blob: f9c97d10b873ae4a896c3a22266707653d941c90 [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
Cyril Hrubise8420a82013-04-29 15:08:33 -07009#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070011#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/mm.h>
13#include <linux/shm.h>
14#include <linux/mman.h>
15#include <linux/pagemap.h>
16#include <linux/swap.h>
17#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/init.h>
20#include <linux/file.h>
21#include <linux/fs.h>
22#include <linux/personality.h>
23#include <linux/security.h>
24#include <linux/hugetlb.h>
25#include <linux/profile.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/mempolicy.h>
29#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070030#include <linux/mmu_notifier.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020031#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040032#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080033#include <linux/khugepaged.h>
Srikar Dronamraju2b144492012-02-09 14:56:42 +053034#include <linux/uprobes.h>
Michel Lespinassed3737182012-12-11 16:01:38 -080035#include <linux/rbtree_augmented.h>
Clark Williamscf4aebc22013-02-07 09:46:59 -060036#include <linux/sched/sysctl.h>
Andrew Shewmaker16408792013-04-29 15:08:12 -070037#include <linux/notifier.h>
38#include <linux/memory.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#include <asm/uaccess.h>
41#include <asm/cacheflush.h>
42#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020043#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Jan Beulich42b77722008-07-23 21:27:10 -070045#include "internal.h"
46
Kirill Korotaev3a459752006-09-07 14:17:04 +040047#ifndef arch_mmap_check
48#define arch_mmap_check(addr, len, flags) (0)
49#endif
50
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080051#ifndef arch_rebalance_pgtables
52#define arch_rebalance_pgtables(addr, len) (addr)
53#endif
54
Hugh Dickinse0da3822005-04-19 13:29:15 -070055static void unmap_region(struct mm_struct *mm,
56 struct vm_area_struct *vma, struct vm_area_struct *prev,
57 unsigned long start, unsigned long end);
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/* 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;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070090unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070091unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -070092/*
93 * Make sure vm_committed_as in one cacheline and not cacheline shared with
94 * other variables. It can be updated by several CPUs frequently.
95 */
96struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -080099 * The global memory commitment made in the system can be a metric
100 * that can be used to drive ballooning decisions when Linux is hosted
101 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
102 * balancing memory across competing virtual machines that are hosted.
103 * Several metrics drive this policy engine including the guest reported
104 * memory commitment.
105 */
106unsigned long vm_memory_committed(void)
107{
108 return percpu_counter_read_positive(&vm_committed_as);
109}
110EXPORT_SYMBOL_GPL(vm_memory_committed);
111
112/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * Check that a process has enough memory to allocate a new virtual
114 * mapping. 0 means there is enough memory for the allocation to
115 * succeed and -ENOMEM implies there is not.
116 *
117 * We currently support three overcommit policies, which are set via the
118 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
119 *
120 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
121 * Additional code 2002 Jul 20 by Robert Love.
122 *
123 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
124 *
125 * Note this is a helper function intended to be used by LSMs which
126 * wish to use this logic.
127 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700128int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700130 unsigned long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 vm_acct_memory(pages);
133
134 /*
135 * Sometimes we want to use more memory than we have
136 */
137 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
138 return 0;
139
140 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700141 free = global_page_state(NR_FREE_PAGES);
142 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Dmitry Finkc15bef32011-07-25 17:12:19 -0700144 /*
145 * shmem pages shouldn't be counted as free in this
146 * case, they can't be purged, only swapped out, and
147 * that won't affect the overall amount of available
148 * memory in the system.
149 */
150 free -= global_page_state(NR_SHMEM);
151
Shaohua Liec8acf22013-02-22 16:34:38 -0800152 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154 /*
155 * Any slabs which are created with the
156 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
157 * which are reclaimable, under pressure. The dentry
158 * cache and most inode caches should fall into this
159 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700160 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700163 * Leave reserved pages. The pages are not for anonymous pages.
164 */
165 if (free <= totalreserve_pages)
166 goto error;
167 else
168 free -= totalreserve_pages;
169
170 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700171 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 */
173 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700174 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 if (free > pages)
177 return 0;
178
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700179 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 }
181
182 allowed = (totalram_pages - hugetlb_total_pages())
183 * sysctl_overcommit_ratio / 100;
184 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700185 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 */
187 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700188 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 allowed += total_swap_pages;
190
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700191 /*
192 * Don't let a single process grow so big a user can't recover
193 */
194 if (mm) {
195 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
196 allowed -= min(mm->total_vm / 32, reserve);
197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700199 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700201error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 vm_unacct_memory(pages);
203
204 return -ENOMEM;
205}
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700208 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 */
210static void __remove_shared_vm_struct(struct vm_area_struct *vma,
211 struct file *file, struct address_space *mapping)
212{
213 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500214 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (vma->vm_flags & VM_SHARED)
216 mapping->i_mmap_writable--;
217
218 flush_dcache_mmap_lock(mapping);
219 if (unlikely(vma->vm_flags & VM_NONLINEAR))
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700220 list_del_init(&vma->shared.nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700222 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 flush_dcache_mmap_unlock(mapping);
224}
225
226/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700227 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700228 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700230void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 struct file *file = vma->vm_file;
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 if (file) {
235 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700236 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700238 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700240}
241
242/*
243 * Close a vm structure and free it, returning the next.
244 */
245static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
246{
247 struct vm_area_struct *next = vma->vm_next;
248
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700249 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 if (vma->vm_ops && vma->vm_ops->close)
251 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700252 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700253 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700254 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700256 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700259static unsigned long do_brk(unsigned long addr, unsigned long len);
260
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100261SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
263 unsigned long rlim, retval;
264 unsigned long newbrk, oldbrk;
265 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700266 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800267 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269 down_write(&mm->mmap_sem);
270
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700271#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800272 /*
273 * CONFIG_COMPAT_BRK can still be overridden by setting
274 * randomize_va_space to 2, which will still cause mm->start_brk
275 * to be arbitrarily shifted
276 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700277 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800278 min_brk = mm->start_brk;
279 else
280 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700281#else
282 min_brk = mm->start_brk;
283#endif
284 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700286
287 /*
288 * Check against rlimit here. If this check is done later after the test
289 * of oldbrk with newbrk then it can escape the test and let the data
290 * segment grow beyond its set limit the in case where the limit is
291 * not page aligned -Ram Gupta
292 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800293 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100294 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
295 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700296 goto out;
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 newbrk = PAGE_ALIGN(brk);
299 oldbrk = PAGE_ALIGN(mm->brk);
300 if (oldbrk == newbrk)
301 goto set_brk;
302
303 /* Always allow shrinking brk. */
304 if (brk <= mm->brk) {
305 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
306 goto set_brk;
307 goto out;
308 }
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 /* Check against existing mmap mappings. */
311 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
312 goto out;
313
314 /* Ok, looks good - let it rip. */
315 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
316 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318set_brk:
319 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800320 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
321 up_write(&mm->mmap_sem);
322 if (populate)
323 mm_populate(oldbrk, newbrk - oldbrk);
324 return brk;
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326out:
327 retval = mm->brk;
328 up_write(&mm->mmap_sem);
329 return retval;
330}
331
Michel Lespinassed3737182012-12-11 16:01:38 -0800332static long vma_compute_subtree_gap(struct vm_area_struct *vma)
333{
334 unsigned long max, subtree_gap;
335 max = vma->vm_start;
336 if (vma->vm_prev)
337 max -= vma->vm_prev->vm_end;
338 if (vma->vm_rb.rb_left) {
339 subtree_gap = rb_entry(vma->vm_rb.rb_left,
340 struct vm_area_struct, vm_rb)->rb_subtree_gap;
341 if (subtree_gap > max)
342 max = subtree_gap;
343 }
344 if (vma->vm_rb.rb_right) {
345 subtree_gap = rb_entry(vma->vm_rb.rb_right,
346 struct vm_area_struct, vm_rb)->rb_subtree_gap;
347 if (subtree_gap > max)
348 max = subtree_gap;
349 }
350 return max;
351}
352
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700353#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354static int browse_rb(struct rb_root *root)
355{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800356 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 struct rb_node *nd, *pn = NULL;
358 unsigned long prev = 0, pend = 0;
359
360 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
361 struct vm_area_struct *vma;
362 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800363 if (vma->vm_start < prev) {
364 printk("vm_start %lx prev %lx\n", vma->vm_start, prev);
365 bug = 1;
366 }
367 if (vma->vm_start < pend) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800369 bug = 1;
370 }
371 if (vma->vm_start > vma->vm_end) {
372 printk("vm_end %lx < vm_start %lx\n",
373 vma->vm_end, vma->vm_start);
374 bug = 1;
375 }
376 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
377 printk("free gap %lx, correct %lx\n",
378 vma->rb_subtree_gap,
379 vma_compute_subtree_gap(vma));
380 bug = 1;
381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 i++;
383 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800384 prev = vma->vm_start;
385 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 }
387 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800388 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800390 if (i != j) {
391 printk("backwards %d, forwards %d\n", j, i);
392 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800394 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Michel Lespinassed3737182012-12-11 16:01:38 -0800397static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
398{
399 struct rb_node *nd;
400
401 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
402 struct vm_area_struct *vma;
403 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
404 BUG_ON(vma != ignore &&
405 vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407}
408
409void validate_mm(struct mm_struct *mm)
410{
411 int bug = 0;
412 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800413 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700414 struct vm_area_struct *vma = mm->mmap;
415 while (vma) {
416 struct anon_vma_chain *avc;
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800417 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700418 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
419 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800420 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800421 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700422 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 i++;
424 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800425 if (i != mm->map_count) {
426 printk("map_count %d vm_next %d\n", mm->map_count, i);
427 bug = 1;
428 }
429 if (highest_address != mm->highest_vm_end) {
430 printk("mm->highest_vm_end %lx, found %lx\n",
431 mm->highest_vm_end, highest_address);
432 bug = 1;
433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800435 if (i != mm->map_count) {
436 printk("map_count %d rb %d\n", mm->map_count, i);
437 bug = 1;
438 }
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200439 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
441#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800442#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443#define validate_mm(mm) do { } while (0)
444#endif
445
Michel Lespinassed3737182012-12-11 16:01:38 -0800446RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
447 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
448
449/*
450 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
451 * vma->vm_prev->vm_end values changed, without modifying the vma's position
452 * in the rbtree.
453 */
454static void vma_gap_update(struct vm_area_struct *vma)
455{
456 /*
457 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
458 * function that does exacltly what we want.
459 */
460 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
461}
462
463static inline void vma_rb_insert(struct vm_area_struct *vma,
464 struct rb_root *root)
465{
466 /* All rb_subtree_gap values must be consistent prior to insertion */
467 validate_mm_rb(root, NULL);
468
469 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
470}
471
472static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
473{
474 /*
475 * All rb_subtree_gap values must be consistent prior to erase,
476 * with the possible exception of the vma being erased.
477 */
478 validate_mm_rb(root, vma);
479
480 /*
481 * Note rb_erase_augmented is a fairly large inline function,
482 * so make sure we instantiate it only once with our desired
483 * augmented rbtree callbacks.
484 */
485 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
486}
487
Michel Lespinassebf181b92012-10-08 16:31:39 -0700488/*
489 * vma has some anon_vma assigned, and is already inserted on that
490 * anon_vma's interval trees.
491 *
492 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
493 * vma must be removed from the anon_vma's interval trees using
494 * anon_vma_interval_tree_pre_update_vma().
495 *
496 * After the update, the vma will be reinserted using
497 * anon_vma_interval_tree_post_update_vma().
498 *
499 * The entire update must be protected by exclusive mmap_sem and by
500 * the root anon_vma's mutex.
501 */
502static inline void
503anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
504{
505 struct anon_vma_chain *avc;
506
507 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
508 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
509}
510
511static inline void
512anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
513{
514 struct anon_vma_chain *avc;
515
516 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
517 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
518}
519
Hugh Dickins6597d782012-10-08 16:29:07 -0700520static int find_vma_links(struct mm_struct *mm, unsigned long addr,
521 unsigned long end, struct vm_area_struct **pprev,
522 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
Hugh Dickins6597d782012-10-08 16:29:07 -0700524 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
526 __rb_link = &mm->mm_rb.rb_node;
527 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
529 while (*__rb_link) {
530 struct vm_area_struct *vma_tmp;
531
532 __rb_parent = *__rb_link;
533 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
534
535 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700536 /* Fail if an existing vma overlaps the area */
537 if (vma_tmp->vm_start < end)
538 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 __rb_link = &__rb_parent->rb_left;
540 } else {
541 rb_prev = __rb_parent;
542 __rb_link = &__rb_parent->rb_right;
543 }
544 }
545
546 *pprev = NULL;
547 if (rb_prev)
548 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
549 *rb_link = __rb_link;
550 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700551 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
Cyril Hrubise8420a82013-04-29 15:08:33 -0700554static unsigned long count_vma_pages_range(struct mm_struct *mm,
555 unsigned long addr, unsigned long end)
556{
557 unsigned long nr_pages = 0;
558 struct vm_area_struct *vma;
559
560 /* Find first overlaping mapping */
561 vma = find_vma_intersection(mm, addr, end);
562 if (!vma)
563 return 0;
564
565 nr_pages = (min(end, vma->vm_end) -
566 max(addr, vma->vm_start)) >> PAGE_SHIFT;
567
568 /* Iterate over the rest of the overlaps */
569 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
570 unsigned long overlap_len;
571
572 if (vma->vm_start > end)
573 break;
574
575 overlap_len = min(end, vma->vm_end) - vma->vm_start;
576 nr_pages += overlap_len >> PAGE_SHIFT;
577 }
578
579 return nr_pages;
580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
583 struct rb_node **rb_link, struct rb_node *rb_parent)
584{
Michel Lespinassed3737182012-12-11 16:01:38 -0800585 /* Update tracking information for the gap following the new vma. */
586 if (vma->vm_next)
587 vma_gap_update(vma->vm_next);
588 else
589 mm->highest_vm_end = vma->vm_end;
590
591 /*
592 * vma->vm_prev wasn't known when we followed the rbtree to find the
593 * correct insertion point for that vma. As a result, we could not
594 * update the vma vm_rb parents rb_subtree_gap values on the way down.
595 * So, we first insert the vma with a zero rb_subtree_gap value
596 * (to be consistent with what we did on the way down), and then
597 * immediately update the gap to the correct value. Finally we
598 * rebalance the rbtree after all augmented values have been set.
599 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800601 vma->rb_subtree_gap = 0;
602 vma_gap_update(vma);
603 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700606static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
ZhenwenXu48aae422009-01-06 14:40:21 -0800608 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 file = vma->vm_file;
611 if (file) {
612 struct address_space *mapping = file->f_mapping;
613
614 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500615 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (vma->vm_flags & VM_SHARED)
617 mapping->i_mmap_writable++;
618
619 flush_dcache_mmap_lock(mapping);
620 if (unlikely(vma->vm_flags & VM_NONLINEAR))
621 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
622 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700623 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 flush_dcache_mmap_unlock(mapping);
625 }
626}
627
628static void
629__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
630 struct vm_area_struct *prev, struct rb_node **rb_link,
631 struct rb_node *rb_parent)
632{
633 __vma_link_list(mm, vma, prev, rb_parent);
634 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
637static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
638 struct vm_area_struct *prev, struct rb_node **rb_link,
639 struct rb_node *rb_parent)
640{
641 struct address_space *mapping = NULL;
642
643 if (vma->vm_file)
644 mapping = vma->vm_file->f_mapping;
645
Peter Zijlstra97a89412011-05-24 17:12:04 -0700646 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700647 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649 __vma_link(mm, vma, prev, rb_link, rb_parent);
650 __vma_link_file(vma);
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700653 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
655 mm->map_count++;
656 validate_mm(mm);
657}
658
659/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700660 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700661 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800663static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Hugh Dickins6597d782012-10-08 16:29:07 -0700665 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800666 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Hugh Dickins6597d782012-10-08 16:29:07 -0700668 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
669 &prev, &rb_link, &rb_parent))
670 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 __vma_link(mm, vma, prev, rb_link, rb_parent);
672 mm->map_count++;
673}
674
675static inline void
676__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
677 struct vm_area_struct *prev)
678{
Michel Lespinassed3737182012-12-11 16:01:38 -0800679 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700680
Michel Lespinassed3737182012-12-11 16:01:38 -0800681 vma_rb_erase(vma, &mm->mm_rb);
682 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700683 if (next)
684 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 if (mm->mmap_cache == vma)
686 mm->mmap_cache = prev;
687}
688
689/*
690 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
691 * is already present in an i_mmap tree without adjusting the tree.
692 * The following helper function should be used when such adjustments
693 * are necessary. The "insert" vma (if any) is to be inserted
694 * before we drop the necessary locks.
695 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800696int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
698{
699 struct mm_struct *mm = vma->vm_mm;
700 struct vm_area_struct *next = vma->vm_next;
701 struct vm_area_struct *importer = NULL;
702 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700703 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700704 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800706 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 long adjust_next = 0;
708 int remove_next = 0;
709
710 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700711 struct vm_area_struct *exporter = NULL;
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (end >= next->vm_end) {
714 /*
715 * vma expands, overlapping all the next, and
716 * perhaps the one after too (mprotect case 6).
717 */
718again: remove_next = 1 + (end > next->vm_end);
719 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700720 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 importer = vma;
722 } else if (end > next->vm_start) {
723 /*
724 * vma expands, overlapping part of the next:
725 * mprotect case 5 shifting the boundary up.
726 */
727 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700728 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 importer = vma;
730 } else if (end < vma->vm_end) {
731 /*
732 * vma shrinks, and !insert tells it's not
733 * split_vma inserting another: so it must be
734 * mprotect case 4 shifting the boundary down.
735 */
736 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700737 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 importer = next;
739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Rik van Riel5beb4932010-03-05 13:42:07 -0800741 /*
742 * Easily overlooked: when mprotect shifts the boundary,
743 * make sure the expanding vma has anon_vma set if the
744 * shrinking vma had, to cover any anon pages imported.
745 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700746 if (exporter && exporter->anon_vma && !importer->anon_vma) {
747 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800748 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700749 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800750 }
751 }
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 if (file) {
754 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530755 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530757 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530758
759 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530760 uprobe_munmap(next, next->vm_start,
761 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530762 }
763
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700764 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700767 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 * are visible to arm/parisc __flush_dcache_page
769 * throughout; but we cannot insert into address
770 * space until vma start or end is updated.
771 */
772 __vma_link_file(insert);
773 }
774 }
775
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800776 vma_adjust_trans_huge(vma, start, end, adjust_next);
777
Michel Lespinassebf181b92012-10-08 16:31:39 -0700778 anon_vma = vma->anon_vma;
779 if (!anon_vma && adjust_next)
780 anon_vma = next->anon_vma;
781 if (anon_vma) {
Michel Lespinasseca42b262012-10-08 16:30:01 -0700782 VM_BUG_ON(adjust_next && next->anon_vma &&
783 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000784 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700785 anon_vma_interval_tree_pre_update_vma(vma);
786 if (adjust_next)
787 anon_vma_interval_tree_pre_update_vma(next);
788 }
Rik van Riel012f18002010-08-09 17:18:40 -0700789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 if (root) {
791 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700792 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700794 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
796
Michel Lespinassed3737182012-12-11 16:01:38 -0800797 if (start != vma->vm_start) {
798 vma->vm_start = start;
799 start_changed = true;
800 }
801 if (end != vma->vm_end) {
802 vma->vm_end = end;
803 end_changed = true;
804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 vma->vm_pgoff = pgoff;
806 if (adjust_next) {
807 next->vm_start += adjust_next << PAGE_SHIFT;
808 next->vm_pgoff += adjust_next;
809 }
810
811 if (root) {
812 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700813 vma_interval_tree_insert(next, root);
814 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 flush_dcache_mmap_unlock(mapping);
816 }
817
818 if (remove_next) {
819 /*
820 * vma_merge has merged next into vma, and needs
821 * us to remove next before dropping the locks.
822 */
823 __vma_unlink(mm, next, vma);
824 if (file)
825 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 } else if (insert) {
827 /*
828 * split_vma has split insert from vma, and needs
829 * us to insert it before dropping the locks
830 * (it may either follow vma or precede it).
831 */
832 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800833 } else {
834 if (start_changed)
835 vma_gap_update(vma);
836 if (end_changed) {
837 if (!next)
838 mm->highest_vm_end = end;
839 else if (!adjust_next)
840 vma_gap_update(next);
841 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 }
843
Michel Lespinassebf181b92012-10-08 16:31:39 -0700844 if (anon_vma) {
845 anon_vma_interval_tree_post_update_vma(vma);
846 if (adjust_next)
847 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800848 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700851 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530853 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100854 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530855
856 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100857 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530858 }
859
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700861 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530862 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700864 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800865 if (next->anon_vma)
866 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700868 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 kmem_cache_free(vm_area_cachep, next);
870 /*
871 * In mprotect's case 6 (see comments on vma_merge),
872 * we must remove another next too. It would clutter
873 * up the code too much to do both in one go.
874 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800875 next = vma->vm_next;
876 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800878 else if (next)
879 vma_gap_update(next);
880 else
881 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530883 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100884 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
886 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800887
888 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889}
890
891/*
892 * If the vma has a ->close operation then the driver probably needs to release
893 * per-vma resources, so we don't attempt to merge those.
894 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895static inline int is_mergeable_vma(struct vm_area_struct *vma,
896 struct file *file, unsigned long vm_flags)
897{
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700898 if (vma->vm_flags ^ vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 return 0;
900 if (vma->vm_file != file)
901 return 0;
902 if (vma->vm_ops && vma->vm_ops->close)
903 return 0;
904 return 1;
905}
906
907static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700908 struct anon_vma *anon_vma2,
909 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
Shaohua Li965f55d2011-05-24 17:11:20 -0700911 /*
912 * The list_is_singular() test is to avoid merging VMA cloned from
913 * parents. This can improve scalability caused by anon_vma lock.
914 */
915 if ((!anon_vma1 || !anon_vma2) && (!vma ||
916 list_is_singular(&vma->anon_vma_chain)))
917 return 1;
918 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919}
920
921/*
922 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
923 * in front of (at a lower virtual address and file offset than) the vma.
924 *
925 * We cannot merge two vmas if they have differently assigned (non-NULL)
926 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
927 *
928 * We don't check here for the merged mmap wrapping around the end of pagecache
929 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
930 * wrap, nor mmaps which cover the final page at index -1UL.
931 */
932static int
933can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
934 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
935{
936 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700937 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 if (vma->vm_pgoff == vm_pgoff)
939 return 1;
940 }
941 return 0;
942}
943
944/*
945 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
946 * beyond (at a higher virtual address and file offset than) the vma.
947 *
948 * We cannot merge two vmas if they have differently assigned (non-NULL)
949 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
950 */
951static int
952can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
953 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
954{
955 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700956 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700958 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
960 return 1;
961 }
962 return 0;
963}
964
965/*
966 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
967 * whether that can be merged with its predecessor or its successor.
968 * Or both (it neatly fills a hole).
969 *
970 * In most cases - when called for mmap, brk or mremap - [addr,end) is
971 * certain not to be mapped by the time vma_merge is called; but when
972 * called for mprotect, it is certain to be already mapped (either at
973 * an offset within prev, or at the start of next), and the flags of
974 * this area are about to be changed to vm_flags - and the no-change
975 * case has already been eliminated.
976 *
977 * The following mprotect cases have to be considered, where AAAA is
978 * the area passed down from mprotect_fixup, never extending beyond one
979 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
980 *
981 * AAAA AAAA AAAA AAAA
982 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
983 * cannot merge might become might become might become
984 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
985 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
986 * mremap move: PPPPNNNNNNNN 8
987 * AAAA
988 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
989 * might become case 1 below case 2 below case 3 below
990 *
991 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
992 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
993 */
994struct vm_area_struct *vma_merge(struct mm_struct *mm,
995 struct vm_area_struct *prev, unsigned long addr,
996 unsigned long end, unsigned long vm_flags,
997 struct anon_vma *anon_vma, struct file *file,
998 pgoff_t pgoff, struct mempolicy *policy)
999{
1000 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1001 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001002 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004 /*
1005 * We later require that vma->vm_flags == vm_flags,
1006 * so this tests vma->vm_flags & VM_SPECIAL, too.
1007 */
1008 if (vm_flags & VM_SPECIAL)
1009 return NULL;
1010
1011 if (prev)
1012 next = prev->vm_next;
1013 else
1014 next = mm->mmap;
1015 area = next;
1016 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1017 next = next->vm_next;
1018
1019 /*
1020 * Can it merge with the predecessor?
1021 */
1022 if (prev && prev->vm_end == addr &&
1023 mpol_equal(vma_policy(prev), policy) &&
1024 can_vma_merge_after(prev, vm_flags,
1025 anon_vma, file, pgoff)) {
1026 /*
1027 * OK, it can. Can we now merge in the successor as well?
1028 */
1029 if (next && end == next->vm_start &&
1030 mpol_equal(policy, vma_policy(next)) &&
1031 can_vma_merge_before(next, vm_flags,
1032 anon_vma, file, pgoff+pglen) &&
1033 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001034 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001036 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 next->vm_end, prev->vm_pgoff, NULL);
1038 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001039 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001041 if (err)
1042 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001043 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 return prev;
1045 }
1046
1047 /*
1048 * Can this new request be merged in front of next?
1049 */
1050 if (next && end == next->vm_start &&
1051 mpol_equal(policy, vma_policy(next)) &&
1052 can_vma_merge_before(next, vm_flags,
1053 anon_vma, file, pgoff+pglen)) {
1054 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001055 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 addr, prev->vm_pgoff, NULL);
1057 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001058 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001060 if (err)
1061 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001062 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 return area;
1064 }
1065
1066 return NULL;
1067}
1068
1069/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001070 * Rough compatbility check to quickly see if it's even worth looking
1071 * at sharing an anon_vma.
1072 *
1073 * They need to have the same vm_file, and the flags can only differ
1074 * in things that mprotect may change.
1075 *
1076 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1077 * we can merge the two vma's. For example, we refuse to merge a vma if
1078 * there is a vm_ops->close() function, because that indicates that the
1079 * driver is doing some kind of reference counting. But that doesn't
1080 * really matter for the anon_vma sharing case.
1081 */
1082static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1083{
1084 return a->vm_end == b->vm_start &&
1085 mpol_equal(vma_policy(a), vma_policy(b)) &&
1086 a->vm_file == b->vm_file &&
1087 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
1088 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1089}
1090
1091/*
1092 * Do some basic sanity checking to see if we can re-use the anon_vma
1093 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1094 * the same as 'old', the other will be the new one that is trying
1095 * to share the anon_vma.
1096 *
1097 * NOTE! This runs with mm_sem held for reading, so it is possible that
1098 * the anon_vma of 'old' is concurrently in the process of being set up
1099 * by another page fault trying to merge _that_. But that's ok: if it
1100 * is being set up, that automatically means that it will be a singleton
1101 * acceptable for merging, so we can do all of this optimistically. But
1102 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1103 *
1104 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1105 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1106 * is to return an anon_vma that is "complex" due to having gone through
1107 * a fork).
1108 *
1109 * We also make sure that the two vma's are compatible (adjacent,
1110 * and with the same memory policies). That's all stable, even with just
1111 * a read lock on the mm_sem.
1112 */
1113static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1114{
1115 if (anon_vma_compatible(a, b)) {
1116 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1117
1118 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1119 return anon_vma;
1120 }
1121 return NULL;
1122}
1123
1124/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1126 * neighbouring vmas for a suitable anon_vma, before it goes off
1127 * to allocate a new anon_vma. It checks because a repetitive
1128 * sequence of mprotects and faults may otherwise lead to distinct
1129 * anon_vmas being allocated, preventing vma merge in subsequent
1130 * mprotect.
1131 */
1132struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1133{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001134 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 near = vma->vm_next;
1138 if (!near)
1139 goto try_prev;
1140
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001141 anon_vma = reusable_anon_vma(near, vma, near);
1142 if (anon_vma)
1143 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001145 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 if (!near)
1147 goto none;
1148
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001149 anon_vma = reusable_anon_vma(near, near, vma);
1150 if (anon_vma)
1151 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152none:
1153 /*
1154 * There's no absolute need to look only at touching neighbours:
1155 * we could search further afield for "compatible" anon_vmas.
1156 * But it would probably just be a waste of time searching,
1157 * or lead to too many vmas hanging off the same anon_vma.
1158 * We're trying to allow mprotect remerging later on,
1159 * not trying to minimize memory used for anon_vmas.
1160 */
1161 return NULL;
1162}
1163
1164#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001165void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 struct file *file, long pages)
1167{
1168 const unsigned long stack_flags
1169 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1170
Huang Shijie44de9d02012-07-31 16:41:49 -07001171 mm->total_vm += pages;
1172
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 if (file) {
1174 mm->shared_vm += pages;
1175 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1176 mm->exec_vm += pages;
1177 } else if (flags & stack_flags)
1178 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179}
1180#endif /* CONFIG_PROC_FS */
1181
1182/*
Al Viro40401532012-02-13 03:58:52 +00001183 * If a hint addr is less than mmap_min_addr change hint to be as
1184 * low as possible but still greater than mmap_min_addr
1185 */
1186static inline unsigned long round_hint_to_min(unsigned long hint)
1187{
1188 hint &= PAGE_MASK;
1189 if (((void *)hint != NULL) &&
1190 (hint < mmap_min_addr))
1191 return PAGE_ALIGN(mmap_min_addr);
1192 return hint;
1193}
1194
1195/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001196 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 */
1198
Al Viroe3fc6292012-05-30 20:08:42 -04001199unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001201 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001202 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203{
1204 struct mm_struct * mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 struct inode *inode;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001206 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
Michel Lespinasse41badc12013-02-22 16:32:47 -08001208 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 /*
1211 * Does the application expect PROT_READ to imply PROT_EXEC?
1212 *
1213 * (the exception is when the underlying filesystem is noexec
1214 * mounted, in which case we dont add PROT_EXEC.)
1215 */
1216 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001217 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 prot |= PROT_EXEC;
1219
1220 if (!len)
1221 return -EINVAL;
1222
Eric Paris7cd94142007-11-26 18:47:40 -05001223 if (!(flags & MAP_FIXED))
1224 addr = round_hint_to_min(addr);
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 /* Careful about overflows.. */
1227 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001228 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 return -ENOMEM;
1230
1231 /* offset overflow? */
1232 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1233 return -EOVERFLOW;
1234
1235 /* Too many mappings? */
1236 if (mm->map_count > sysctl_max_map_count)
1237 return -ENOMEM;
1238
1239 /* Obtain the address to map to. we verify (or select) it and ensure
1240 * that it represents a valid section of the address space.
1241 */
1242 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1243 if (addr & ~PAGE_MASK)
1244 return addr;
1245
1246 /* Do simple checking here so the lower-level routines won't have
1247 * to. we assume access permissions have been handled by the open
1248 * of the memory object, so we don't do any here.
1249 */
1250 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1251 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1252
Huang Shijiecdf7b342009-09-21 17:03:36 -07001253 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (!can_do_mlock())
1255 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 /* mlock MCL_FUTURE? */
1258 if (vm_flags & VM_LOCKED) {
1259 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001260 locked = len >> PAGE_SHIFT;
1261 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001262 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001263 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1265 return -EAGAIN;
1266 }
1267
Al Viro496ad9a2013-01-23 17:07:38 -05001268 inode = file ? file_inode(file) : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
1270 if (file) {
1271 switch (flags & MAP_TYPE) {
1272 case MAP_SHARED:
1273 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1274 return -EACCES;
1275
1276 /*
1277 * Make sure we don't allow writing to an append-only
1278 * file..
1279 */
1280 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1281 return -EACCES;
1282
1283 /*
1284 * Make sure there are no mandatory locks on the file.
1285 */
1286 if (locks_verify_locked(inode))
1287 return -EAGAIN;
1288
1289 vm_flags |= VM_SHARED | VM_MAYSHARE;
1290 if (!(file->f_mode & FMODE_WRITE))
1291 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1292
1293 /* fall through */
1294 case MAP_PRIVATE:
1295 if (!(file->f_mode & FMODE_READ))
1296 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001297 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001298 if (vm_flags & VM_EXEC)
1299 return -EPERM;
1300 vm_flags &= ~VM_MAYEXEC;
1301 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001302
1303 if (!file->f_op || !file->f_op->mmap)
1304 return -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 break;
1306
1307 default:
1308 return -EINVAL;
1309 }
1310 } else {
1311 switch (flags & MAP_TYPE) {
1312 case MAP_SHARED:
Tejun Heoce363942008-09-03 16:09:47 +02001313 /*
1314 * Ignore pgoff.
1315 */
1316 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 vm_flags |= VM_SHARED | VM_MAYSHARE;
1318 break;
1319 case MAP_PRIVATE:
1320 /*
1321 * Set pgoff according to addr for anon_vma.
1322 */
1323 pgoff = addr >> PAGE_SHIFT;
1324 break;
1325 default:
1326 return -EINVAL;
1327 }
1328 }
1329
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001330 /*
1331 * Set 'VM_NORESERVE' if we should not account for the
1332 * memory use of this mapping.
1333 */
1334 if (flags & MAP_NORESERVE) {
1335 /* We honor MAP_NORESERVE if allowed to overcommit */
1336 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1337 vm_flags |= VM_NORESERVE;
1338
1339 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1340 if (file && is_file_hugepages(file))
1341 vm_flags |= VM_NORESERVE;
1342 }
1343
1344 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001345 if (!IS_ERR_VALUE(addr) &&
1346 ((vm_flags & VM_LOCKED) ||
1347 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001348 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001349 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001350}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001351
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001352SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1353 unsigned long, prot, unsigned long, flags,
1354 unsigned long, fd, unsigned long, pgoff)
1355{
1356 struct file *file = NULL;
1357 unsigned long retval = -EBADF;
1358
1359 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001360 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001361 file = fget(fd);
1362 if (!file)
1363 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001364 if (is_file_hugepages(file))
1365 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001366 retval = -EINVAL;
1367 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1368 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001369 } else if (flags & MAP_HUGETLB) {
1370 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001371 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001372
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001373 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001374 if (!hs)
1375 return -EINVAL;
1376
1377 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001378 /*
1379 * VM_NORESERVE is used because the reservations will be
1380 * taken when vm_ops->mmap() is called
1381 * A dummy user value is used because we are not locking
1382 * memory so no accounting is necessary
1383 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001384 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001385 VM_NORESERVE,
1386 &user, HUGETLB_ANONHUGE_INODE,
1387 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001388 if (IS_ERR(file))
1389 return PTR_ERR(file);
1390 }
1391
1392 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1393
Al Viroeb36c582012-05-30 20:17:35 -04001394 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001395out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001396 if (file)
1397 fput(file);
1398out:
1399 return retval;
1400}
1401
Christoph Hellwiga4679372010-03-10 15:21:15 -08001402#ifdef __ARCH_WANT_SYS_OLD_MMAP
1403struct mmap_arg_struct {
1404 unsigned long addr;
1405 unsigned long len;
1406 unsigned long prot;
1407 unsigned long flags;
1408 unsigned long fd;
1409 unsigned long offset;
1410};
1411
1412SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1413{
1414 struct mmap_arg_struct a;
1415
1416 if (copy_from_user(&a, arg, sizeof(a)))
1417 return -EFAULT;
1418 if (a.offset & ~PAGE_MASK)
1419 return -EINVAL;
1420
1421 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1422 a.offset >> PAGE_SHIFT);
1423}
1424#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1425
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001426/*
1427 * Some shared mappigns will want the pages marked read-only
1428 * to track write events. If so, we'll downgrade vm_page_prot
1429 * to the private version (using protection_map[] without the
1430 * VM_SHARED bit).
1431 */
1432int vma_wants_writenotify(struct vm_area_struct *vma)
1433{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001434 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001435
1436 /* If it was private or non-writable, the write bit is already clear */
1437 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1438 return 0;
1439
1440 /* The backer wishes to know when pages are first written to? */
1441 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1442 return 1;
1443
1444 /* The open routine did something to the protections already? */
1445 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001446 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001447 return 0;
1448
1449 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001450 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001451 return 0;
1452
1453 /* Can the mapping track the dirty pages? */
1454 return vma->vm_file && vma->vm_file->f_mapping &&
1455 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1456}
1457
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001458/*
1459 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001460 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001461 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001462static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001463{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001464 /*
1465 * hugetlb has its own accounting separate from the core VM
1466 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1467 */
1468 if (file && is_file_hugepages(file))
1469 return 0;
1470
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001471 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1472}
1473
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001474unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001475 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001476{
1477 struct mm_struct *mm = current->mm;
1478 struct vm_area_struct *vma, *prev;
1479 int correct_wcount = 0;
1480 int error;
1481 struct rb_node **rb_link, *rb_parent;
1482 unsigned long charged = 0;
Al Viro496ad9a2013-01-23 17:07:38 -05001483 struct inode *inode = file ? file_inode(file) : NULL;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001484
Cyril Hrubise8420a82013-04-29 15:08:33 -07001485 /* Check against address space limit. */
1486 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1487 unsigned long nr_pages;
1488
1489 /*
1490 * MAP_FIXED may remove pages of mappings that intersects with
1491 * requested mapping. Account for the pages it would unmap.
1492 */
1493 if (!(vm_flags & MAP_FIXED))
1494 return -ENOMEM;
1495
1496 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1497
1498 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1499 return -ENOMEM;
1500 }
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 /* Clear old maps */
1503 error = -ENOMEM;
1504munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001505 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 if (do_munmap(mm, addr, len))
1507 return -ENOMEM;
1508 goto munmap_back;
1509 }
1510
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001511 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001512 * Private writable mapping: check memory availability
1513 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001514 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001515 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001516 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001517 return -ENOMEM;
1518 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 }
1520
1521 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001522 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001524 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1525 if (vma)
1526 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 /*
1529 * Determine the object being mapped and call the appropriate
1530 * specific mapper. the address has already been validated, but
1531 * not unmapped, but the maps are removed from the list.
1532 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001533 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 if (!vma) {
1535 error = -ENOMEM;
1536 goto unacct_error;
1537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
1539 vma->vm_mm = mm;
1540 vma->vm_start = addr;
1541 vma->vm_end = addr + len;
1542 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001543 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001545 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Hugh Dickinsce8fea72012-03-06 12:28:52 -08001547 error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1551 goto free_vma;
1552 if (vm_flags & VM_DENYWRITE) {
1553 error = deny_write_access(file);
1554 if (error)
1555 goto free_vma;
1556 correct_wcount = 1;
1557 }
Al Virocb0942b2012-08-27 14:48:26 -04001558 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 error = file->f_op->mmap(file, vma);
1560 if (error)
1561 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001562
1563 /* Can addr have changed??
1564 *
1565 * Answer: Yes, several device drivers can do it in their
1566 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001567 * Bug: If addr is changed, prev, rb_link, rb_parent should
1568 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001569 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001570 WARN_ON_ONCE(addr != vma->vm_start);
1571
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001572 addr = vma->vm_start;
1573 pgoff = vma->vm_pgoff;
1574 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 } else if (vm_flags & VM_SHARED) {
Al Viro835ee792012-03-05 06:39:47 +00001576 if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1577 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 error = shmem_zero_setup(vma);
1579 if (error)
1580 goto free_vma;
1581 }
1582
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001583 if (vma_wants_writenotify(vma)) {
1584 pgprot_t pprot = vma->vm_page_prot;
1585
1586 /* Can vma->vm_page_prot have changed??
1587 *
1588 * Answer: Yes, drivers may have changed it in their
1589 * f_op->mmap method.
1590 *
1591 * Ensures that vmas marked as uncached stay that way.
1592 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001593 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001594 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1595 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1596 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001597
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001598 vma_link(mm, vma, prev, rb_link, rb_parent);
1599 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001600
1601 /* Once vma denies write, undo our temporary denial count */
1602 if (correct_wcount)
1603 atomic_inc(&inode->i_writecount);
1604out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001605 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001606
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001607 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001609 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1610 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001611 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001612 else
1613 vma->vm_flags &= ~VM_LOCKED;
1614 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301615
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001616 if (file)
1617 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 return addr;
1620
1621unmap_and_free_vma:
1622 if (correct_wcount)
1623 atomic_inc(&inode->i_writecount);
1624 vma->vm_file = NULL;
1625 fput(file);
1626
1627 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001628 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1629 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630free_vma:
1631 kmem_cache_free(vm_area_cachep, vma);
1632unacct_error:
1633 if (charged)
1634 vm_unacct_memory(charged);
1635 return error;
1636}
1637
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001638unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1639{
1640 /*
1641 * We implement the search by looking for an rbtree node that
1642 * immediately follows a suitable gap. That is,
1643 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1644 * - gap_end = vma->vm_start >= info->low_limit + length;
1645 * - gap_end - gap_start >= length
1646 */
1647
1648 struct mm_struct *mm = current->mm;
1649 struct vm_area_struct *vma;
1650 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1651
1652 /* Adjust search length to account for worst case alignment overhead */
1653 length = info->length + info->align_mask;
1654 if (length < info->length)
1655 return -ENOMEM;
1656
1657 /* Adjust search limits by the desired length */
1658 if (info->high_limit < length)
1659 return -ENOMEM;
1660 high_limit = info->high_limit - length;
1661
1662 if (info->low_limit > high_limit)
1663 return -ENOMEM;
1664 low_limit = info->low_limit + length;
1665
1666 /* Check if rbtree root looks promising */
1667 if (RB_EMPTY_ROOT(&mm->mm_rb))
1668 goto check_highest;
1669 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1670 if (vma->rb_subtree_gap < length)
1671 goto check_highest;
1672
1673 while (true) {
1674 /* Visit left subtree if it looks promising */
1675 gap_end = vma->vm_start;
1676 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1677 struct vm_area_struct *left =
1678 rb_entry(vma->vm_rb.rb_left,
1679 struct vm_area_struct, vm_rb);
1680 if (left->rb_subtree_gap >= length) {
1681 vma = left;
1682 continue;
1683 }
1684 }
1685
1686 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1687check_current:
1688 /* Check if current node has a suitable gap */
1689 if (gap_start > high_limit)
1690 return -ENOMEM;
1691 if (gap_end >= low_limit && gap_end - gap_start >= length)
1692 goto found;
1693
1694 /* Visit right subtree if it looks promising */
1695 if (vma->vm_rb.rb_right) {
1696 struct vm_area_struct *right =
1697 rb_entry(vma->vm_rb.rb_right,
1698 struct vm_area_struct, vm_rb);
1699 if (right->rb_subtree_gap >= length) {
1700 vma = right;
1701 continue;
1702 }
1703 }
1704
1705 /* Go back up the rbtree to find next candidate node */
1706 while (true) {
1707 struct rb_node *prev = &vma->vm_rb;
1708 if (!rb_parent(prev))
1709 goto check_highest;
1710 vma = rb_entry(rb_parent(prev),
1711 struct vm_area_struct, vm_rb);
1712 if (prev == vma->vm_rb.rb_left) {
1713 gap_start = vma->vm_prev->vm_end;
1714 gap_end = vma->vm_start;
1715 goto check_current;
1716 }
1717 }
1718 }
1719
1720check_highest:
1721 /* Check highest gap, which does not precede any rbtree node */
1722 gap_start = mm->highest_vm_end;
1723 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1724 if (gap_start > high_limit)
1725 return -ENOMEM;
1726
1727found:
1728 /* We found a suitable gap. Clip it with the original low_limit. */
1729 if (gap_start < info->low_limit)
1730 gap_start = info->low_limit;
1731
1732 /* Adjust gap address to the desired alignment */
1733 gap_start += (info->align_offset - gap_start) & info->align_mask;
1734
1735 VM_BUG_ON(gap_start + info->length > info->high_limit);
1736 VM_BUG_ON(gap_start + info->length > gap_end);
1737 return gap_start;
1738}
1739
1740unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1741{
1742 struct mm_struct *mm = current->mm;
1743 struct vm_area_struct *vma;
1744 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1745
1746 /* Adjust search length to account for worst case alignment overhead */
1747 length = info->length + info->align_mask;
1748 if (length < info->length)
1749 return -ENOMEM;
1750
1751 /*
1752 * Adjust search limits by the desired length.
1753 * See implementation comment at top of unmapped_area().
1754 */
1755 gap_end = info->high_limit;
1756 if (gap_end < length)
1757 return -ENOMEM;
1758 high_limit = gap_end - length;
1759
1760 if (info->low_limit > high_limit)
1761 return -ENOMEM;
1762 low_limit = info->low_limit + length;
1763
1764 /* Check highest gap, which does not precede any rbtree node */
1765 gap_start = mm->highest_vm_end;
1766 if (gap_start <= high_limit)
1767 goto found_highest;
1768
1769 /* Check if rbtree root looks promising */
1770 if (RB_EMPTY_ROOT(&mm->mm_rb))
1771 return -ENOMEM;
1772 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1773 if (vma->rb_subtree_gap < length)
1774 return -ENOMEM;
1775
1776 while (true) {
1777 /* Visit right subtree if it looks promising */
1778 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1779 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1780 struct vm_area_struct *right =
1781 rb_entry(vma->vm_rb.rb_right,
1782 struct vm_area_struct, vm_rb);
1783 if (right->rb_subtree_gap >= length) {
1784 vma = right;
1785 continue;
1786 }
1787 }
1788
1789check_current:
1790 /* Check if current node has a suitable gap */
1791 gap_end = vma->vm_start;
1792 if (gap_end < low_limit)
1793 return -ENOMEM;
1794 if (gap_start <= high_limit && gap_end - gap_start >= length)
1795 goto found;
1796
1797 /* Visit left subtree if it looks promising */
1798 if (vma->vm_rb.rb_left) {
1799 struct vm_area_struct *left =
1800 rb_entry(vma->vm_rb.rb_left,
1801 struct vm_area_struct, vm_rb);
1802 if (left->rb_subtree_gap >= length) {
1803 vma = left;
1804 continue;
1805 }
1806 }
1807
1808 /* Go back up the rbtree to find next candidate node */
1809 while (true) {
1810 struct rb_node *prev = &vma->vm_rb;
1811 if (!rb_parent(prev))
1812 return -ENOMEM;
1813 vma = rb_entry(rb_parent(prev),
1814 struct vm_area_struct, vm_rb);
1815 if (prev == vma->vm_rb.rb_right) {
1816 gap_start = vma->vm_prev ?
1817 vma->vm_prev->vm_end : 0;
1818 goto check_current;
1819 }
1820 }
1821 }
1822
1823found:
1824 /* We found a suitable gap. Clip it with the original high_limit. */
1825 if (gap_end > info->high_limit)
1826 gap_end = info->high_limit;
1827
1828found_highest:
1829 /* Compute highest gap address at the desired alignment */
1830 gap_end -= info->length;
1831 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1832
1833 VM_BUG_ON(gap_end < info->low_limit);
1834 VM_BUG_ON(gap_end < gap_start);
1835 return gap_end;
1836}
1837
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838/* Get an address range which is currently unmapped.
1839 * For shmat() with addr=0.
1840 *
1841 * Ugly calling convention alert:
1842 * Return value with the low bits set means error value,
1843 * ie
1844 * if (ret & ~PAGE_MASK)
1845 * error = ret;
1846 *
1847 * This function "knows" that -ENOMEM has the bits set.
1848 */
1849#ifndef HAVE_ARCH_UNMAPPED_AREA
1850unsigned long
1851arch_get_unmapped_area(struct file *filp, unsigned long addr,
1852 unsigned long len, unsigned long pgoff, unsigned long flags)
1853{
1854 struct mm_struct *mm = current->mm;
1855 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001856 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
1858 if (len > TASK_SIZE)
1859 return -ENOMEM;
1860
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001861 if (flags & MAP_FIXED)
1862 return addr;
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 if (addr) {
1865 addr = PAGE_ALIGN(addr);
1866 vma = find_vma(mm, addr);
1867 if (TASK_SIZE - len >= addr &&
1868 (!vma || addr + len <= vma->vm_start))
1869 return addr;
1870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001872 info.flags = 0;
1873 info.length = len;
1874 info.low_limit = TASK_UNMAPPED_BASE;
1875 info.high_limit = TASK_SIZE;
1876 info.align_mask = 0;
1877 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878}
1879#endif
1880
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881/*
1882 * This mmap-allocator allocates new areas top-down from below the
1883 * stack's low limit (the base):
1884 */
1885#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1886unsigned long
1887arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1888 const unsigned long len, const unsigned long pgoff,
1889 const unsigned long flags)
1890{
1891 struct vm_area_struct *vma;
1892 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001893 unsigned long addr = addr0;
1894 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896 /* requested length too big for entire address space */
1897 if (len > TASK_SIZE)
1898 return -ENOMEM;
1899
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001900 if (flags & MAP_FIXED)
1901 return addr;
1902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 /* requesting a specific address */
1904 if (addr) {
1905 addr = PAGE_ALIGN(addr);
1906 vma = find_vma(mm, addr);
1907 if (TASK_SIZE - len >= addr &&
1908 (!vma || addr + len <= vma->vm_start))
1909 return addr;
1910 }
1911
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001912 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1913 info.length = len;
1914 info.low_limit = PAGE_SIZE;
1915 info.high_limit = mm->mmap_base;
1916 info.align_mask = 0;
1917 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 /*
1920 * A failed mmap() very likely causes application failure,
1921 * so fall back to the bottom-up function here. This scenario
1922 * can happen with large stack limits and large mmap()
1923 * allocations.
1924 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001925 if (addr & ~PAGE_MASK) {
1926 VM_BUG_ON(addr != -ENOMEM);
1927 info.flags = 0;
1928 info.low_limit = TASK_UNMAPPED_BASE;
1929 info.high_limit = TASK_SIZE;
1930 addr = vm_unmapped_area(&info);
1931 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
1933 return addr;
1934}
1935#endif
1936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937unsigned long
1938get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1939 unsigned long pgoff, unsigned long flags)
1940{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001941 unsigned long (*get_area)(struct file *, unsigned long,
1942 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
Al Viro9206de92009-12-03 15:23:11 -05001944 unsigned long error = arch_mmap_check(addr, len, flags);
1945 if (error)
1946 return error;
1947
1948 /* Careful about overflows.. */
1949 if (len > TASK_SIZE)
1950 return -ENOMEM;
1951
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001952 get_area = current->mm->get_unmapped_area;
1953 if (file && file->f_op && file->f_op->get_unmapped_area)
1954 get_area = file->f_op->get_unmapped_area;
1955 addr = get_area(file, addr, len, pgoff, flags);
1956 if (IS_ERR_VALUE(addr))
1957 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001959 if (addr > TASK_SIZE - len)
1960 return -ENOMEM;
1961 if (addr & ~PAGE_MASK)
1962 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001963
Al Viro9ac4ed42012-05-30 17:13:15 -04001964 addr = arch_rebalance_pgtables(addr, len);
1965 error = security_mmap_addr(addr);
1966 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967}
1968
1969EXPORT_SYMBOL(get_unmapped_area);
1970
1971/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001972struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
1974 struct vm_area_struct *vma = NULL;
1975
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001976 /* Check the cache first. */
1977 /* (Cache hit rate is typically around 35%.) */
Jan Stancekb6a9b7f2013-04-04 11:35:10 -07001978 vma = ACCESS_ONCE(mm->mmap_cache);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001979 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1980 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001982 rb_node = mm->mm_rb.rb_node;
1983 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001985 while (rb_node) {
1986 struct vm_area_struct *vma_tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001988 vma_tmp = rb_entry(rb_node,
1989 struct vm_area_struct, vm_rb);
1990
1991 if (vma_tmp->vm_end > addr) {
1992 vma = vma_tmp;
1993 if (vma_tmp->vm_start <= addr)
1994 break;
1995 rb_node = rb_node->rb_left;
1996 } else
1997 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 }
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001999 if (vma)
2000 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 }
2002 return vma;
2003}
2004
2005EXPORT_SYMBOL(find_vma);
2006
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002007/*
2008 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002009 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010struct vm_area_struct *
2011find_vma_prev(struct mm_struct *mm, unsigned long addr,
2012 struct vm_area_struct **pprev)
2013{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002014 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002016 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002017 if (vma) {
2018 *pprev = vma->vm_prev;
2019 } else {
2020 struct rb_node *rb_node = mm->mm_rb.rb_node;
2021 *pprev = NULL;
2022 while (rb_node) {
2023 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2024 rb_node = rb_node->rb_right;
2025 }
2026 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002027 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028}
2029
2030/*
2031 * Verify that the stack growth is acceptable and
2032 * update accounting. This is shared with both the
2033 * grow-up and grow-down cases.
2034 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002035static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036{
2037 struct mm_struct *mm = vma->vm_mm;
2038 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08002039 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
2041 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002042 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return -ENOMEM;
2044
2045 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08002046 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return -ENOMEM;
2048
2049 /* mlock limit tests */
2050 if (vma->vm_flags & VM_LOCKED) {
2051 unsigned long locked;
2052 unsigned long limit;
2053 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002054 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2055 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (locked > limit && !capable(CAP_IPC_LOCK))
2057 return -ENOMEM;
2058 }
2059
Adam Litke0d59a012007-01-30 14:35:39 -08002060 /* Check to ensure the stack will not grow into a hugetlb-only region */
2061 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2062 vma->vm_end - size;
2063 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2064 return -EFAULT;
2065
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 /*
2067 * Overcommit.. This must be the final test, as it will
2068 * update security statistics.
2069 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002070 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 return -ENOMEM;
2072
2073 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 if (vma->vm_flags & VM_LOCKED)
2075 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002076 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 return 0;
2078}
2079
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002080#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002082 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2083 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002085int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086{
2087 int error;
2088
2089 if (!(vma->vm_flags & VM_GROWSUP))
2090 return -EFAULT;
2091
2092 /*
2093 * We must make sure the anon_vma is allocated
2094 * so that the anon_vma locking is not a noop.
2095 */
2096 if (unlikely(anon_vma_prepare(vma)))
2097 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002098 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
2100 /*
2101 * vma->vm_start/vm_end cannot change under us because the caller
2102 * is required to hold the mmap_sem in read mode. We need the
2103 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002104 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 */
Helge Deller06b32f32006-12-19 19:28:33 +01002106 if (address < PAGE_ALIGN(address+4))
2107 address = PAGE_ALIGN(address+4);
2108 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002109 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002110 return -ENOMEM;
2111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 error = 0;
2113
2114 /* Somebody else might have raced and expanded it already */
2115 if (address > vma->vm_end) {
2116 unsigned long size, grow;
2117
2118 size = address - vma->vm_start;
2119 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2120
Hugh Dickins42c36f62011-05-09 17:44:42 -07002121 error = -ENOMEM;
2122 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2123 error = acct_stack_growth(vma, size, grow);
2124 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002125 /*
2126 * vma_gap_update() doesn't support concurrent
2127 * updates, but we only hold a shared mmap_sem
2128 * lock here, so we need to protect against
2129 * concurrent vma expansions.
2130 * vma_lock_anon_vma() doesn't help here, as
2131 * we don't guarantee that all growable vmas
2132 * in a mm share the same root anon vma.
2133 * So, we reuse mm->page_table_lock to guard
2134 * against concurrent vma expansions.
2135 */
2136 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002137 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002138 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002139 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002140 if (vma->vm_next)
2141 vma_gap_update(vma->vm_next);
2142 else
2143 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002144 spin_unlock(&vma->vm_mm->page_table_lock);
2145
Hugh Dickins42c36f62011-05-09 17:44:42 -07002146 perf_event_mmap(vma);
2147 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002150 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002151 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002152 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 return error;
2154}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002155#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2156
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157/*
2158 * vma is the first one with address < vma->vm_start. Have to extend vma.
2159 */
Michal Hockod05f3162011-05-24 17:11:44 -07002160int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002161 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162{
2163 int error;
2164
2165 /*
2166 * We must make sure the anon_vma is allocated
2167 * so that the anon_vma locking is not a noop.
2168 */
2169 if (unlikely(anon_vma_prepare(vma)))
2170 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002171
2172 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002173 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002174 if (error)
2175 return error;
2176
Rik van Rielbb4a3402010-08-09 17:18:37 -07002177 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
2179 /*
2180 * vma->vm_start/vm_end cannot change under us because the caller
2181 * is required to hold the mmap_sem in read mode. We need the
2182 * anon_vma lock to serialize against concurrent expand_stacks.
2183 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
2185 /* Somebody else might have raced and expanded it already */
2186 if (address < vma->vm_start) {
2187 unsigned long size, grow;
2188
2189 size = vma->vm_end - address;
2190 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2191
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002192 error = -ENOMEM;
2193 if (grow <= vma->vm_pgoff) {
2194 error = acct_stack_growth(vma, size, grow);
2195 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002196 /*
2197 * vma_gap_update() doesn't support concurrent
2198 * updates, but we only hold a shared mmap_sem
2199 * lock here, so we need to protect against
2200 * concurrent vma expansions.
2201 * vma_lock_anon_vma() doesn't help here, as
2202 * we don't guarantee that all growable vmas
2203 * in a mm share the same root anon vma.
2204 * So, we reuse mm->page_table_lock to guard
2205 * against concurrent vma expansions.
2206 */
2207 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002208 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002209 vma->vm_start = address;
2210 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002211 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002212 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002213 spin_unlock(&vma->vm_mm->page_table_lock);
2214
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002215 perf_event_mmap(vma);
2216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 }
2218 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002219 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002220 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002221 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 return error;
2223}
2224
Linus Torvalds09884962013-02-27 08:36:04 -08002225/*
2226 * Note how expand_stack() refuses to expand the stack all the way to
2227 * abut the next virtual mapping, *unless* that mapping itself is also
2228 * a stack mapping. We want to leave room for a guard page, after all
2229 * (the guard page itself is not added here, that is done by the
2230 * actual page faulting logic)
2231 *
2232 * This matches the behavior of the guard page logic (see mm/memory.c:
2233 * check_stack_guard_page()), which only allows the guard page to be
2234 * removed under these circumstances.
2235 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002236#ifdef CONFIG_STACK_GROWSUP
2237int expand_stack(struct vm_area_struct *vma, unsigned long address)
2238{
Linus Torvalds09884962013-02-27 08:36:04 -08002239 struct vm_area_struct *next;
2240
2241 address &= PAGE_MASK;
2242 next = vma->vm_next;
2243 if (next && next->vm_start == address + PAGE_SIZE) {
2244 if (!(next->vm_flags & VM_GROWSUP))
2245 return -ENOMEM;
2246 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002247 return expand_upwards(vma, address);
2248}
2249
2250struct vm_area_struct *
2251find_extend_vma(struct mm_struct *mm, unsigned long addr)
2252{
2253 struct vm_area_struct *vma, *prev;
2254
2255 addr &= PAGE_MASK;
2256 vma = find_vma_prev(mm, addr, &prev);
2257 if (vma && (vma->vm_start <= addr))
2258 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002259 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002260 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002261 if (prev->vm_flags & VM_LOCKED)
2262 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002263 return prev;
2264}
2265#else
2266int expand_stack(struct vm_area_struct *vma, unsigned long address)
2267{
Linus Torvalds09884962013-02-27 08:36:04 -08002268 struct vm_area_struct *prev;
2269
2270 address &= PAGE_MASK;
2271 prev = vma->vm_prev;
2272 if (prev && prev->vm_end == address) {
2273 if (!(prev->vm_flags & VM_GROWSDOWN))
2274 return -ENOMEM;
2275 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002276 return expand_downwards(vma, address);
2277}
2278
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279struct vm_area_struct *
2280find_extend_vma(struct mm_struct * mm, unsigned long addr)
2281{
2282 struct vm_area_struct * vma;
2283 unsigned long start;
2284
2285 addr &= PAGE_MASK;
2286 vma = find_vma(mm,addr);
2287 if (!vma)
2288 return NULL;
2289 if (vma->vm_start <= addr)
2290 return vma;
2291 if (!(vma->vm_flags & VM_GROWSDOWN))
2292 return NULL;
2293 start = vma->vm_start;
2294 if (expand_stack(vma, addr))
2295 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002296 if (vma->vm_flags & VM_LOCKED)
2297 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 return vma;
2299}
2300#endif
2301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002303 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002305 *
2306 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002308static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002310 unsigned long nr_accounted = 0;
2311
Hugh Dickins365e9c872005-10-29 18:16:18 -07002312 /* Update high watermark before we lower total_vm */
2313 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002315 long nrpages = vma_pages(vma);
2316
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002317 if (vma->vm_flags & VM_ACCOUNT)
2318 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002319 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002320 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002321 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002322 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 validate_mm(mm);
2324}
2325
2326/*
2327 * Get rid of page table information in the indicated region.
2328 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002329 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 */
2331static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002332 struct vm_area_struct *vma, struct vm_area_struct *prev,
2333 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334{
Hugh Dickinse0da3822005-04-19 13:29:15 -07002335 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002336 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
2338 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002339 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002340 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002341 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002342 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002343 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002344 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345}
2346
2347/*
2348 * Create a list of vma's touched by the unmap, removing them from the mm's
2349 * vma list as we go..
2350 */
2351static void
2352detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2353 struct vm_area_struct *prev, unsigned long end)
2354{
2355 struct vm_area_struct **insertion_point;
2356 struct vm_area_struct *tail_vma = NULL;
2357
2358 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002359 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002361 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 mm->map_count--;
2363 tail_vma = vma;
2364 vma = vma->vm_next;
2365 } while (vma && vma->vm_start < end);
2366 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002367 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002368 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002369 vma_gap_update(vma);
2370 } else
2371 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 tail_vma->vm_next = NULL;
2373 mm->mmap_cache = NULL; /* Kill the cache. */
2374}
2375
2376/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002377 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2378 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002380static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 unsigned long addr, int new_below)
2382{
2383 struct mempolicy *pol;
2384 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08002385 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
Andi Kleena5516432008-07-23 21:27:41 -07002387 if (is_vm_hugetlb_page(vma) && (addr &
2388 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 return -EINVAL;
2390
Christoph Lametere94b1762006-12-06 20:33:17 -08002391 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002393 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394
2395 /* most fields are the same, copy all, and then fixup */
2396 *new = *vma;
2397
Rik van Riel5beb4932010-03-05 13:42:07 -08002398 INIT_LIST_HEAD(&new->anon_vma_chain);
2399
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 if (new_below)
2401 new->vm_end = addr;
2402 else {
2403 new->vm_start = addr;
2404 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2405 }
2406
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002407 pol = mpol_dup(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 if (IS_ERR(pol)) {
Rik van Riel5beb4932010-03-05 13:42:07 -08002409 err = PTR_ERR(pol);
2410 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 }
2412 vma_set_policy(new, pol);
2413
Rik van Riel5beb4932010-03-05 13:42:07 -08002414 if (anon_vma_clone(new, vma))
2415 goto out_free_mpol;
2416
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002417 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 get_file(new->vm_file);
2419
2420 if (new->vm_ops && new->vm_ops->open)
2421 new->vm_ops->open(new);
2422
2423 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002424 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2426 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002427 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
Rik van Riel5beb4932010-03-05 13:42:07 -08002429 /* Success. */
2430 if (!err)
2431 return 0;
2432
2433 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002434 if (new->vm_ops && new->vm_ops->close)
2435 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002436 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002437 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002438 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002439 out_free_mpol:
2440 mpol_put(pol);
2441 out_free_vma:
2442 kmem_cache_free(vm_area_cachep, new);
2443 out_err:
2444 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445}
2446
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002447/*
2448 * Split a vma into two pieces at address 'addr', a new vma is allocated
2449 * either for the first part or the tail.
2450 */
2451int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2452 unsigned long addr, int new_below)
2453{
2454 if (mm->map_count >= sysctl_max_map_count)
2455 return -ENOMEM;
2456
2457 return __split_vma(mm, vma, addr, new_below);
2458}
2459
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460/* Munmap is split into 2 main parts -- this part which finds
2461 * what needs doing, and the areas themselves, which do the
2462 * work. This now handles partial unmappings.
2463 * Jeremy Fitzhardinge <jeremy@goop.org>
2464 */
2465int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2466{
2467 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002468 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469
2470 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2471 return -EINVAL;
2472
2473 if ((len = PAGE_ALIGN(len)) == 0)
2474 return -EINVAL;
2475
2476 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002477 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002478 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002480 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002481 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
2483 /* if it doesn't overlap, we have nothing.. */
2484 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002485 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 return 0;
2487
2488 /*
2489 * If we need to split any vma, do it now to save pain later.
2490 *
2491 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2492 * unmapped vm_area_struct will remain in use: so lower split_vma
2493 * places tmp vma above, and higher split_vma places tmp vma below.
2494 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002495 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002496 int error;
2497
2498 /*
2499 * Make sure that map_count on return from munmap() will
2500 * not exceed its limit; but let map_count go just above
2501 * its limit temporarily, to help free resources as expected.
2502 */
2503 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2504 return -ENOMEM;
2505
2506 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 if (error)
2508 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002509 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 }
2511
2512 /* Does it split the last one? */
2513 last = find_vma(mm, end);
2514 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002515 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 if (error)
2517 return error;
2518 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002519 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520
2521 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002522 * unlock any mlock()ed ranges before detaching vmas
2523 */
2524 if (mm->locked_vm) {
2525 struct vm_area_struct *tmp = vma;
2526 while (tmp && tmp->vm_start < end) {
2527 if (tmp->vm_flags & VM_LOCKED) {
2528 mm->locked_vm -= vma_pages(tmp);
2529 munlock_vma_pages_all(tmp);
2530 }
2531 tmp = tmp->vm_next;
2532 }
2533 }
2534
2535 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 * Remove the vma's, and unmap the actual pages
2537 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002538 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2539 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
2541 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002542 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
2544 return 0;
2545}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Al Virobfce2812012-04-20 21:57:04 -04002547int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002548{
2549 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002550 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002551
2552 down_write(&mm->mmap_sem);
2553 ret = do_munmap(mm, start, len);
2554 up_write(&mm->mmap_sem);
2555 return ret;
2556}
2557EXPORT_SYMBOL(vm_munmap);
2558
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002559SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002562 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563}
2564
2565static inline void verify_mm_writelocked(struct mm_struct *mm)
2566{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002567#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2569 WARN_ON(1);
2570 up_read(&mm->mmap_sem);
2571 }
2572#endif
2573}
2574
2575/*
2576 * this is really a simplified "do_mmap". it only handles
2577 * anonymous maps. eventually we may be able to do some
2578 * brk-specific accounting here.
2579 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002580static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581{
2582 struct mm_struct * mm = current->mm;
2583 struct vm_area_struct * vma, * prev;
2584 unsigned long flags;
2585 struct rb_node ** rb_link, * rb_parent;
2586 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002587 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
2589 len = PAGE_ALIGN(len);
2590 if (!len)
2591 return addr;
2592
Kirill Korotaev3a459752006-09-07 14:17:04 +04002593 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2594
Al Viro2c6a1012009-12-03 19:40:46 -05002595 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2596 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002597 return error;
2598
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 /*
2600 * mlock MCL_FUTURE?
2601 */
2602 if (mm->def_flags & VM_LOCKED) {
2603 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002604 locked = len >> PAGE_SHIFT;
2605 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002606 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002607 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2609 return -EAGAIN;
2610 }
2611
2612 /*
2613 * mm->mmap_sem is required to protect against another thread
2614 * changing the mappings in case we sleep.
2615 */
2616 verify_mm_writelocked(mm);
2617
2618 /*
2619 * Clear old maps. this also does some error checking for us
2620 */
2621 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002622 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 if (do_munmap(mm, addr, len))
2624 return -ENOMEM;
2625 goto munmap_back;
2626 }
2627
2628 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002629 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return -ENOMEM;
2631
2632 if (mm->map_count > sysctl_max_map_count)
2633 return -ENOMEM;
2634
Al Viro191c5422012-02-13 03:58:52 +00002635 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 return -ENOMEM;
2637
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002639 vma = vma_merge(mm, prev, addr, addr + len, flags,
2640 NULL, NULL, pgoff, NULL);
2641 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 goto out;
2643
2644 /*
2645 * create a vma struct for an anonymous mapping
2646 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002647 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 if (!vma) {
2649 vm_unacct_memory(len >> PAGE_SHIFT);
2650 return -ENOMEM;
2651 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
Rik van Riel5beb4932010-03-05 13:42:07 -08002653 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 vma->vm_mm = mm;
2655 vma->vm_start = addr;
2656 vma->vm_end = addr + len;
2657 vma->vm_pgoff = pgoff;
2658 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002659 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 vma_link(mm, vma, prev, rb_link, rb_parent);
2661out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002662 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002664 if (flags & VM_LOCKED)
2665 mm->locked_vm += (len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 return addr;
2667}
2668
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002669unsigned long vm_brk(unsigned long addr, unsigned long len)
2670{
2671 struct mm_struct *mm = current->mm;
2672 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002673 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002674
2675 down_write(&mm->mmap_sem);
2676 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002677 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002678 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002679 if (populate)
2680 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002681 return ret;
2682}
2683EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684
2685/* Release all mmaps. */
2686void exit_mmap(struct mm_struct *mm)
2687{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002688 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002689 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 unsigned long nr_accounted = 0;
2691
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002692 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002693 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002694
Rik van Rielba470de2008-10-18 20:26:50 -07002695 if (mm->locked_vm) {
2696 vma = mm->mmap;
2697 while (vma) {
2698 if (vma->vm_flags & VM_LOCKED)
2699 munlock_vma_pages_all(vma);
2700 vma = vma->vm_next;
2701 }
2702 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002703
2704 arch_exit_mmap(mm);
2705
Rik van Rielba470de2008-10-18 20:26:50 -07002706 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002707 if (!vma) /* Can happen if dup_mmap() received an OOM */
2708 return;
2709
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002712 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002713 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002714 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002715 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002716
Hugh Dickins6ee86302013-04-29 15:07:44 -07002717 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002718 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002721 * Walk the list again, actually closing and freeing it,
2722 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002724 while (vma) {
2725 if (vma->vm_flags & VM_ACCOUNT)
2726 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002727 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002728 }
2729 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002730
Hugh Dickinsf9aed622012-08-21 16:15:45 -07002731 WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732}
2733
2734/* Insert vm structure into process list sorted by address
2735 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002736 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002738int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739{
Hugh Dickins6597d782012-10-08 16:29:07 -07002740 struct vm_area_struct *prev;
2741 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742
2743 /*
2744 * The vm_pgoff of a purely anonymous vma should be irrelevant
2745 * until its first write fault, when page's anon_vma and index
2746 * are set. But now set the vm_pgoff it will almost certainly
2747 * end up with (unless mremap moves it elsewhere before that
2748 * first wfault), so /proc/pid/maps tells a consistent story.
2749 *
2750 * By setting it to reflect the virtual start address of the
2751 * vma, merges and splits can happen in a seamless way, just
2752 * using the existing file pgoff checks and manipulations.
2753 * Similarly in do_mmap_pgoff and in do_brk.
2754 */
2755 if (!vma->vm_file) {
2756 BUG_ON(vma->anon_vma);
2757 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2758 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002759 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2760 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002762 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002763 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002764 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 vma_link(mm, vma, prev, rb_link, rb_parent);
2767 return 0;
2768}
2769
2770/*
2771 * Copy the vma structure to a new location in the same mm,
2772 * prior to moving page table entries, to effect an mremap move.
2773 */
2774struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002775 unsigned long addr, unsigned long len, pgoff_t pgoff,
2776 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777{
2778 struct vm_area_struct *vma = *vmap;
2779 unsigned long vma_start = vma->vm_start;
2780 struct mm_struct *mm = vma->vm_mm;
2781 struct vm_area_struct *new_vma, *prev;
2782 struct rb_node **rb_link, *rb_parent;
2783 struct mempolicy *pol;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002784 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
2786 /*
2787 * If anonymous vma has not yet been faulted, update new pgoff
2788 * to match new location, to increase its chance of merging.
2789 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002790 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002792 faulted_in_anon_vma = false;
2793 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
Hugh Dickins6597d782012-10-08 16:29:07 -07002795 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2796 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2798 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2799 if (new_vma) {
2800 /*
2801 * Source vma may have been merged into new_vma
2802 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002803 if (unlikely(vma_start >= new_vma->vm_start &&
2804 vma_start < new_vma->vm_end)) {
2805 /*
2806 * The only way we can get a vma_merge with
2807 * self during an mremap is if the vma hasn't
2808 * been faulted in yet and we were allowed to
2809 * reset the dst vma->vm_pgoff to the
2810 * destination address of the mremap to allow
2811 * the merge to happen. mremap must change the
2812 * vm_pgoff linearity between src and dst vmas
2813 * (in turn preventing a vma_merge) to be
2814 * safe. It is only safe to keep the vm_pgoff
2815 * linear if there are no pages mapped yet.
2816 */
2817 VM_BUG_ON(faulted_in_anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002818 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002819 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002820 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002822 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 if (new_vma) {
2824 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 new_vma->vm_start = addr;
2826 new_vma->vm_end = addr + len;
2827 new_vma->vm_pgoff = pgoff;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002828 pol = mpol_dup(vma_policy(vma));
2829 if (IS_ERR(pol))
2830 goto out_free_vma;
2831 vma_set_policy(new_vma, pol);
2832 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2833 if (anon_vma_clone(new_vma, vma))
2834 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002835 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 get_file(new_vma->vm_file);
2837 if (new_vma->vm_ops && new_vma->vm_ops->open)
2838 new_vma->vm_ops->open(new_vma);
2839 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002840 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 }
2842 }
2843 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002844
2845 out_free_mempol:
2846 mpol_put(pol);
2847 out_free_vma:
2848 kmem_cache_free(vm_area_cachep, new_vma);
2849 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002851
2852/*
2853 * Return true if the calling process may expand its vm space by the passed
2854 * number of pages
2855 */
2856int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2857{
2858 unsigned long cur = mm->total_vm; /* pages */
2859 unsigned long lim;
2860
Jiri Slaby59e99e52010-03-05 13:41:44 -08002861 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002862
2863 if (cur + npages > lim)
2864 return 0;
2865 return 1;
2866}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002867
2868
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002869static int special_mapping_fault(struct vm_area_struct *vma,
2870 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002871{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002872 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002873 struct page **pages;
2874
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002875 /*
2876 * special mappings have no vm_file, and in that case, the mm
2877 * uses vm_pgoff internally. So we have to subtract it from here.
2878 * We are allowed to do this because we are the mm; do not copy
2879 * this code into drivers!
2880 */
2881 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002882
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002883 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2884 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002885
2886 if (*pages) {
2887 struct page *page = *pages;
2888 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002889 vmf->page = page;
2890 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002891 }
2892
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002893 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002894}
2895
2896/*
2897 * Having a close hook prevents vma merging regardless of flags.
2898 */
2899static void special_mapping_close(struct vm_area_struct *vma)
2900{
2901}
2902
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002903static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002904 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002905 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002906};
2907
2908/*
2909 * Called with mm->mmap_sem held for writing.
2910 * Insert a new vma covering the given region, with the given flags.
2911 * Its pages are supplied by the given array of struct page *.
2912 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2913 * The region past the last page supplied will always produce SIGBUS.
2914 * The array pointer and the pages it points to are assumed to stay alive
2915 * for as long as this mapping might exist.
2916 */
2917int install_special_mapping(struct mm_struct *mm,
2918 unsigned long addr, unsigned long len,
2919 unsigned long vm_flags, struct page **pages)
2920{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002921 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002922 struct vm_area_struct *vma;
2923
2924 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2925 if (unlikely(vma == NULL))
2926 return -ENOMEM;
2927
Rik van Riel5beb4932010-03-05 13:42:07 -08002928 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002929 vma->vm_mm = mm;
2930 vma->vm_start = addr;
2931 vma->vm_end = addr + len;
2932
Nick Piggin2f987352008-02-02 03:08:53 +01002933 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
Coly Li3ed75eb2007-10-18 23:39:15 -07002934 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002935
2936 vma->vm_ops = &special_mapping_vmops;
2937 vma->vm_private_data = pages;
2938
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002939 ret = insert_vm_struct(mm, vma);
2940 if (ret)
2941 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002942
2943 mm->total_vm += len >> PAGE_SHIFT;
2944
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002945 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002946
Roland McGrathfa5dc222007-02-08 14:20:41 -08002947 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002948
2949out:
2950 kmem_cache_free(vm_area_cachep, vma);
2951 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002952}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002953
2954static DEFINE_MUTEX(mm_all_locks_mutex);
2955
Peter Zijlstra454ed842008-08-11 09:30:25 +02002956static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002957{
Michel Lespinassebf181b92012-10-08 16:31:39 -07002958 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002959 /*
2960 * The LSB of head.next can't change from under us
2961 * because we hold the mm_all_locks_mutex.
2962 */
Jiri Kosina572043c2013-01-11 14:31:59 -08002963 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002964 /*
2965 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00002966 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002967 * the same anon_vma we won't take it again.
2968 *
2969 * No need of atomic instructions here, head.next
2970 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00002971 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002972 */
2973 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07002974 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002975 BUG();
2976 }
2977}
2978
Peter Zijlstra454ed842008-08-11 09:30:25 +02002979static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002980{
2981 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2982 /*
2983 * AS_MM_ALL_LOCKS can't change from under us because
2984 * we hold the mm_all_locks_mutex.
2985 *
2986 * Operations on ->flags have to be atomic because
2987 * even if AS_MM_ALL_LOCKS is stable thanks to the
2988 * mm_all_locks_mutex, there may be other cpus
2989 * changing other bitflags in parallel to us.
2990 */
2991 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2992 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002993 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002994 }
2995}
2996
2997/*
2998 * This operation locks against the VM for all pte/vma/mm related
2999 * operations that could ever happen on a certain mm. This includes
3000 * vmtruncate, try_to_unmap, and all page faults.
3001 *
3002 * The caller must take the mmap_sem in write mode before calling
3003 * mm_take_all_locks(). The caller isn't allowed to release the
3004 * mmap_sem until mm_drop_all_locks() returns.
3005 *
3006 * mmap_sem in write mode is required in order to block all operations
3007 * that could modify pagetables and free pages without need of
3008 * altering the vma layout (for example populate_range() with
3009 * nonlinear vmas). It's also needed in write mode to avoid new
3010 * anon_vmas to be associated with existing vmas.
3011 *
3012 * A single task can't take more than one mm_take_all_locks() in a row
3013 * or it would deadlock.
3014 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003015 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003016 * mapping->flags avoid to take the same lock twice, if more than one
3017 * vma in this mm is backed by the same anon_vma or address_space.
3018 *
3019 * We can take all the locks in random order because the VM code
Yuanhan Liu631b0cf2013-02-04 14:28:48 -08003020 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003021 * takes more than one of them in a row. Secondly we're protected
3022 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3023 *
3024 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3025 * that may have to take thousand of locks.
3026 *
3027 * mm_take_all_locks() can fail if it's interrupted by signals.
3028 */
3029int mm_take_all_locks(struct mm_struct *mm)
3030{
3031 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003032 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003033
3034 BUG_ON(down_read_trylock(&mm->mmap_sem));
3035
3036 mutex_lock(&mm_all_locks_mutex);
3037
3038 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3039 if (signal_pending(current))
3040 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003041 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003042 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003043 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003044
3045 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3046 if (signal_pending(current))
3047 goto out_unlock;
3048 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003049 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3050 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003051 }
3052
Kautuk Consul584cff52011-10-31 17:08:59 -07003053 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003054
3055out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003056 mm_drop_all_locks(mm);
3057 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003058}
3059
3060static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3061{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003062 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003063 /*
3064 * The LSB of head.next can't change to 0 from under
3065 * us because we hold the mm_all_locks_mutex.
3066 *
3067 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003068 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003069 * never see our bitflag.
3070 *
3071 * No need of atomic instructions here, head.next
3072 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003073 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003074 */
3075 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003076 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003077 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003078 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003079 }
3080}
3081
3082static void vm_unlock_mapping(struct address_space *mapping)
3083{
3084 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3085 /*
3086 * AS_MM_ALL_LOCKS can't change to 0 from under us
3087 * because we hold the mm_all_locks_mutex.
3088 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003089 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003090 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3091 &mapping->flags))
3092 BUG();
3093 }
3094}
3095
3096/*
3097 * The mmap_sem cannot be released by the caller until
3098 * mm_drop_all_locks() returns.
3099 */
3100void mm_drop_all_locks(struct mm_struct *mm)
3101{
3102 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003103 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003104
3105 BUG_ON(down_read_trylock(&mm->mmap_sem));
3106 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3107
3108 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3109 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003110 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3111 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003112 if (vma->vm_file && vma->vm_file->f_mapping)
3113 vm_unlock_mapping(vma->vm_file->f_mapping);
3114 }
3115
3116 mutex_unlock(&mm_all_locks_mutex);
3117}
David Howells8feae132009-01-08 12:04:47 +00003118
3119/*
3120 * initialise the VMA slab
3121 */
3122void __init mmap_init(void)
3123{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003124 int ret;
3125
3126 ret = percpu_counter_init(&vm_committed_as, 0);
3127 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003128}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003129
3130/*
3131 * Initialise sysctl_user_reserve_kbytes.
3132 *
3133 * This is intended to prevent a user from starting a single memory hogging
3134 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3135 * mode.
3136 *
3137 * The default value is min(3% of free memory, 128MB)
3138 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3139 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003140static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003141{
3142 unsigned long free_kbytes;
3143
3144 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3145
3146 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3147 return 0;
3148}
3149module_init(init_user_reserve)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003150
3151/*
3152 * Initialise sysctl_admin_reserve_kbytes.
3153 *
3154 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3155 * to log in and kill a memory hogging process.
3156 *
3157 * Systems with more than 256MB will reserve 8MB, enough to recover
3158 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3159 * only reserve 3% of free pages by default.
3160 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003161static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003162{
3163 unsigned long free_kbytes;
3164
3165 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3166
3167 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3168 return 0;
3169}
3170module_init(init_admin_reserve)
Andrew Shewmaker16408792013-04-29 15:08:12 -07003171
3172/*
3173 * Reinititalise user and admin reserves if memory is added or removed.
3174 *
3175 * The default user reserve max is 128MB, and the default max for the
3176 * admin reserve is 8MB. These are usually, but not always, enough to
3177 * enable recovery from a memory hogging process using login/sshd, a shell,
3178 * and tools like top. It may make sense to increase or even disable the
3179 * reserve depending on the existence of swap or variations in the recovery
3180 * tools. So, the admin may have changed them.
3181 *
3182 * If memory is added and the reserves have been eliminated or increased above
3183 * the default max, then we'll trust the admin.
3184 *
3185 * If memory is removed and there isn't enough free memory, then we
3186 * need to reset the reserves.
3187 *
3188 * Otherwise keep the reserve set by the admin.
3189 */
3190static int reserve_mem_notifier(struct notifier_block *nb,
3191 unsigned long action, void *data)
3192{
3193 unsigned long tmp, free_kbytes;
3194
3195 switch (action) {
3196 case MEM_ONLINE:
3197 /* Default max is 128MB. Leave alone if modified by operator. */
3198 tmp = sysctl_user_reserve_kbytes;
3199 if (0 < tmp && tmp < (1UL << 17))
3200 init_user_reserve();
3201
3202 /* Default max is 8MB. Leave alone if modified by operator. */
3203 tmp = sysctl_admin_reserve_kbytes;
3204 if (0 < tmp && tmp < (1UL << 13))
3205 init_admin_reserve();
3206
3207 break;
3208 case MEM_OFFLINE:
3209 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3210
3211 if (sysctl_user_reserve_kbytes > free_kbytes) {
3212 init_user_reserve();
3213 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3214 sysctl_user_reserve_kbytes);
3215 }
3216
3217 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3218 init_admin_reserve();
3219 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3220 sysctl_admin_reserve_kbytes);
3221 }
3222 break;
3223 default:
3224 break;
3225 }
3226 return NOTIFY_OK;
3227}
3228
3229static struct notifier_block reserve_mem_nb = {
3230 .notifier_call = reserve_mem_notifier,
3231};
3232
3233static int __meminit init_reserve_notifier(void)
3234{
3235 if (register_hotmemory_notifier(&reserve_mem_nb))
3236 printk("Failed registering memory add/remove notifier for admin reserve");
3237
3238 return 0;
3239}
3240module_init(init_reserve_notifier)