsigned long, fd, unsigned long, pgoff)
1122{
1123 struct file *file = NULL;
1124 unsigned long retval = -EBADF;
1125
1126 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001127 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001128 if (unlikely(flags & MAP_HUGETLB))
1129 return -EINVAL;
1130 file = fget(fd);
1131 if (!file)
1132 goto out;
1133 } else if (flags & MAP_HUGETLB) {
1134 struct user_struct *user = NULL;
1135 /*
1136 * VM_NORESERVE is used because the reservations will be
1137 * taken when vm_ops->mmap() is called
1138 * A dummy user value is used because we are not locking
1139 * memory so no accounting is necessary
1140 */
Steven Truelove40716e22012-03-21 16:34:14 -07001141 file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1142 VM_NORESERVE, &user,
1143 HUGETLB_ANONHUGE_INODE);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001144 if (IS_ERR(file))
1145 return PTR_ERR(file);
1146 }
1147
1148 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1149
1150 down_write(&current->mm->mmap_sem);
1151 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1152 up_write(&current->mm->mmap_sem);
1153
1154 if (file)
1155 fput(file);
1156out:
1157 return retval;
1158}
1159
Christoph Hellwiga4679372010-03-10 15:21:15 -08001160#ifdef __ARCH_WANT_SYS_OLD_MMAP
1161struct mmap_arg_struct {
1162 unsigned long addr;
1163 unsigned long len;
1164 unsigned long prot;
1165 unsigned long flags;
1166 unsigned long fd;
1167 unsigned long offset;
1168};
1169
1170SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1171{
1172 struct mmap_arg_struct a;
1173
1174 if (copy_from_user(&a, arg, sizeof(a)))
1175 return -EFAULT;
1176 if (a.offset & ~PAGE_MASK)
1177 return -EINVAL;
1178
1179 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1180 a.offset >> PAGE_SHIFT);
1181}
1182#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1183
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001184/*
1185 * Some shared mappigns will want the pages marked read-only
1186 * to track write events. If so, we'll downgrade vm_page_prot
1187 * to the private version (using protection_map[] without the
1188 * VM_SHARED bit).
1189 */
1190int vma_wants_writenotify(struct vm_area_struct *vma)
1191{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001192 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001193
1194 /* If it was private or non-writable, the write bit is already clear */
1195 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1196 return 0;
1197
1198 /* The backer wishes to know when pages are first written to? */
1199 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1200 return 1;
1201
1202 /* The open routine did something to the protections already? */
1203 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001204 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001205 return 0;
1206
1207 /* Specialty mapping? */
1208 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1209 return 0;
1210
1211 /* Can the mapping track the dirty pages? */
1212 return vma->vm_file && vma->vm_file->f_mapping &&
1213 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1214}
1215
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001216/*
1217 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001218 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001219 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001220static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001221{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001222 /*
1223 * hugetlb has its own accounting separate from the core VM
1224 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1225 */
1226 if (file && is_file_hugepages(file))
1227 return 0;
1228
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001229 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1230}
1231
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001232unsigned long mmap_region(struct file *file, unsigned long addr,
1233 unsigned long len, unsigned long flags,
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001234 vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001235{
1236 struct mm_struct *mm = current->mm;
1237 struct vm_area_struct *vma, *prev;
1238 int correct_wcount = 0;
1239 int error;
1240 struct rb_node **rb_link, *rb_parent;
1241 unsigned long charged = 0;
1242 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 /* Clear old maps */
1245 error = -ENOMEM;
1246munmap_back:
1247 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1248 if (vma && vma->vm_start < addr + len) {
1249 if (do_munmap(mm, addr, len))
1250 return -ENOMEM;
1251 goto munmap_back;
1252 }
1253
1254 /* Check against address space limit. */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001255 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 return -ENOMEM;
1257
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001258 /*
1259 * Set 'VM_NORESERVE' if we should not account for the
Mel Gorman5a6fe122009-02-10 14:02:27 +00001260 * memory use of this mapping.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001261 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001262 if ((flags & MAP_NORESERVE)) {
1263 /* We honor MAP_NORESERVE if allowed to overcommit */
1264 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1265 vm_flags |= VM_NORESERVE;
1266
1267 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1268 if (file && is_file_hugepages(file))
1269 vm_flags |= VM_NORESERVE;
1270 }
Andy Whitcroftcdfd4322008-07-23 21:27:28 -07001271
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001272 /*
1273 * Private writable mapping: check memory availability
1274 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001275 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001276 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001277 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001278 return -ENOMEM;
1279 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
1281
1282 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001283 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001285 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1286 if (vma)
1287 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
1289 /*
1290 * Determine the object being mapped and call the appropriate
1291 * specific mapper. the address has already been validated, but
1292 * not unmapped, but the maps are removed from the list.
1293 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001294 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 if (!vma) {
1296 error = -ENOMEM;
1297 goto unacct_error;
1298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
1300 vma->vm_mm = mm;
1301 vma->vm_start = addr;
1302 vma->vm_end = addr + len;
1303 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001304 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001306 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Hugh Dickinsce8fea72012-03-06 12:28:52 -08001308 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1312 goto free_vma;
1313 if (vm_flags & VM_DENYWRITE) {
1314 error = deny_write_access(file);
1315 if (error)
1316 goto free_vma;
1317 correct_wcount = 1;
1318 }
1319 vma->vm_file = file;
1320 get_file(file);
1321 error = file->f_op->mmap(file, vma);
1322 if (error)
1323 goto unmap_and_free_vma;
Matt Helsley925d1c42008-04-29 01:01:36 -07001324 if (vm_flags & VM_EXECUTABLE)
1325 added_exe_file_vma(mm);
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001326
1327 /* Can addr have changed??
1328 *
1329 * Answer: Yes, several device drivers can do it in their
1330 * f_op->mmap method. -DaveM
1331 */
1332 addr = vma->vm_start;
1333 pgoff = vma->vm_pgoff;
1334 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 } else if (vm_flags & VM_SHARED) {
Al Viro835ee792012-03-05 06:39:47 +00001336 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1337 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 error = shmem_zero_setup(vma);
1339 if (error)
1340 goto free_vma;
1341 }
1342
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001343 if (vma_wants_writenotify(vma)) {
1344 pgprot_t pprot = vma->vm_page_prot;
1345
1346 /* Can vma->vm_page_prot have changed??
1347 *
1348 * Answer: Yes, drivers may have changed it in their
1349 * f_op->mmap method.
1350 *
1351 * Ensures that vmas marked as uncached stay that way.
1352 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001353 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001354 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1355 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1356 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001357
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001358 vma_link(mm, vma, prev, rb_link, rb_parent);
1359 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001360
1361 /* Once vma denies write, undo our temporary denial count */
1362 if (correct_wcount)
1363 atomic_inc(&inode->i_writecount);
1364out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001365 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 mm->total_vm += len >> PAGE_SHIFT;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001368 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 if (vm_flags & VM_LOCKED) {
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001370 if (!mlock_vma_pages_range(vma, addr, addr + len))
1371 mm->locked_vm += (len >> PAGE_SHIFT);
Rik van Rielba470de2008-10-18 20:26:50 -07001372 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
Nick Piggin54cb8822007-07-19 01:46:59 -07001373 make_pages_present(addr, addr + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 return addr;
1375
1376unmap_and_free_vma:
1377 if (correct_wcount)
1378 atomic_inc(&inode->i_writecount);
1379 vma->vm_file = NULL;
1380 fput(file);
1381
1382 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001383 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1384 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385free_vma:
1386 kmem_cache_free(vm_area_cachep, vma);
1387unacct_error:
1388 if (charged)
1389 vm_unacct_memory(charged);
1390 return error;
1391}
1392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393/* Get an address range which is currently unmapped.
1394 * For shmat() with addr=0.
1395 *
1396 * Ugly calling convention alert:
1397 * Return value with the low bits set means error value,
1398 * ie
1399 * if (ret & ~PAGE_MASK)
1400 * error = ret;
1401 *
1402 * This function "knows" that -ENOMEM has the bits set.
1403 */
1404#ifndef HAVE_ARCH_UNMAPPED_AREA
1405unsigned long
1406arch_get_unmapped_area(struct file *filp, unsigned long addr,
1407 unsigned long len, unsigned long pgoff, unsigned long flags)
1408{
1409 struct mm_struct *mm = current->mm;
1410 struct vm_area_struct *vma;
1411 unsigned long start_addr;
1412
1413 if (len > TASK_SIZE)
1414 return -ENOMEM;
1415
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001416 if (flags & MAP_FIXED)
1417 return addr;
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (addr) {
1420 addr = PAGE_ALIGN(addr);
1421 vma = find_vma(mm, addr);
1422 if (TASK_SIZE - len >= addr &&
1423 (!vma || addr + len <= vma->vm_start))
1424 return addr;
1425 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001426 if (len > mm->cached_hole_size) {
1427 start_addr = addr = mm->free_area_cache;
1428 } else {
1429 start_addr = addr = TASK_UNMAPPED_BASE;
1430 mm->cached_hole_size = 0;
1431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
1433full_search:
1434 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1435 /* At this point: (!vma || addr < vma->vm_end). */
1436 if (TASK_SIZE - len < addr) {
1437 /*
1438 * Start a new search - just in case we missed
1439 * some holes.
1440 */
1441 if (start_addr != TASK_UNMAPPED_BASE) {
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001442 addr = TASK_UNMAPPED_BASE;
1443 start_addr = addr;
1444 mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 goto full_search;
1446 }
1447 return -ENOMEM;
1448 }
1449 if (!vma || addr + len <= vma->vm_start) {
1450 /*
1451 * Remember the place where we stopped the search:
1452 */
1453 mm->free_area_cache = addr + len;
1454 return addr;
1455 }
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001456 if (addr + mm->cached_hole_size < vma->vm_start)
1457 mm->cached_hole_size = vma->vm_start - addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 addr = vma->vm_end;
1459 }
1460}
1461#endif
1462
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001463void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464{
1465 /*
1466 * Is this a new hole at the lowest possible address?
1467 */
Xiao Guangrongf44d2192012-03-21 16:33:56 -07001468 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001469 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
1472/*
1473 * This mmap-allocator allocates new areas top-down from below the
1474 * stack's low limit (the base):
1475 */
1476#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1477unsigned long
1478arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1479 const unsigned long len, const unsigned long pgoff,
1480 const unsigned long flags)
1481{
1482 struct vm_area_struct *vma;
1483 struct mm_struct *mm = current->mm;
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001484 unsigned long addr = addr0, start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
1486 /* requested length too big for entire address space */
1487 if (len > TASK_SIZE)
1488 return -ENOMEM;
1489
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001490 if (flags & MAP_FIXED)
1491 return addr;
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 /* requesting a specific address */
1494 if (addr) {
1495 addr = PAGE_ALIGN(addr);
1496 vma = find_vma(mm, addr);
1497 if (TASK_SIZE - len >= addr &&
1498 (!vma || addr + len <= vma->vm_start))
1499 return addr;
1500 }
1501
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001502 /* check if free_area_cache is useful for us */
1503 if (len <= mm->cached_hole_size) {
1504 mm->cached_hole_size = 0;
1505 mm->free_area_cache = mm->mmap_base;
1506 }
1507
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001508try_again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 /* either no address requested or can't fit in requested address hole */
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001510 start_addr = addr = mm->free_area_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001512 if (addr < len)
1513 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001515 addr -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 do {
1517 /*
1518 * Lookup failure means no vma is above this address,
1519 * else if new region fits below vma->vm_start,
1520 * return with success:
1521 */
1522 vma = find_vma(mm, addr);
1523 if (!vma || addr+len <= vma->vm_start)
1524 /* remember the address as a hint for next time */
1525 return (mm->free_area_cache = addr);
1526
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001527 /* remember the largest hole we saw so far */
1528 if (addr + mm->cached_hole_size < vma->vm_start)
1529 mm->cached_hole_size = vma->vm_start - addr;
1530
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 /* try just below the current vma->vm_start */
1532 addr = vma->vm_start-len;
Linus Torvalds49a43872005-05-18 15:39:33 -07001533 } while (len < vma->vm_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001535fail:
1536 /*
1537 * if hint left us with no space for the requested
1538 * mapping then try again:
1539 *
1540 * Note: this is different with the case of bottomup
1541 * which does the fully line-search, but we use find_vma
1542 * here that causes some holes skipped.
1543 */
1544 if (start_addr != mm->mmap_base) {
1545 mm->free_area_cache = mm->mmap_base;
1546 mm->cached_hole_size = 0;
1547 goto try_again;
1548 }
1549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 /*
1551 * A failed mmap() very likely causes application failure,
1552 * so fall back to the bottom-up function here. This scenario
1553 * can happen with large stack limits and large mmap()
1554 * allocations.
1555 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001556 mm->cached_hole_size = ~0UL;
1557 mm->free_area_cache = TASK_UNMAPPED_BASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1559 /*
1560 * Restore the topdown base:
1561 */
1562 mm->free_area_cache = mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001563 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
1565 return addr;
1566}
1567#endif
1568
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001569void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570{
1571 /*
1572 * Is this a new hole at the highest possible address?
1573 */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001574 if (addr > mm->free_area_cache)
1575 mm->free_area_cache = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
1577 /* dont allow allocations above current base */
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001578 if (mm->free_area_cache > mm->mmap_base)
1579 mm->free_area_cache = mm->mmap_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580}
1581
1582unsigned long
1583get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1584 unsigned long pgoff, unsigned long flags)
1585{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001586 unsigned long (*get_area)(struct file *, unsigned long,
1587 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
Al Viro9206de92009-12-03 15:23:11 -05001589 unsigned long error = arch_mmap_check(addr, len, flags);
1590 if (error)
1591 return error;
1592
1593 /* Careful about overflows.. */
1594 if (len > TASK_SIZE)
1595 return -ENOMEM;
1596
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001597 get_area = current->mm->get_unmapped_area;
1598 if (file && file->f_op && file->f_op->get_unmapped_area)
1599 get_area = file->f_op->get_unmapped_area;
1600 addr = get_area(file, addr, len, pgoff, flags);
1601 if (IS_ERR_VALUE(addr))
1602 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001604 if (addr > TASK_SIZE - len)
1605 return -ENOMEM;
1606 if (addr & ~PAGE_MASK)
1607 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001608
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -08001609 return arch_rebalance_pgtables(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610}
1611
1612EXPORT_SYMBOL(get_unmapped_area);
1613
1614/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001615struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616{
1617 struct vm_area_struct *vma = NULL;
1618
1619 if (mm) {
1620 /* Check the cache first. */
1621 /* (Cache hit rate is typically around 35%.) */
1622 vma = mm->mmap_cache;
1623 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1624 struct rb_node * rb_node;
1625
1626 rb_node = mm->mm_rb.rb_node;
1627 vma = NULL;
1628
1629 while (rb_node) {
1630 struct vm_area_struct * vma_tmp;
1631
1632 vma_tmp = rb_entry(rb_node,
1633 struct vm_area_struct, vm_rb);
1634
1635 if (vma_tmp->vm_end > addr) {
1636 vma = vma_tmp;
1637 if (vma_tmp->vm_start <= addr)
1638 break;
1639 rb_node = rb_node->rb_left;
1640 } else
1641 rb_node = rb_node->rb_right;
1642 }
1643 if (vma)
1644 mm->mmap_cache = vma;
1645 }
1646 }
1647 return vma;
1648}
1649
1650EXPORT_SYMBOL(find_vma);
1651
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001652/*
1653 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001654 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655struct vm_area_struct *
1656find_vma_prev(struct mm_struct *mm, unsigned long addr,
1657 struct vm_area_struct **pprev)
1658{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001659 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001661 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05001662 if (vma) {
1663 *pprev = vma->vm_prev;
1664 } else {
1665 struct rb_node *rb_node = mm->mm_rb.rb_node;
1666 *pprev = NULL;
1667 while (rb_node) {
1668 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1669 rb_node = rb_node->rb_right;
1670 }
1671 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001672 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
1674
1675/*
1676 * Verify that the stack growth is acceptable and
1677 * update accounting. This is shared with both the
1678 * grow-up and grow-down cases.
1679 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001680static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
1682 struct mm_struct *mm = vma->vm_mm;
1683 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08001684 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
1686 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07001687 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 return -ENOMEM;
1689
1690 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08001691 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 return -ENOMEM;
1693
1694 /* mlock limit tests */
1695 if (vma->vm_flags & VM_LOCKED) {
1696 unsigned long locked;
1697 unsigned long limit;
1698 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001699 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1700 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 if (locked > limit && !capable(CAP_IPC_LOCK))
1702 return -ENOMEM;
1703 }
1704
Adam Litke0d59a012007-01-30 14:35:39 -08001705 /* Check to ensure the stack will not grow into a hugetlb-only region */
1706 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1707 vma->vm_end - size;
1708 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1709 return -EFAULT;
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 /*
1712 * Overcommit.. This must be the final test, as it will
1713 * update security statistics.
1714 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01001715 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 return -ENOMEM;
1717
1718 /* Ok, everything looks good - let it rip */
1719 mm->total_vm += grow;
1720 if (vma->vm_flags & VM_LOCKED)
1721 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001722 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 return 0;
1724}
1725
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001726#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001728 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1729 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001731int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732{
1733 int error;
1734
1735 if (!(vma->vm_flags & VM_GROWSUP))
1736 return -EFAULT;
1737
1738 /*
1739 * We must make sure the anon_vma is allocated
1740 * so that the anon_vma locking is not a noop.
1741 */
1742 if (unlikely(anon_vma_prepare(vma)))
1743 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07001744 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
1746 /*
1747 * vma->vm_start/vm_end cannot change under us because the caller
1748 * is required to hold the mmap_sem in read mode. We need the
1749 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01001750 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 */
Helge Deller06b32f32006-12-19 19:28:33 +01001752 if (address < PAGE_ALIGN(address+4))
1753 address = PAGE_ALIGN(address+4);
1754 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07001755 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01001756 return -ENOMEM;
1757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 error = 0;
1759
1760 /* Somebody else might have raced and expanded it already */
1761 if (address > vma->vm_end) {
1762 unsigned long size, grow;
1763
1764 size = address - vma->vm_start;
1765 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1766
Hugh Dickins42c36f62011-05-09 17:44:42 -07001767 error = -ENOMEM;
1768 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1769 error = acct_stack_growth(vma, size, grow);
1770 if (!error) {
1771 vma->vm_end = address;
1772 perf_event_mmap(vma);
1773 }
Eric B Munson3af9e852010-05-18 15:30:49 +01001774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001776 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001777 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 return error;
1779}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07001780#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782/*
1783 * vma is the first one with address < vma->vm_start. Have to extend vma.
1784 */
Michal Hockod05f3162011-05-24 17:11:44 -07001785int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001786 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787{
1788 int error;
1789
1790 /*
1791 * We must make sure the anon_vma is allocated
1792 * so that the anon_vma locking is not a noop.
1793 */
1794 if (unlikely(anon_vma_prepare(vma)))
1795 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05001796
1797 address &= PAGE_MASK;
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01001798 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
Eric Paris88694772007-11-26 18:47:26 -05001799 if (error)
1800 return error;
1801
Rik van Rielbb4a3402010-08-09 17:18:37 -07001802 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 /*
1805 * vma->vm_start/vm_end cannot change under us because the caller
1806 * is required to hold the mmap_sem in read mode. We need the
1807 * anon_vma lock to serialize against concurrent expand_stacks.
1808 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
1810 /* Somebody else might have raced and expanded it already */
1811 if (address < vma->vm_start) {
1812 unsigned long size, grow;
1813
1814 size = vma->vm_end - address;
1815 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1816
Linus Torvaldsa626ca62011-04-13 08:07:28 -07001817 error = -ENOMEM;
1818 if (grow <= vma->vm_pgoff) {
1819 error = acct_stack_growth(vma, size, grow);
1820 if (!error) {
1821 vma->vm_start = address;
1822 vma->vm_pgoff -= grow;
1823 perf_event_mmap(vma);
1824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 }
1826 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07001827 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001828 khugepaged_enter_vma_merge(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 return error;
1830}
1831
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001832#ifdef CONFIG_STACK_GROWSUP
1833int expand_stack(struct vm_area_struct *vma, unsigned long address)
1834{
1835 return expand_upwards(vma, address);
1836}
1837
1838struct vm_area_struct *
1839find_extend_vma(struct mm_struct *mm, unsigned long addr)
1840{
1841 struct vm_area_struct *vma, *prev;
1842
1843 addr &= PAGE_MASK;
1844 vma = find_vma_prev(mm, addr, &prev);
1845 if (vma && (vma->vm_start <= addr))
1846 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01001847 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001848 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001849 if (prev->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001850 mlock_vma_pages_range(prev, addr, prev->vm_end);
Rik van Rielba470de2008-10-18 20:26:50 -07001851 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001852 return prev;
1853}
1854#else
1855int expand_stack(struct vm_area_struct *vma, unsigned long address)
1856{
1857 return expand_downwards(vma, address);
1858}
1859
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860struct vm_area_struct *
1861find_extend_vma(struct mm_struct * mm, unsigned long addr)
1862{
1863 struct vm_area_struct * vma;
1864 unsigned long start;
1865
1866 addr &= PAGE_MASK;
1867 vma = find_vma(mm,addr);
1868 if (!vma)
1869 return NULL;
1870 if (vma->vm_start <= addr)
1871 return vma;
1872 if (!(vma->vm_flags & VM_GROWSDOWN))
1873 return NULL;
1874 start = vma->vm_start;
1875 if (expand_stack(vma, addr))
1876 return NULL;
Rik van Rielba470de2008-10-18 20:26:50 -07001877 if (vma->vm_flags & VM_LOCKED) {
KOSAKI Motohiroc58267c2010-03-05 13:41:43 -08001878 mlock_vma_pages_range(vma, addr, start);
Rik van Rielba470de2008-10-18 20:26:50 -07001879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 return vma;
1881}
1882#endif
1883
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001885 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001887 *
1888 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001890static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891{
Hugh Dickins365e9c872005-10-29 18:16:18 -07001892 /* Update high watermark before we lower total_vm */
1893 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001895 long nrpages = vma_pages(vma);
1896
1897 mm->total_vm -= nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07001898 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07001899 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07001900 } while (vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 validate_mm(mm);
1902}
1903
1904/*
1905 * Get rid of page table information in the indicated region.
1906 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07001907 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 */
1909static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07001910 struct vm_area_struct *vma, struct vm_area_struct *prev,
1911 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912{
Hugh Dickinse0da3822005-04-19 13:29:15 -07001913 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001914 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 unsigned long nr_accounted = 0;
1916
1917 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001918 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001919 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001920 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 vm_unacct_memory(nr_accounted);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001922 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1923 next ? next->vm_start : 0);
1924 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925}
1926
1927/*
1928 * Create a list of vma's touched by the unmap, removing them from the mm's
1929 * vma list as we go..
1930 */
1931static void
1932detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1933 struct vm_area_struct *prev, unsigned long end)
1934{
1935 struct vm_area_struct **insertion_point;
1936 struct vm_area_struct *tail_vma = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001937 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
1939 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001940 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 do {
1942 rb_erase(&vma->vm_rb, &mm->mm_rb);
1943 mm->map_count--;
1944 tail_vma = vma;
1945 vma = vma->vm_next;
1946 } while (vma && vma->vm_start < end);
1947 *insertion_point = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -07001948 if (vma)
1949 vma->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 tail_vma->vm_next = NULL;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001951 if (mm->unmap_area == arch_unmap_area)
1952 addr = prev ? prev->vm_end : mm->mmap_base;
1953 else
1954 addr = vma ? vma->vm_start : mm->mmap_base;
1955 mm->unmap_area(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 mm->mmap_cache = NULL; /* Kill the cache. */
1957}
1958
1959/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001960 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1961 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08001963static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 unsigned long addr, int new_below)
1965{
1966 struct mempolicy *pol;
1967 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08001968 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
Andi Kleena5516432008-07-23 21:27:41 -07001970 if (is_vm_hugetlb_page(vma) && (addr &
1971 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 return -EINVAL;
1973
Christoph Lametere94b1762006-12-06 20:33:17 -08001974 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08001976 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
1978 /* most fields are the same, copy all, and then fixup */
1979 *new = *vma;
1980
Rik van Riel5beb4932010-03-05 13:42:07 -08001981 INIT_LIST_HEAD(&new->anon_vma_chain);
1982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (new_below)
1984 new->vm_end = addr;
1985 else {
1986 new->vm_start = addr;
1987 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1988 }
1989
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001990 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001992 err = PTR_ERR(pol);
1993 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
1995 vma_set_policy(new, pol);
1996
Rik van Riel5beb4932010-03-05 13:42:07 -08001997 if (anon_vma_clone(new, vma))
1998 goto out_free_mpol;
1999
Matt Helsley925d1c42008-04-29 01:01:36 -07002000 if (new->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 get_file(new->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002002 if (vma->vm_flags & VM_EXECUTABLE)
2003 added_exe_file_vma(mm);
2004 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
2006 if (new->vm_ops && new->vm_ops->open)
2007 new->vm_ops->open(new);
2008
2009 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002010 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2012 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002013 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Rik van Riel5beb4932010-03-05 13:42:07 -08002015 /* Success. */
2016 if (!err)
2017 return 0;
2018
2019 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002020 if (new->vm_ops && new->vm_ops->close)
2021 new->vm_ops->close(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002022 if (new->vm_file) {
2023 if (vma->vm_flags & VM_EXECUTABLE)
2024 removed_exe_file_vma(mm);
2025 fput(new->vm_file);
2026 }
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002027 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002028 out_free_mpol:
2029 mpol_put(pol);
2030 out_free_vma:
2031 kmem_cache_free(vm_area_cachep, new);
2032 out_err:
2033 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034}
2035
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002036/*
2037 * Split a vma into two pieces at address 'addr', a new vma is allocated
2038 * either for the first part or the tail.
2039 */
2040int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2041 unsigned long addr, int new_below)
2042{
2043 if (mm->map_count >= sysctl_max_map_count)
2044 return -ENOMEM;
2045
2046 return __split_vma(mm, vma, addr, new_below);
2047}
2048
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049/* Munmap is split into 2 main parts -- this part which finds
2050 * what needs doing, and the areas themselves, which do the
2051 * work. This now handles partial unmappings.
2052 * Jeremy Fitzhardinge <jeremy@goop.org>
2053 */
2054int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2055{
2056 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002057 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
2059 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2060 return -EINVAL;
2061
2062 if ((len = PAGE_ALIGN(len)) == 0)
2063 return -EINVAL;
2064
2065 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002066 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002067 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002069 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002070 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
2072 /* if it doesn't overlap, we have nothing.. */
2073 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002074 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 return 0;
2076
2077 /*
2078 * If we need to split any vma, do it now to save pain later.
2079 *
2080 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2081 * unmapped vm_area_struct will remain in use: so lower split_vma
2082 * places tmp vma above, and higher split_vma places tmp vma below.
2083 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002084 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002085 int error;
2086
2087 /*
2088 * Make sure that map_count on return from munmap() will
2089 * not exceed its limit; but let map_count go just above
2090 * its limit temporarily, to help free resources as expected.
2091 */
2092 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2093 return -ENOMEM;
2094
2095 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 if (error)
2097 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002098 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 }
2100
2101 /* Does it split the last one? */
2102 last = find_vma(mm, end);
2103 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002104 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if (error)
2106 return error;
2107 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002108 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
2110 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002111 * unlock any mlock()ed ranges before detaching vmas
2112 */
2113 if (mm->locked_vm) {
2114 struct vm_area_struct *tmp = vma;
2115 while (tmp && tmp->vm_start < end) {
2116 if (tmp->vm_flags & VM_LOCKED) {
2117 mm->locked_vm -= vma_pages(tmp);
2118 munlock_vma_pages_all(tmp);
2119 }
2120 tmp = tmp->vm_next;
2121 }
2122 }
2123
2124 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 * Remove the vma's, and unmap the actual pages
2126 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002127 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2128 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
2130 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002131 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
2133 return 0;
2134}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135EXPORT_SYMBOL(do_munmap);
2136
Al Virobfce2812012-04-20 21:57:04 -04002137int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002138{
2139 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002140 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002141
2142 down_write(&mm->mmap_sem);
2143 ret = do_munmap(mm, start, len);
2144 up_write(&mm->mmap_sem);
2145 return ret;
2146}
2147EXPORT_SYMBOL(vm_munmap);
2148
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002149SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002152 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153}
2154
2155static inline void verify_mm_writelocked(struct mm_struct *mm)
2156{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002157#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2159 WARN_ON(1);
2160 up_read(&mm->mmap_sem);
2161 }
2162#endif
2163}
2164
2165/*
2166 * this is really a simplified "do_mmap". it only handles
2167 * anonymous maps. eventually we may be able to do some
2168 * brk-specific accounting here.
2169 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002170static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171{
2172 struct mm_struct * mm = current->mm;
2173 struct vm_area_struct * vma, * prev;
2174 unsigned long flags;
2175 struct rb_node ** rb_link, * rb_parent;
2176 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002177 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
2179 len = PAGE_ALIGN(len);
2180 if (!len)
2181 return addr;
2182
Richard Knutsson88c3f7a2007-12-08 12:02:48 +01002183 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
Eric Paris5a211a52007-12-04 11:06:55 -05002184 if (error)
2185 return error;
2186
Kirill Korotaev3a459752006-09-07 14:17:04 +04002187 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2188
Al Viro2c6a10162009-12-03 19:40:46 -05002189 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2190 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002191 return error;
2192
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 /*
2194 * mlock MCL_FUTURE?
2195 */
2196 if (mm->def_flags & VM_LOCKED) {
2197 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002198 locked = len >> PAGE_SHIFT;
2199 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002200 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002201 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2203 return -EAGAIN;
2204 }
2205
2206 /*
2207 * mm->mmap_sem is required to protect against another thread
2208 * changing the mappings in case we sleep.
2209 */
2210 verify_mm_writelocked(mm);
2211
2212 /*
2213 * Clear old maps. this also does some error checking for us
2214 */
2215 munmap_back:
2216 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2217 if (vma && vma->vm_start < addr + len) {
2218 if (do_munmap(mm, addr, len))
2219 return -ENOMEM;
2220 goto munmap_back;
2221 }
2222
2223 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002224 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 return -ENOMEM;
2226
2227 if (mm->map_count > sysctl_max_map_count)
2228 return -ENOMEM;
2229
Al Viro191c5422012-02-13 03:58:52 +00002230 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 return -ENOMEM;
2232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002234 vma = vma_merge(mm, prev, addr, addr + len, flags,
2235 NULL, NULL, pgoff, NULL);
2236 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 goto out;
2238
2239 /*
2240 * create a vma struct for an anonymous mapping
2241 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002242 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 if (!vma) {
2244 vm_unacct_memory(len >> PAGE_SHIFT);
2245 return -ENOMEM;
2246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
Rik van Riel5beb4932010-03-05 13:42:07 -08002248 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 vma->vm_mm = mm;
2250 vma->vm_start = addr;
2251 vma->vm_end = addr + len;
2252 vma->vm_pgoff = pgoff;
2253 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002254 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 vma_link(mm, vma, prev, rb_link, rb_parent);
2256out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002257 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 mm->total_vm += len >> PAGE_SHIFT;
2259 if (flags & VM_LOCKED) {
Rik van Rielba470de2008-10-18 20:26:50 -07002260 if (!mlock_vma_pages_range(vma, addr, addr + len))
2261 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 }
2263 return addr;
2264}
2265
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002266unsigned long vm_brk(unsigned long addr, unsigned long len)
2267{
2268 struct mm_struct *mm = current->mm;
2269 unsigned long ret;
2270
2271 down_write(&mm->mmap_sem);
2272 ret = do_brk(addr, len);
2273 up_write(&mm->mmap_sem);
2274 return ret;
2275}
2276EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
2278/* Release all mmaps. */
2279void exit_mmap(struct mm_struct *mm)
2280{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002281 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002282 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 unsigned long nr_accounted = 0;
2284
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002285 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002286 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002287
Rik van Rielba470de2008-10-18 20:26:50 -07002288 if (mm->locked_vm) {
2289 vma = mm->mmap;
2290 while (vma) {
2291 if (vma->vm_flags & VM_LOCKED)
2292 munlock_vma_pages_all(vma);
2293 vma = vma->vm_next;
2294 }
2295 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002296
2297 arch_exit_mmap(mm);
2298
Rik van Rielba470de2008-10-18 20:26:50 -07002299 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002300 if (!vma) /* Can happen if dup_mmap() received an OOM */
2301 return;
2302
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 flush_cache_mm(mm);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002305 tlb_gather_mmu(&tlb, mm, 1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002306 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002307 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Al Viro6e8bb012012-03-05 13:41:15 -05002308 unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 vm_unacct_memory(nr_accounted);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002310
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002311 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
Al Viro853f5e22012-03-05 14:03:47 -05002312 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002315 * Walk the list again, actually closing and freeing it,
2316 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002318 while (vma)
2319 vma = remove_vma(vma);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002320
Hugh Dickinse2cdef82005-04-19 13:29:19 -07002321 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322}
2323
2324/* Insert vm structure into process list sorted by address
2325 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002326 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 */
2328int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2329{
2330 struct vm_area_struct * __vma, * prev;
2331 struct rb_node ** rb_link, * rb_parent;
2332
2333 /*
2334 * The vm_pgoff of a purely anonymous vma should be irrelevant
2335 * until its first write fault, when page's anon_vma and index
2336 * are set. But now set the vm_pgoff it will almost certainly
2337 * end up with (unless mremap moves it elsewhere before that
2338 * first wfault), so /proc/pid/maps tells a consistent story.
2339 *
2340 * By setting it to reflect the virtual start address of the
2341 * vma, merges and splits can happen in a seamless way, just
2342 * using the existing file pgoff checks and manipulations.
2343 * Similarly in do_mmap_pgoff and in do_brk.
2344 */
2345 if (!vma->vm_file) {
2346 BUG_ON(vma->anon_vma);
2347 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2348 }
2349 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2350 if (__vma && __vma->vm_start < vma->vm_end)
2351 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002352 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002353 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002354 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 vma_link(mm, vma, prev, rb_link, rb_parent);
2356 return 0;
2357}
2358
2359/*
2360 * Copy the vma structure to a new location in the same mm,
2361 * prior to moving page table entries, to effect an mremap move.
2362 */
2363struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2364 unsigned long addr, unsigned long len, pgoff_t pgoff)
2365{
2366 struct vm_area_struct *vma = *vmap;
2367 unsigned long vma_start = vma->vm_start;
2368 struct mm_struct *mm = vma->vm_mm;
2369 struct vm_area_struct *new_vma, *prev;
2370 struct rb_node **rb_link, *rb_parent;
2371 struct mempolicy *pol;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002372 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
2374 /*
2375 * If anonymous vma has not yet been faulted, update new pgoff
2376 * to match new location, to increase its chance of merging.
2377 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002378 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002380 faulted_in_anon_vma = false;
2381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
2383 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2384 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2385 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2386 if (new_vma) {
2387 /*
2388 * Source vma may have been merged into new_vma
2389 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002390 if (unlikely(vma_start >= new_vma->vm_start &&
2391 vma_start < new_vma->vm_end)) {
2392 /*
2393 * The only way we can get a vma_merge with
2394 * self during an mremap is if the vma hasn't
2395 * been faulted in yet and we were allowed to
2396 * reset the dst vma->vm_pgoff to the
2397 * destination address of the mremap to allow
2398 * the merge to happen. mremap must change the
2399 * vm_pgoff linearity between src and dst vmas
2400 * (in turn preventing a vma_merge) to be
2401 * safe. It is only safe to keep the vm_pgoff
2402 * linear if there are no pages mapped yet.
2403 */
2404 VM_BUG_ON(faulted_in_anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 *vmap = new_vma;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002406 } else
2407 anon_vma_moveto_tail(new_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002409 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 if (new_vma) {
2411 *new_vma = *vma;
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002412 pol = mpol_dup(vma_policy(vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002413 if (IS_ERR(pol))
2414 goto out_free_vma;
2415 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2416 if (anon_vma_clone(new_vma, vma))
2417 goto out_free_mempol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 vma_set_policy(new_vma, pol);
2419 new_vma->vm_start = addr;
2420 new_vma->vm_end = addr + len;
2421 new_vma->vm_pgoff = pgoff;
Matt Helsley925d1c42008-04-29 01:01:36 -07002422 if (new_vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 get_file(new_vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07002424 if (vma->vm_flags & VM_EXECUTABLE)
2425 added_exe_file_vma(mm);
2426 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 if (new_vma->vm_ops && new_vma->vm_ops->open)
2428 new_vma->vm_ops->open(new_vma);
2429 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2430 }
2431 }
2432 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002433
2434 out_free_mempol:
2435 mpol_put(pol);
2436 out_free_vma:
2437 kmem_cache_free(vm_area_cachep, new_vma);
2438 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002440
2441/*
2442 * Return true if the calling process may expand its vm space by the passed
2443 * number of pages
2444 */
2445int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2446{
2447 unsigned long cur = mm->total_vm; /* pages */
2448 unsigned long lim;
2449
Jiri Slaby59e99e52010-03-05 13:41:44 -08002450 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002451
2452 if (cur + npages > lim)
2453 return 0;
2454 return 1;
2455}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002456
2457
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002458static int special_mapping_fault(struct vm_area_struct *vma,
2459 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002460{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002461 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002462 struct page **pages;
2463
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002464 /*
2465 * special mappings have no vm_file, and in that case, the mm
2466 * uses vm_pgoff internally. So we have to subtract it from here.
2467 * We are allowed to do this because we are the mm; do not copy
2468 * this code into drivers!
2469 */
2470 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002471
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002472 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2473 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002474
2475 if (*pages) {
2476 struct page *page = *pages;
2477 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002478 vmf->page = page;
2479 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002480 }
2481
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002482 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002483}
2484
2485/*
2486 * Having a close hook prevents vma merging regardless of flags.
2487 */
2488static void special_mapping_close(struct vm_area_struct *vma)
2489{
2490}
2491
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002492static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002493 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002494 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002495};
2496
2497/*
2498 * Called with mm->mmap_sem held for writing.
2499 * Insert a new vma covering the given region, with the given flags.
2500 * Its pages are supplied by the given array of struct page *.
2501 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2502 * The region past the last page supplied will always produce SIGBUS.
2503 * The array pointer and the pages it points to are assumed to stay alive
2504 * for as long as this mapping might exist.
2505 */
2506int install_special_mapping(struct mm_struct *mm,
2507 unsigned long addr, unsigned long len,
2508 unsigned long vm_flags, struct page **pages)
2509{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002510 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002511 struct vm_area_struct *vma;
2512
2513 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2514 if (unlikely(vma == NULL))
2515 return -ENOMEM;
2516
Rik van Riel5beb4932010-03-05 13:42:07 -08002517 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002518 vma->vm_mm = mm;
2519 vma->vm_start = addr;
2520 vma->vm_end = addr + len;
2521
Nick Piggin2f987352008-02-02 03:08:53 +01002522 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002523 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002524
2525 vma->vm_ops = &special_mapping_vmops;
2526 vma->vm_private_data = pages;
2527
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002528 ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2529 if (ret)
2530 goto out;
2531
2532 ret = insert_vm_struct(mm, vma);
2533 if (ret)
2534 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002535
2536 mm->total_vm += len >> PAGE_SHIFT;
2537
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002538 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002539
Roland McGrathfa5dc222007-02-08 14:20:41 -08002540 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002541
2542out:
2543 kmem_cache_free(vm_area_cachep, vma);
2544 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002545}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002546
2547static DEFINE_MUTEX(mm_all_locks_mutex);
2548
Peter Zijlstra454ed842008-08-11 09:30:25 +02002549static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002550{
Rik van Riel012f1802010-08-09 17:18:40 -07002551 if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002552 /*
2553 * The LSB of head.next can't change from under us
2554 * because we hold the mm_all_locks_mutex.
2555 */
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002556 mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002557 /*
2558 * We can safely modify head.next after taking the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002559 * anon_vma->root->mutex. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002560 * the same anon_vma we won't take it again.
2561 *
2562 * No need of atomic instructions here, head.next
2563 * can't change from under us thanks to the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002564 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002565 */
2566 if (__test_and_set_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002567 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002568 BUG();
2569 }
2570}
2571
Peter Zijlstra454ed842008-08-11 09:30:25 +02002572static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002573{
2574 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2575 /*
2576 * AS_MM_ALL_LOCKS can't change from under us because
2577 * we hold the mm_all_locks_mutex.
2578 *
2579 * Operations on ->flags have to be atomic because
2580 * even if AS_MM_ALL_LOCKS is stable thanks to the
2581 * mm_all_locks_mutex, there may be other cpus
2582 * changing other bitflags in parallel to us.
2583 */
2584 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2585 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002586 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002587 }
2588}
2589
2590/*
2591 * This operation locks against the VM for all pte/vma/mm related
2592 * operations that could ever happen on a certain mm. This includes
2593 * vmtruncate, try_to_unmap, and all page faults.
2594 *
2595 * The caller must take the mmap_sem in write mode before calling
2596 * mm_take_all_locks(). The caller isn't allowed to release the
2597 * mmap_sem until mm_drop_all_locks() returns.
2598 *
2599 * mmap_sem in write mode is required in order to block all operations
2600 * that could modify pagetables and free pages without need of
2601 * altering the vma layout (for example populate_range() with
2602 * nonlinear vmas). It's also needed in write mode to avoid new
2603 * anon_vmas to be associated with existing vmas.
2604 *
2605 * A single task can't take more than one mm_take_all_locks() in a row
2606 * or it would deadlock.
2607 *
2608 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2609 * mapping->flags avoid to take the same lock twice, if more than one
2610 * vma in this mm is backed by the same anon_vma or address_space.
2611 *
2612 * We can take all the locks in random order because the VM code
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002613 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002614 * takes more than one of them in a row. Secondly we're protected
2615 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2616 *
2617 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2618 * that may have to take thousand of locks.
2619 *
2620 * mm_take_all_locks() can fail if it's interrupted by signals.
2621 */
2622int mm_take_all_locks(struct mm_struct *mm)
2623{
2624 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002625 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002626
2627 BUG_ON(down_read_trylock(&mm->mmap_sem));
2628
2629 mutex_lock(&mm_all_locks_mutex);
2630
2631 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2632 if (signal_pending(current))
2633 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002634 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02002635 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002636 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002637
2638 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2639 if (signal_pending(current))
2640 goto out_unlock;
2641 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002642 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2643 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02002644 }
2645
Kautuk Consul584cff52011-10-31 17:08:59 -07002646 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002647
2648out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07002649 mm_drop_all_locks(mm);
2650 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002651}
2652
2653static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2654{
Rik van Riel012f1802010-08-09 17:18:40 -07002655 if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002656 /*
2657 * The LSB of head.next can't change to 0 from under
2658 * us because we hold the mm_all_locks_mutex.
2659 *
2660 * We must however clear the bitflag before unlocking
2661 * the vma so the users using the anon_vma->head will
2662 * never see our bitflag.
2663 *
2664 * No need of atomic instructions here, head.next
2665 * can't change from under us until we release the
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07002666 * anon_vma->root->mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002667 */
2668 if (!__test_and_clear_bit(0, (unsigned long *)
Rik van Riel012f1802010-08-09 17:18:40 -07002669 &anon_vma->root->head.next))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002670 BUG();
Rik van Rielcba48b92010-08-09 17:18:38 -07002671 anon_vma_unlock(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002672 }
2673}
2674
2675static void vm_unlock_mapping(struct address_space *mapping)
2676{
2677 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2678 /*
2679 * AS_MM_ALL_LOCKS can't change to 0 from under us
2680 * because we hold the mm_all_locks_mutex.
2681 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002682 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002683 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2684 &mapping->flags))
2685 BUG();
2686 }
2687}
2688
2689/*
2690 * The mmap_sem cannot be released by the caller until
2691 * mm_drop_all_locks() returns.
2692 */
2693void mm_drop_all_locks(struct mm_struct *mm)
2694{
2695 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002696 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002697
2698 BUG_ON(down_read_trylock(&mm->mmap_sem));
2699 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2700
2701 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2702 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08002703 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2704 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002705 if (vma->vm_file && vma->vm_file->f_mapping)
2706 vm_unlock_mapping(vma->vm_file->f_mapping);
2707 }
2708
2709 mutex_unlock(&mm_all_locks_mutex);
2710}
David Howells8feae132009-01-08 12:04:47 +00002711
2712/*
2713 * initialise the VMA slab
2714 */
2715void __init mmap_init(void)
2716{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07002717 int ret;
2718
2719 ret = percpu_counter_init(&vm_committed_as, 0);
2720 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00002721}