blob: fa206abb5c9de3614090732c9576303ea343facf [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;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001205 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
Michel Lespinasse41badc12013-02-22 16:32:47 -08001207 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 /*
1210 * Does the application expect PROT_READ to imply PROT_EXEC?
1211 *
1212 * (the exception is when the underlying filesystem is noexec
1213 * mounted, in which case we dont add PROT_EXEC.)
1214 */
1215 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001216 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 prot |= PROT_EXEC;
1218
1219 if (!len)
1220 return -EINVAL;
1221
Eric Paris7cd94142007-11-26 18:47:40 -05001222 if (!(flags & MAP_FIXED))
1223 addr = round_hint_to_min(addr);
1224
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 /* Careful about overflows.. */
1226 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001227 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 return -ENOMEM;
1229
1230 /* offset overflow? */
1231 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1232 return -EOVERFLOW;
1233
1234 /* Too many mappings? */
1235 if (mm->map_count > sysctl_max_map_count)
1236 return -ENOMEM;
1237
1238 /* Obtain the address to map to. we verify (or select) it and ensure
1239 * that it represents a valid section of the address space.
1240 */
1241 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1242 if (addr & ~PAGE_MASK)
1243 return addr;
1244
1245 /* Do simple checking here so the lower-level routines won't have
1246 * to. we assume access permissions have been handled by the open
1247 * of the memory object, so we don't do any here.
1248 */
1249 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1250 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1251
Huang Shijiecdf7b342009-09-21 17:03:36 -07001252 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 if (!can_do_mlock())
1254 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 /* mlock MCL_FUTURE? */
1257 if (vm_flags & VM_LOCKED) {
1258 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07001259 locked = len >> PAGE_SHIFT;
1260 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08001261 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07001262 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1264 return -EAGAIN;
1265 }
1266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001268 struct inode *inode = file_inode(file);
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 switch (flags & MAP_TYPE) {
1271 case MAP_SHARED:
1272 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1273 return -EACCES;
1274
1275 /*
1276 * Make sure we don't allow writing to an append-only
1277 * file..
1278 */
1279 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1280 return -EACCES;
1281
1282 /*
1283 * Make sure there are no mandatory locks on the file.
1284 */
1285 if (locks_verify_locked(inode))
1286 return -EAGAIN;
1287
1288 vm_flags |= VM_SHARED | VM_MAYSHARE;
1289 if (!(file->f_mode & FMODE_WRITE))
1290 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1291
1292 /* fall through */
1293 case MAP_PRIVATE:
1294 if (!(file->f_mode & FMODE_READ))
1295 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001296 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001297 if (vm_flags & VM_EXEC)
1298 return -EPERM;
1299 vm_flags &= ~VM_MAYEXEC;
1300 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001301
1302 if (!file->f_op || !file->f_op->mmap)
1303 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001304 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1305 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 break;
1307
1308 default:
1309 return -EINVAL;
1310 }
1311 } else {
1312 switch (flags & MAP_TYPE) {
1313 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001314 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1315 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001316 /*
1317 * Ignore pgoff.
1318 */
1319 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 vm_flags |= VM_SHARED | VM_MAYSHARE;
1321 break;
1322 case MAP_PRIVATE:
1323 /*
1324 * Set pgoff according to addr for anon_vma.
1325 */
1326 pgoff = addr >> PAGE_SHIFT;
1327 break;
1328 default:
1329 return -EINVAL;
1330 }
1331 }
1332
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001333 /*
1334 * Set 'VM_NORESERVE' if we should not account for the
1335 * memory use of this mapping.
1336 */
1337 if (flags & MAP_NORESERVE) {
1338 /* We honor MAP_NORESERVE if allowed to overcommit */
1339 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1340 vm_flags |= VM_NORESERVE;
1341
1342 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1343 if (file && is_file_hugepages(file))
1344 vm_flags |= VM_NORESERVE;
1345 }
1346
1347 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001348 if (!IS_ERR_VALUE(addr) &&
1349 ((vm_flags & VM_LOCKED) ||
1350 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001351 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001352 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001353}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001354
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001355SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1356 unsigned long, prot, unsigned long, flags,
1357 unsigned long, fd, unsigned long, pgoff)
1358{
1359 struct file *file = NULL;
1360 unsigned long retval = -EBADF;
1361
1362 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001363 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001364 file = fget(fd);
1365 if (!file)
1366 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001367 if (is_file_hugepages(file))
1368 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001369 retval = -EINVAL;
1370 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1371 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001372 } else if (flags & MAP_HUGETLB) {
1373 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001374 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001375
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001376 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001377 if (!hs)
1378 return -EINVAL;
1379
1380 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001381 /*
1382 * VM_NORESERVE is used because the reservations will be
1383 * taken when vm_ops->mmap() is called
1384 * A dummy user value is used because we are not locking
1385 * memory so no accounting is necessary
1386 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001387 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001388 VM_NORESERVE,
1389 &user, HUGETLB_ANONHUGE_INODE,
1390 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001391 if (IS_ERR(file))
1392 return PTR_ERR(file);
1393 }
1394
1395 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1396
Al Viroeb36c582012-05-30 20:17:35 -04001397 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001398out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001399 if (file)
1400 fput(file);
1401out:
1402 return retval;
1403}
1404
Christoph Hellwiga4679372010-03-10 15:21:15 -08001405#ifdef __ARCH_WANT_SYS_OLD_MMAP
1406struct mmap_arg_struct {
1407 unsigned long addr;
1408 unsigned long len;
1409 unsigned long prot;
1410 unsigned long flags;
1411 unsigned long fd;
1412 unsigned long offset;
1413};
1414
1415SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1416{
1417 struct mmap_arg_struct a;
1418
1419 if (copy_from_user(&a, arg, sizeof(a)))
1420 return -EFAULT;
1421 if (a.offset & ~PAGE_MASK)
1422 return -EINVAL;
1423
1424 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1425 a.offset >> PAGE_SHIFT);
1426}
1427#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1428
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001429/*
1430 * Some shared mappigns will want the pages marked read-only
1431 * to track write events. If so, we'll downgrade vm_page_prot
1432 * to the private version (using protection_map[] without the
1433 * VM_SHARED bit).
1434 */
1435int vma_wants_writenotify(struct vm_area_struct *vma)
1436{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001437 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001438
1439 /* If it was private or non-writable, the write bit is already clear */
1440 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1441 return 0;
1442
1443 /* The backer wishes to know when pages are first written to? */
1444 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1445 return 1;
1446
1447 /* The open routine did something to the protections already? */
1448 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001449 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001450 return 0;
1451
1452 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001453 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001454 return 0;
1455
1456 /* Can the mapping track the dirty pages? */
1457 return vma->vm_file && vma->vm_file->f_mapping &&
1458 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1459}
1460
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001461/*
1462 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001463 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001464 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001465static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001466{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001467 /*
1468 * hugetlb has its own accounting separate from the core VM
1469 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1470 */
1471 if (file && is_file_hugepages(file))
1472 return 0;
1473
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001474 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1475}
1476
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001477unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001478 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001479{
1480 struct mm_struct *mm = current->mm;
1481 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001482 int error;
1483 struct rb_node **rb_link, *rb_parent;
1484 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001485
Cyril Hrubise8420a82013-04-29 15:08:33 -07001486 /* Check against address space limit. */
1487 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1488 unsigned long nr_pages;
1489
1490 /*
1491 * MAP_FIXED may remove pages of mappings that intersects with
1492 * requested mapping. Account for the pages it would unmap.
1493 */
1494 if (!(vm_flags & MAP_FIXED))
1495 return -ENOMEM;
1496
1497 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1498
1499 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1500 return -ENOMEM;
1501 }
1502
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 /* Clear old maps */
1504 error = -ENOMEM;
1505munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001506 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 if (do_munmap(mm, addr, len))
1508 return -ENOMEM;
1509 goto munmap_back;
1510 }
1511
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001512 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001513 * Private writable mapping: check memory availability
1514 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001515 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001516 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001517 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001518 return -ENOMEM;
1519 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 }
1521
1522 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001523 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001525 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1526 if (vma)
1527 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529 /*
1530 * Determine the object being mapped and call the appropriate
1531 * specific mapper. the address has already been validated, but
1532 * not unmapped, but the maps are removed from the list.
1533 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001534 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 if (!vma) {
1536 error = -ENOMEM;
1537 goto unacct_error;
1538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
1540 vma->vm_mm = mm;
1541 vma->vm_start = addr;
1542 vma->vm_end = addr + len;
1543 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001544 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001546 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 if (vm_flags & VM_DENYWRITE) {
1550 error = deny_write_access(file);
1551 if (error)
1552 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 }
Al Virocb0942b2012-08-27 14:48:26 -04001554 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 error = file->f_op->mmap(file, vma);
1556 if (error)
1557 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001558
1559 /* Can addr have changed??
1560 *
1561 * Answer: Yes, several device drivers can do it in their
1562 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001563 * Bug: If addr is changed, prev, rb_link, rb_parent should
1564 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001565 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001566 WARN_ON_ONCE(addr != vma->vm_start);
1567
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001568 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001569 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 } else if (vm_flags & VM_SHARED) {
1571 error = shmem_zero_setup(vma);
1572 if (error)
1573 goto free_vma;
1574 }
1575
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001576 if (vma_wants_writenotify(vma)) {
1577 pgprot_t pprot = vma->vm_page_prot;
1578
1579 /* Can vma->vm_page_prot have changed??
1580 *
1581 * Answer: Yes, drivers may have changed it in their
1582 * f_op->mmap method.
1583 *
1584 * Ensures that vmas marked as uncached stay that way.
1585 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001586 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001587 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1588 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1589 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001590
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001591 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001592 /* Once vma denies write, undo our temporary denial count */
Oleg Nesterove8686772013-09-11 14:20:20 -07001593 if (vm_flags & VM_DENYWRITE)
1594 allow_write_access(file);
1595 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001596out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001597 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001598
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001599 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001601 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1602 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001603 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001604 else
1605 vma->vm_flags &= ~VM_LOCKED;
1606 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301607
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001608 if (file)
1609 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301610
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001611 /*
1612 * New (or expanded) vma always get soft dirty status.
1613 * Otherwise user-space soft-dirty page tracker won't
1614 * be able to distinguish situation when vma area unmapped,
1615 * then new mapped in-place (which must be aimed as
1616 * a completely new data area).
1617 */
1618 vma->vm_flags |= VM_SOFTDIRTY;
1619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 return addr;
1621
1622unmap_and_free_vma:
Oleg Nesterove8686772013-09-11 14:20:20 -07001623 if (vm_flags & VM_DENYWRITE)
1624 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 vma->vm_file = NULL;
1626 fput(file);
1627
1628 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001629 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1630 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631free_vma:
1632 kmem_cache_free(vm_area_cachep, vma);
1633unacct_error:
1634 if (charged)
1635 vm_unacct_memory(charged);
1636 return error;
1637}
1638
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001639unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1640{
1641 /*
1642 * We implement the search by looking for an rbtree node that
1643 * immediately follows a suitable gap. That is,
1644 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1645 * - gap_end = vma->vm_start >= info->low_limit + length;
1646 * - gap_end - gap_start >= length
1647 */
1648
1649 struct mm_struct *mm = current->mm;
1650 struct vm_area_struct *vma;
1651 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1652
1653 /* Adjust search length to account for worst case alignment overhead */
1654 length = info->length + info->align_mask;
1655 if (length < info->length)
1656 return -ENOMEM;
1657
1658 /* Adjust search limits by the desired length */
1659 if (info->high_limit < length)
1660 return -ENOMEM;
1661 high_limit = info->high_limit - length;
1662
1663 if (info->low_limit > high_limit)
1664 return -ENOMEM;
1665 low_limit = info->low_limit + length;
1666
1667 /* Check if rbtree root looks promising */
1668 if (RB_EMPTY_ROOT(&mm->mm_rb))
1669 goto check_highest;
1670 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1671 if (vma->rb_subtree_gap < length)
1672 goto check_highest;
1673
1674 while (true) {
1675 /* Visit left subtree if it looks promising */
1676 gap_end = vma->vm_start;
1677 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1678 struct vm_area_struct *left =
1679 rb_entry(vma->vm_rb.rb_left,
1680 struct vm_area_struct, vm_rb);
1681 if (left->rb_subtree_gap >= length) {
1682 vma = left;
1683 continue;
1684 }
1685 }
1686
1687 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1688check_current:
1689 /* Check if current node has a suitable gap */
1690 if (gap_start > high_limit)
1691 return -ENOMEM;
1692 if (gap_end >= low_limit && gap_end - gap_start >= length)
1693 goto found;
1694
1695 /* Visit right subtree if it looks promising */
1696 if (vma->vm_rb.rb_right) {
1697 struct vm_area_struct *right =
1698 rb_entry(vma->vm_rb.rb_right,
1699 struct vm_area_struct, vm_rb);
1700 if (right->rb_subtree_gap >= length) {
1701 vma = right;
1702 continue;
1703 }
1704 }
1705
1706 /* Go back up the rbtree to find next candidate node */
1707 while (true) {
1708 struct rb_node *prev = &vma->vm_rb;
1709 if (!rb_parent(prev))
1710 goto check_highest;
1711 vma = rb_entry(rb_parent(prev),
1712 struct vm_area_struct, vm_rb);
1713 if (prev == vma->vm_rb.rb_left) {
1714 gap_start = vma->vm_prev->vm_end;
1715 gap_end = vma->vm_start;
1716 goto check_current;
1717 }
1718 }
1719 }
1720
1721check_highest:
1722 /* Check highest gap, which does not precede any rbtree node */
1723 gap_start = mm->highest_vm_end;
1724 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1725 if (gap_start > high_limit)
1726 return -ENOMEM;
1727
1728found:
1729 /* We found a suitable gap. Clip it with the original low_limit. */
1730 if (gap_start < info->low_limit)
1731 gap_start = info->low_limit;
1732
1733 /* Adjust gap address to the desired alignment */
1734 gap_start += (info->align_offset - gap_start) & info->align_mask;
1735
1736 VM_BUG_ON(gap_start + info->length > info->high_limit);
1737 VM_BUG_ON(gap_start + info->length > gap_end);
1738 return gap_start;
1739}
1740
1741unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1742{
1743 struct mm_struct *mm = current->mm;
1744 struct vm_area_struct *vma;
1745 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1746
1747 /* Adjust search length to account for worst case alignment overhead */
1748 length = info->length + info->align_mask;
1749 if (length < info->length)
1750 return -ENOMEM;
1751
1752 /*
1753 * Adjust search limits by the desired length.
1754 * See implementation comment at top of unmapped_area().
1755 */
1756 gap_end = info->high_limit;
1757 if (gap_end < length)
1758 return -ENOMEM;
1759 high_limit = gap_end - length;
1760
1761 if (info->low_limit > high_limit)
1762 return -ENOMEM;
1763 low_limit = info->low_limit + length;
1764
1765 /* Check highest gap, which does not precede any rbtree node */
1766 gap_start = mm->highest_vm_end;
1767 if (gap_start <= high_limit)
1768 goto found_highest;
1769
1770 /* Check if rbtree root looks promising */
1771 if (RB_EMPTY_ROOT(&mm->mm_rb))
1772 return -ENOMEM;
1773 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1774 if (vma->rb_subtree_gap < length)
1775 return -ENOMEM;
1776
1777 while (true) {
1778 /* Visit right subtree if it looks promising */
1779 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1780 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1781 struct vm_area_struct *right =
1782 rb_entry(vma->vm_rb.rb_right,
1783 struct vm_area_struct, vm_rb);
1784 if (right->rb_subtree_gap >= length) {
1785 vma = right;
1786 continue;
1787 }
1788 }
1789
1790check_current:
1791 /* Check if current node has a suitable gap */
1792 gap_end = vma->vm_start;
1793 if (gap_end < low_limit)
1794 return -ENOMEM;
1795 if (gap_start <= high_limit && gap_end - gap_start >= length)
1796 goto found;
1797
1798 /* Visit left subtree if it looks promising */
1799 if (vma->vm_rb.rb_left) {
1800 struct vm_area_struct *left =
1801 rb_entry(vma->vm_rb.rb_left,
1802 struct vm_area_struct, vm_rb);
1803 if (left->rb_subtree_gap >= length) {
1804 vma = left;
1805 continue;
1806 }
1807 }
1808
1809 /* Go back up the rbtree to find next candidate node */
1810 while (true) {
1811 struct rb_node *prev = &vma->vm_rb;
1812 if (!rb_parent(prev))
1813 return -ENOMEM;
1814 vma = rb_entry(rb_parent(prev),
1815 struct vm_area_struct, vm_rb);
1816 if (prev == vma->vm_rb.rb_right) {
1817 gap_start = vma->vm_prev ?
1818 vma->vm_prev->vm_end : 0;
1819 goto check_current;
1820 }
1821 }
1822 }
1823
1824found:
1825 /* We found a suitable gap. Clip it with the original high_limit. */
1826 if (gap_end > info->high_limit)
1827 gap_end = info->high_limit;
1828
1829found_highest:
1830 /* Compute highest gap address at the desired alignment */
1831 gap_end -= info->length;
1832 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1833
1834 VM_BUG_ON(gap_end < info->low_limit);
1835 VM_BUG_ON(gap_end < gap_start);
1836 return gap_end;
1837}
1838
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839/* Get an address range which is currently unmapped.
1840 * For shmat() with addr=0.
1841 *
1842 * Ugly calling convention alert:
1843 * Return value with the low bits set means error value,
1844 * ie
1845 * if (ret & ~PAGE_MASK)
1846 * error = ret;
1847 *
1848 * This function "knows" that -ENOMEM has the bits set.
1849 */
1850#ifndef HAVE_ARCH_UNMAPPED_AREA
1851unsigned long
1852arch_get_unmapped_area(struct file *filp, unsigned long addr,
1853 unsigned long len, unsigned long pgoff, unsigned long flags)
1854{
1855 struct mm_struct *mm = current->mm;
1856 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001857 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 if (len > TASK_SIZE)
1860 return -ENOMEM;
1861
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001862 if (flags & MAP_FIXED)
1863 return addr;
1864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 if (addr) {
1866 addr = PAGE_ALIGN(addr);
1867 vma = find_vma(mm, addr);
1868 if (TASK_SIZE - len >= addr &&
1869 (!vma || addr + len <= vma->vm_start))
1870 return addr;
1871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001873 info.flags = 0;
1874 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001875 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001876 info.high_limit = TASK_SIZE;
1877 info.align_mask = 0;
1878 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879}
1880#endif
1881
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882/*
1883 * This mmap-allocator allocates new areas top-down from below the
1884 * stack's low limit (the base):
1885 */
1886#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1887unsigned long
1888arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1889 const unsigned long len, const unsigned long pgoff,
1890 const unsigned long flags)
1891{
1892 struct vm_area_struct *vma;
1893 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001894 unsigned long addr = addr0;
1895 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
1897 /* requested length too big for entire address space */
1898 if (len > TASK_SIZE)
1899 return -ENOMEM;
1900
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001901 if (flags & MAP_FIXED)
1902 return addr;
1903
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 /* requesting a specific address */
1905 if (addr) {
1906 addr = PAGE_ALIGN(addr);
1907 vma = find_vma(mm, addr);
1908 if (TASK_SIZE - len >= addr &&
1909 (!vma || addr + len <= vma->vm_start))
1910 return addr;
1911 }
1912
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001913 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1914 info.length = len;
1915 info.low_limit = PAGE_SIZE;
1916 info.high_limit = mm->mmap_base;
1917 info.align_mask = 0;
1918 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001919
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 /*
1921 * A failed mmap() very likely causes application failure,
1922 * so fall back to the bottom-up function here. This scenario
1923 * can happen with large stack limits and large mmap()
1924 * allocations.
1925 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001926 if (addr & ~PAGE_MASK) {
1927 VM_BUG_ON(addr != -ENOMEM);
1928 info.flags = 0;
1929 info.low_limit = TASK_UNMAPPED_BASE;
1930 info.high_limit = TASK_SIZE;
1931 addr = vm_unmapped_area(&info);
1932 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
1934 return addr;
1935}
1936#endif
1937
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938unsigned long
1939get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1940 unsigned long pgoff, unsigned long flags)
1941{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001942 unsigned long (*get_area)(struct file *, unsigned long,
1943 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
Al Viro9206de92009-12-03 15:23:11 -05001945 unsigned long error = arch_mmap_check(addr, len, flags);
1946 if (error)
1947 return error;
1948
1949 /* Careful about overflows.. */
1950 if (len > TASK_SIZE)
1951 return -ENOMEM;
1952
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001953 get_area = current->mm->get_unmapped_area;
1954 if (file && file->f_op && file->f_op->get_unmapped_area)
1955 get_area = file->f_op->get_unmapped_area;
1956 addr = get_area(file, addr, len, pgoff, flags);
1957 if (IS_ERR_VALUE(addr))
1958 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001960 if (addr > TASK_SIZE - len)
1961 return -ENOMEM;
1962 if (addr & ~PAGE_MASK)
1963 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001964
Al Viro9ac4ed42012-05-30 17:13:15 -04001965 addr = arch_rebalance_pgtables(addr, len);
1966 error = security_mmap_addr(addr);
1967 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968}
1969
1970EXPORT_SYMBOL(get_unmapped_area);
1971
1972/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001973struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974{
1975 struct vm_area_struct *vma = NULL;
1976
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001977 /* Check the cache first. */
1978 /* (Cache hit rate is typically around 35%.) */
Jan Stancekb6a9b7f2013-04-04 11:35:10 -07001979 vma = ACCESS_ONCE(mm->mmap_cache);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001980 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1981 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001983 rb_node = mm->mm_rb.rb_node;
1984 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001986 while (rb_node) {
1987 struct vm_area_struct *vma_tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001989 vma_tmp = rb_entry(rb_node,
1990 struct vm_area_struct, vm_rb);
1991
1992 if (vma_tmp->vm_end > addr) {
1993 vma = vma_tmp;
1994 if (vma_tmp->vm_start <= addr)
1995 break;
1996 rb_node = rb_node->rb_left;
1997 } else
1998 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 }
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002000 if (vma)
2001 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 }
2003 return vma;
2004}
2005
2006EXPORT_SYMBOL(find_vma);
2007
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002008/*
2009 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002010 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011struct vm_area_struct *
2012find_vma_prev(struct mm_struct *mm, unsigned long addr,
2013 struct vm_area_struct **pprev)
2014{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002015 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002017 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002018 if (vma) {
2019 *pprev = vma->vm_prev;
2020 } else {
2021 struct rb_node *rb_node = mm->mm_rb.rb_node;
2022 *pprev = NULL;
2023 while (rb_node) {
2024 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2025 rb_node = rb_node->rb_right;
2026 }
2027 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002028 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029}
2030
2031/*
2032 * Verify that the stack growth is acceptable and
2033 * update accounting. This is shared with both the
2034 * grow-up and grow-down cases.
2035 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002036static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037{
2038 struct mm_struct *mm = vma->vm_mm;
2039 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08002040 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
2042 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002043 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 return -ENOMEM;
2045
2046 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08002047 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return -ENOMEM;
2049
2050 /* mlock limit tests */
2051 if (vma->vm_flags & VM_LOCKED) {
2052 unsigned long locked;
2053 unsigned long limit;
2054 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002055 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2056 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 if (locked > limit && !capable(CAP_IPC_LOCK))
2058 return -ENOMEM;
2059 }
2060
Adam Litke0d59a012007-01-30 14:35:39 -08002061 /* Check to ensure the stack will not grow into a hugetlb-only region */
2062 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2063 vma->vm_end - size;
2064 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2065 return -EFAULT;
2066
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 /*
2068 * Overcommit.. This must be the final test, as it will
2069 * update security statistics.
2070 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002071 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 return -ENOMEM;
2073
2074 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 if (vma->vm_flags & VM_LOCKED)
2076 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002077 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 return 0;
2079}
2080
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002081#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002083 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2084 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002086int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087{
2088 int error;
2089
2090 if (!(vma->vm_flags & VM_GROWSUP))
2091 return -EFAULT;
2092
2093 /*
2094 * We must make sure the anon_vma is allocated
2095 * so that the anon_vma locking is not a noop.
2096 */
2097 if (unlikely(anon_vma_prepare(vma)))
2098 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002099 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
2101 /*
2102 * vma->vm_start/vm_end cannot change under us because the caller
2103 * is required to hold the mmap_sem in read mode. We need the
2104 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002105 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 */
Helge Deller06b32f32006-12-19 19:28:33 +01002107 if (address < PAGE_ALIGN(address+4))
2108 address = PAGE_ALIGN(address+4);
2109 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002110 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002111 return -ENOMEM;
2112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 error = 0;
2114
2115 /* Somebody else might have raced and expanded it already */
2116 if (address > vma->vm_end) {
2117 unsigned long size, grow;
2118
2119 size = address - vma->vm_start;
2120 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2121
Hugh Dickins42c36f62011-05-09 17:44:42 -07002122 error = -ENOMEM;
2123 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2124 error = acct_stack_growth(vma, size, grow);
2125 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002126 /*
2127 * vma_gap_update() doesn't support concurrent
2128 * updates, but we only hold a shared mmap_sem
2129 * lock here, so we need to protect against
2130 * concurrent vma expansions.
2131 * vma_lock_anon_vma() doesn't help here, as
2132 * we don't guarantee that all growable vmas
2133 * in a mm share the same root anon vma.
2134 * So, we reuse mm->page_table_lock to guard
2135 * against concurrent vma expansions.
2136 */
2137 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002138 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002139 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002140 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002141 if (vma->vm_next)
2142 vma_gap_update(vma->vm_next);
2143 else
2144 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002145 spin_unlock(&vma->vm_mm->page_table_lock);
2146
Hugh Dickins42c36f62011-05-09 17:44:42 -07002147 perf_event_mmap(vma);
2148 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002151 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002152 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002153 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 return error;
2155}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002156#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158/*
2159 * vma is the first one with address < vma->vm_start. Have to extend vma.
2160 */
Michal Hockod05f3162011-05-24 17:11:44 -07002161int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002162 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163{
2164 int error;
2165
2166 /*
2167 * We must make sure the anon_vma is allocated
2168 * so that the anon_vma locking is not a noop.
2169 */
2170 if (unlikely(anon_vma_prepare(vma)))
2171 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002172
2173 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002174 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002175 if (error)
2176 return error;
2177
Rik van Rielbb4a3402010-08-09 17:18:37 -07002178 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
2180 /*
2181 * vma->vm_start/vm_end cannot change under us because the caller
2182 * is required to hold the mmap_sem in read mode. We need the
2183 * anon_vma lock to serialize against concurrent expand_stacks.
2184 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185
2186 /* Somebody else might have raced and expanded it already */
2187 if (address < vma->vm_start) {
2188 unsigned long size, grow;
2189
2190 size = vma->vm_end - address;
2191 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2192
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002193 error = -ENOMEM;
2194 if (grow <= vma->vm_pgoff) {
2195 error = acct_stack_growth(vma, size, grow);
2196 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002197 /*
2198 * vma_gap_update() doesn't support concurrent
2199 * updates, but we only hold a shared mmap_sem
2200 * lock here, so we need to protect against
2201 * concurrent vma expansions.
2202 * vma_lock_anon_vma() doesn't help here, as
2203 * we don't guarantee that all growable vmas
2204 * in a mm share the same root anon vma.
2205 * So, we reuse mm->page_table_lock to guard
2206 * against concurrent vma expansions.
2207 */
2208 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002209 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002210 vma->vm_start = address;
2211 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002212 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002213 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002214 spin_unlock(&vma->vm_mm->page_table_lock);
2215
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002216 perf_event_mmap(vma);
2217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 }
2219 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002220 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002221 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002222 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 return error;
2224}
2225
Linus Torvalds09884962013-02-27 08:36:04 -08002226/*
2227 * Note how expand_stack() refuses to expand the stack all the way to
2228 * abut the next virtual mapping, *unless* that mapping itself is also
2229 * a stack mapping. We want to leave room for a guard page, after all
2230 * (the guard page itself is not added here, that is done by the
2231 * actual page faulting logic)
2232 *
2233 * This matches the behavior of the guard page logic (see mm/memory.c:
2234 * check_stack_guard_page()), which only allows the guard page to be
2235 * removed under these circumstances.
2236 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002237#ifdef CONFIG_STACK_GROWSUP
2238int expand_stack(struct vm_area_struct *vma, unsigned long address)
2239{
Linus Torvalds09884962013-02-27 08:36:04 -08002240 struct vm_area_struct *next;
2241
2242 address &= PAGE_MASK;
2243 next = vma->vm_next;
2244 if (next && next->vm_start == address + PAGE_SIZE) {
2245 if (!(next->vm_flags & VM_GROWSUP))
2246 return -ENOMEM;
2247 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002248 return expand_upwards(vma, address);
2249}
2250
2251struct vm_area_struct *
2252find_extend_vma(struct mm_struct *mm, unsigned long addr)
2253{
2254 struct vm_area_struct *vma, *prev;
2255
2256 addr &= PAGE_MASK;
2257 vma = find_vma_prev(mm, addr, &prev);
2258 if (vma && (vma->vm_start <= addr))
2259 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002260 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002261 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002262 if (prev->vm_flags & VM_LOCKED)
2263 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002264 return prev;
2265}
2266#else
2267int expand_stack(struct vm_area_struct *vma, unsigned long address)
2268{
Linus Torvalds09884962013-02-27 08:36:04 -08002269 struct vm_area_struct *prev;
2270
2271 address &= PAGE_MASK;
2272 prev = vma->vm_prev;
2273 if (prev && prev->vm_end == address) {
2274 if (!(prev->vm_flags & VM_GROWSDOWN))
2275 return -ENOMEM;
2276 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002277 return expand_downwards(vma, address);
2278}
2279
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280struct vm_area_struct *
2281find_extend_vma(struct mm_struct * mm, unsigned long addr)
2282{
2283 struct vm_area_struct * vma;
2284 unsigned long start;
2285
2286 addr &= PAGE_MASK;
2287 vma = find_vma(mm,addr);
2288 if (!vma)
2289 return NULL;
2290 if (vma->vm_start <= addr)
2291 return vma;
2292 if (!(vma->vm_flags & VM_GROWSDOWN))
2293 return NULL;
2294 start = vma->vm_start;
2295 if (expand_stack(vma, addr))
2296 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002297 if (vma->vm_flags & VM_LOCKED)
2298 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 return vma;
2300}
2301#endif
2302
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002304 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002306 *
2307 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002309static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002311 unsigned long nr_accounted = 0;
2312
Hugh Dickins365e9c872005-10-29 18:16:18 -07002313 /* Update high watermark before we lower total_vm */
2314 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002316 long nrpages = vma_pages(vma);
2317
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002318 if (vma->vm_flags & VM_ACCOUNT)
2319 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002320 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002321 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002322 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002323 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 validate_mm(mm);
2325}
2326
2327/*
2328 * Get rid of page table information in the indicated region.
2329 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002330 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 */
2332static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002333 struct vm_area_struct *vma, struct vm_area_struct *prev,
2334 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335{
Hugh Dickinse0da3822005-04-19 13:29:15 -07002336 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002337 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
2339 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002340 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002341 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002342 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002343 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002344 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002345 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346}
2347
2348/*
2349 * Create a list of vma's touched by the unmap, removing them from the mm's
2350 * vma list as we go..
2351 */
2352static void
2353detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2354 struct vm_area_struct *prev, unsigned long end)
2355{
2356 struct vm_area_struct **insertion_point;
2357 struct vm_area_struct *tail_vma = NULL;
2358
2359 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002360 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002362 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 mm->map_count--;
2364 tail_vma = vma;
2365 vma = vma->vm_next;
2366 } while (vma && vma->vm_start < end);
2367 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002368 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002369 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002370 vma_gap_update(vma);
2371 } else
2372 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 tail_vma->vm_next = NULL;
2374 mm->mmap_cache = NULL; /* Kill the cache. */
2375}
2376
2377/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002378 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2379 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002381static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 unsigned long addr, int new_below)
2383{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 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
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002407 err = vma_dup_policy(vma, new);
2408 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002409 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
Rik van Riel5beb4932010-03-05 13:42:07 -08002411 if (anon_vma_clone(new, vma))
2412 goto out_free_mpol;
2413
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002414 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 get_file(new->vm_file);
2416
2417 if (new->vm_ops && new->vm_ops->open)
2418 new->vm_ops->open(new);
2419
2420 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002421 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2423 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002424 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Rik van Riel5beb4932010-03-05 13:42:07 -08002426 /* Success. */
2427 if (!err)
2428 return 0;
2429
2430 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002431 if (new->vm_ops && new->vm_ops->close)
2432 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002433 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002434 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002435 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002436 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002437 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002438 out_free_vma:
2439 kmem_cache_free(vm_area_cachep, new);
2440 out_err:
2441 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442}
2443
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002444/*
2445 * Split a vma into two pieces at address 'addr', a new vma is allocated
2446 * either for the first part or the tail.
2447 */
2448int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2449 unsigned long addr, int new_below)
2450{
2451 if (mm->map_count >= sysctl_max_map_count)
2452 return -ENOMEM;
2453
2454 return __split_vma(mm, vma, addr, new_below);
2455}
2456
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457/* Munmap is split into 2 main parts -- this part which finds
2458 * what needs doing, and the areas themselves, which do the
2459 * work. This now handles partial unmappings.
2460 * Jeremy Fitzhardinge <jeremy@goop.org>
2461 */
2462int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2463{
2464 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002465 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
2467 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2468 return -EINVAL;
2469
2470 if ((len = PAGE_ALIGN(len)) == 0)
2471 return -EINVAL;
2472
2473 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002474 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002475 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002477 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002478 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
2480 /* if it doesn't overlap, we have nothing.. */
2481 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002482 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 return 0;
2484
2485 /*
2486 * If we need to split any vma, do it now to save pain later.
2487 *
2488 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2489 * unmapped vm_area_struct will remain in use: so lower split_vma
2490 * places tmp vma above, and higher split_vma places tmp vma below.
2491 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002492 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002493 int error;
2494
2495 /*
2496 * Make sure that map_count on return from munmap() will
2497 * not exceed its limit; but let map_count go just above
2498 * its limit temporarily, to help free resources as expected.
2499 */
2500 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2501 return -ENOMEM;
2502
2503 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 if (error)
2505 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002506 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 }
2508
2509 /* Does it split the last one? */
2510 last = find_vma(mm, end);
2511 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002512 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 if (error)
2514 return error;
2515 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002516 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
2518 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002519 * unlock any mlock()ed ranges before detaching vmas
2520 */
2521 if (mm->locked_vm) {
2522 struct vm_area_struct *tmp = vma;
2523 while (tmp && tmp->vm_start < end) {
2524 if (tmp->vm_flags & VM_LOCKED) {
2525 mm->locked_vm -= vma_pages(tmp);
2526 munlock_vma_pages_all(tmp);
2527 }
2528 tmp = tmp->vm_next;
2529 }
2530 }
2531
2532 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 * Remove the vma's, and unmap the actual pages
2534 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002535 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2536 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
2538 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002539 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
2541 return 0;
2542}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
Al Virobfce2812012-04-20 21:57:04 -04002544int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002545{
2546 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002547 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002548
2549 down_write(&mm->mmap_sem);
2550 ret = do_munmap(mm, start, len);
2551 up_write(&mm->mmap_sem);
2552 return ret;
2553}
2554EXPORT_SYMBOL(vm_munmap);
2555
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002556SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002559 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560}
2561
2562static inline void verify_mm_writelocked(struct mm_struct *mm)
2563{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002564#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2566 WARN_ON(1);
2567 up_read(&mm->mmap_sem);
2568 }
2569#endif
2570}
2571
2572/*
2573 * this is really a simplified "do_mmap". it only handles
2574 * anonymous maps. eventually we may be able to do some
2575 * brk-specific accounting here.
2576 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002577static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578{
2579 struct mm_struct * mm = current->mm;
2580 struct vm_area_struct * vma, * prev;
2581 unsigned long flags;
2582 struct rb_node ** rb_link, * rb_parent;
2583 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002584 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
2586 len = PAGE_ALIGN(len);
2587 if (!len)
2588 return addr;
2589
Kirill Korotaev3a459752006-09-07 14:17:04 +04002590 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2591
Al Viro2c6a1012009-12-03 19:40:46 -05002592 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2593 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002594 return error;
2595
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 /*
2597 * mlock MCL_FUTURE?
2598 */
2599 if (mm->def_flags & VM_LOCKED) {
2600 unsigned long locked, lock_limit;
Chris Wright93ea1d02005-05-01 08:58:38 -07002601 locked = len >> PAGE_SHIFT;
2602 locked += mm->locked_vm;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002603 lock_limit = rlimit(RLIMIT_MEMLOCK);
Chris Wright93ea1d02005-05-01 08:58:38 -07002604 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2606 return -EAGAIN;
2607 }
2608
2609 /*
2610 * mm->mmap_sem is required to protect against another thread
2611 * changing the mappings in case we sleep.
2612 */
2613 verify_mm_writelocked(mm);
2614
2615 /*
2616 * Clear old maps. this also does some error checking for us
2617 */
2618 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002619 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 if (do_munmap(mm, addr, len))
2621 return -ENOMEM;
2622 goto munmap_back;
2623 }
2624
2625 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002626 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 return -ENOMEM;
2628
2629 if (mm->map_count > sysctl_max_map_count)
2630 return -ENOMEM;
2631
Al Viro191c5422012-02-13 03:58:52 +00002632 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 return -ENOMEM;
2634
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002636 vma = vma_merge(mm, prev, addr, addr + len, flags,
2637 NULL, NULL, pgoff, NULL);
2638 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 goto out;
2640
2641 /*
2642 * create a vma struct for an anonymous mapping
2643 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002644 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 if (!vma) {
2646 vm_unacct_memory(len >> PAGE_SHIFT);
2647 return -ENOMEM;
2648 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649
Rik van Riel5beb4932010-03-05 13:42:07 -08002650 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 vma->vm_mm = mm;
2652 vma->vm_start = addr;
2653 vma->vm_end = addr + len;
2654 vma->vm_pgoff = pgoff;
2655 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002656 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 vma_link(mm, vma, prev, rb_link, rb_parent);
2658out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002659 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002661 if (flags & VM_LOCKED)
2662 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002663 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 return addr;
2665}
2666
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002667unsigned long vm_brk(unsigned long addr, unsigned long len)
2668{
2669 struct mm_struct *mm = current->mm;
2670 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002671 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002672
2673 down_write(&mm->mmap_sem);
2674 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002675 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002676 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002677 if (populate)
2678 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002679 return ret;
2680}
2681EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682
2683/* Release all mmaps. */
2684void exit_mmap(struct mm_struct *mm)
2685{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002686 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002687 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 unsigned long nr_accounted = 0;
2689
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002690 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002691 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002692
Rik van Rielba470de2008-10-18 20:26:50 -07002693 if (mm->locked_vm) {
2694 vma = mm->mmap;
2695 while (vma) {
2696 if (vma->vm_flags & VM_LOCKED)
2697 munlock_vma_pages_all(vma);
2698 vma = vma->vm_next;
2699 }
2700 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002701
2702 arch_exit_mmap(mm);
2703
Rik van Rielba470de2008-10-18 20:26:50 -07002704 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002705 if (!vma) /* Can happen if dup_mmap() received an OOM */
2706 return;
2707
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002710 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002711 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002712 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002713 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002714
Hugh Dickins6ee86302013-04-29 15:07:44 -07002715 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002716 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002719 * Walk the list again, actually closing and freeing it,
2720 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002722 while (vma) {
2723 if (vma->vm_flags & VM_ACCOUNT)
2724 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002725 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002726 }
2727 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002728
Hugh Dickinsf9aed622012-08-21 16:15:45 -07002729 WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730}
2731
2732/* Insert vm structure into process list sorted by address
2733 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002734 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002736int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737{
Hugh Dickins6597d782012-10-08 16:29:07 -07002738 struct vm_area_struct *prev;
2739 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740
2741 /*
2742 * The vm_pgoff of a purely anonymous vma should be irrelevant
2743 * until its first write fault, when page's anon_vma and index
2744 * are set. But now set the vm_pgoff it will almost certainly
2745 * end up with (unless mremap moves it elsewhere before that
2746 * first wfault), so /proc/pid/maps tells a consistent story.
2747 *
2748 * By setting it to reflect the virtual start address of the
2749 * vma, merges and splits can happen in a seamless way, just
2750 * using the existing file pgoff checks and manipulations.
2751 * Similarly in do_mmap_pgoff and in do_brk.
2752 */
2753 if (!vma->vm_file) {
2754 BUG_ON(vma->anon_vma);
2755 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2756 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002757 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2758 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002760 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002761 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002762 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302763
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 vma_link(mm, vma, prev, rb_link, rb_parent);
2765 return 0;
2766}
2767
2768/*
2769 * Copy the vma structure to a new location in the same mm,
2770 * prior to moving page table entries, to effect an mremap move.
2771 */
2772struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002773 unsigned long addr, unsigned long len, pgoff_t pgoff,
2774 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775{
2776 struct vm_area_struct *vma = *vmap;
2777 unsigned long vma_start = vma->vm_start;
2778 struct mm_struct *mm = vma->vm_mm;
2779 struct vm_area_struct *new_vma, *prev;
2780 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002781 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782
2783 /*
2784 * If anonymous vma has not yet been faulted, update new pgoff
2785 * to match new location, to increase its chance of merging.
2786 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002787 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002789 faulted_in_anon_vma = false;
2790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
Hugh Dickins6597d782012-10-08 16:29:07 -07002792 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2793 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2795 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2796 if (new_vma) {
2797 /*
2798 * Source vma may have been merged into new_vma
2799 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002800 if (unlikely(vma_start >= new_vma->vm_start &&
2801 vma_start < new_vma->vm_end)) {
2802 /*
2803 * The only way we can get a vma_merge with
2804 * self during an mremap is if the vma hasn't
2805 * been faulted in yet and we were allowed to
2806 * reset the dst vma->vm_pgoff to the
2807 * destination address of the mremap to allow
2808 * the merge to happen. mremap must change the
2809 * vm_pgoff linearity between src and dst vmas
2810 * (in turn preventing a vma_merge) to be
2811 * safe. It is only safe to keep the vm_pgoff
2812 * linear if there are no pages mapped yet.
2813 */
2814 VM_BUG_ON(faulted_in_anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002815 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002816 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002817 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002819 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 if (new_vma) {
2821 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 new_vma->vm_start = addr;
2823 new_vma->vm_end = addr + len;
2824 new_vma->vm_pgoff = pgoff;
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002825 if (vma_dup_policy(vma, new_vma))
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002826 goto out_free_vma;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002827 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2828 if (anon_vma_clone(new_vma, vma))
2829 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002830 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 get_file(new_vma->vm_file);
2832 if (new_vma->vm_ops && new_vma->vm_ops->open)
2833 new_vma->vm_ops->open(new_vma);
2834 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002835 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 }
2837 }
2838 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002839
2840 out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002841 mpol_put(vma_policy(new_vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002842 out_free_vma:
2843 kmem_cache_free(vm_area_cachep, new_vma);
2844 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002846
2847/*
2848 * Return true if the calling process may expand its vm space by the passed
2849 * number of pages
2850 */
2851int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2852{
2853 unsigned long cur = mm->total_vm; /* pages */
2854 unsigned long lim;
2855
Jiri Slaby59e99e52010-03-05 13:41:44 -08002856 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002857
2858 if (cur + npages > lim)
2859 return 0;
2860 return 1;
2861}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002862
2863
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002864static int special_mapping_fault(struct vm_area_struct *vma,
2865 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002866{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002867 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002868 struct page **pages;
2869
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002870 /*
2871 * special mappings have no vm_file, and in that case, the mm
2872 * uses vm_pgoff internally. So we have to subtract it from here.
2873 * We are allowed to do this because we are the mm; do not copy
2874 * this code into drivers!
2875 */
2876 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002877
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002878 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2879 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002880
2881 if (*pages) {
2882 struct page *page = *pages;
2883 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002884 vmf->page = page;
2885 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002886 }
2887
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002888 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002889}
2890
2891/*
2892 * Having a close hook prevents vma merging regardless of flags.
2893 */
2894static void special_mapping_close(struct vm_area_struct *vma)
2895{
2896}
2897
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04002898static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002899 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002900 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002901};
2902
2903/*
2904 * Called with mm->mmap_sem held for writing.
2905 * Insert a new vma covering the given region, with the given flags.
2906 * Its pages are supplied by the given array of struct page *.
2907 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2908 * The region past the last page supplied will always produce SIGBUS.
2909 * The array pointer and the pages it points to are assumed to stay alive
2910 * for as long as this mapping might exist.
2911 */
2912int install_special_mapping(struct mm_struct *mm,
2913 unsigned long addr, unsigned long len,
2914 unsigned long vm_flags, struct page **pages)
2915{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002916 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002917 struct vm_area_struct *vma;
2918
2919 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2920 if (unlikely(vma == NULL))
2921 return -ENOMEM;
2922
Rik van Riel5beb4932010-03-05 13:42:07 -08002923 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002924 vma->vm_mm = mm;
2925 vma->vm_start = addr;
2926 vma->vm_end = addr + len;
2927
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002928 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07002929 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002930
2931 vma->vm_ops = &special_mapping_vmops;
2932 vma->vm_private_data = pages;
2933
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002934 ret = insert_vm_struct(mm, vma);
2935 if (ret)
2936 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002937
2938 mm->total_vm += len >> PAGE_SHIFT;
2939
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002940 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002941
Roland McGrathfa5dc222007-02-08 14:20:41 -08002942 return 0;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002943
2944out:
2945 kmem_cache_free(vm_area_cachep, vma);
2946 return ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002947}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002948
2949static DEFINE_MUTEX(mm_all_locks_mutex);
2950
Peter Zijlstra454ed842008-08-11 09:30:25 +02002951static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002952{
Michel Lespinassebf181b92012-10-08 16:31:39 -07002953 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002954 /*
2955 * The LSB of head.next can't change from under us
2956 * because we hold the mm_all_locks_mutex.
2957 */
Jiri Kosina572043c2013-01-11 14:31:59 -08002958 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002959 /*
2960 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00002961 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002962 * the same anon_vma we won't take it again.
2963 *
2964 * No need of atomic instructions here, head.next
2965 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00002966 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002967 */
2968 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07002969 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002970 BUG();
2971 }
2972}
2973
Peter Zijlstra454ed842008-08-11 09:30:25 +02002974static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002975{
2976 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2977 /*
2978 * AS_MM_ALL_LOCKS can't change from under us because
2979 * we hold the mm_all_locks_mutex.
2980 *
2981 * Operations on ->flags have to be atomic because
2982 * even if AS_MM_ALL_LOCKS is stable thanks to the
2983 * mm_all_locks_mutex, there may be other cpus
2984 * changing other bitflags in parallel to us.
2985 */
2986 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2987 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002988 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002989 }
2990}
2991
2992/*
2993 * This operation locks against the VM for all pte/vma/mm related
2994 * operations that could ever happen on a certain mm. This includes
2995 * vmtruncate, try_to_unmap, and all page faults.
2996 *
2997 * The caller must take the mmap_sem in write mode before calling
2998 * mm_take_all_locks(). The caller isn't allowed to release the
2999 * mmap_sem until mm_drop_all_locks() returns.
3000 *
3001 * mmap_sem in write mode is required in order to block all operations
3002 * that could modify pagetables and free pages without need of
3003 * altering the vma layout (for example populate_range() with
3004 * nonlinear vmas). It's also needed in write mode to avoid new
3005 * anon_vmas to be associated with existing vmas.
3006 *
3007 * A single task can't take more than one mm_take_all_locks() in a row
3008 * or it would deadlock.
3009 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003010 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003011 * mapping->flags avoid to take the same lock twice, if more than one
3012 * vma in this mm is backed by the same anon_vma or address_space.
3013 *
3014 * We can take all the locks in random order because the VM code
Yuanhan Liu631b0cf2013-02-04 14:28:48 -08003015 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003016 * takes more than one of them in a row. Secondly we're protected
3017 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3018 *
3019 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3020 * that may have to take thousand of locks.
3021 *
3022 * mm_take_all_locks() can fail if it's interrupted by signals.
3023 */
3024int mm_take_all_locks(struct mm_struct *mm)
3025{
3026 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003027 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003028
3029 BUG_ON(down_read_trylock(&mm->mmap_sem));
3030
3031 mutex_lock(&mm_all_locks_mutex);
3032
3033 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3034 if (signal_pending(current))
3035 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003036 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003037 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003038 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003039
3040 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3041 if (signal_pending(current))
3042 goto out_unlock;
3043 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003044 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3045 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003046 }
3047
Kautuk Consul584cff52011-10-31 17:08:59 -07003048 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003049
3050out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003051 mm_drop_all_locks(mm);
3052 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003053}
3054
3055static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3056{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003057 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003058 /*
3059 * The LSB of head.next can't change to 0 from under
3060 * us because we hold the mm_all_locks_mutex.
3061 *
3062 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003063 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003064 * never see our bitflag.
3065 *
3066 * No need of atomic instructions here, head.next
3067 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003068 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003069 */
3070 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003071 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003072 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003073 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003074 }
3075}
3076
3077static void vm_unlock_mapping(struct address_space *mapping)
3078{
3079 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3080 /*
3081 * AS_MM_ALL_LOCKS can't change to 0 from under us
3082 * because we hold the mm_all_locks_mutex.
3083 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003084 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003085 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3086 &mapping->flags))
3087 BUG();
3088 }
3089}
3090
3091/*
3092 * The mmap_sem cannot be released by the caller until
3093 * mm_drop_all_locks() returns.
3094 */
3095void mm_drop_all_locks(struct mm_struct *mm)
3096{
3097 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003098 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003099
3100 BUG_ON(down_read_trylock(&mm->mmap_sem));
3101 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3102
3103 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3104 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003105 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3106 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003107 if (vma->vm_file && vma->vm_file->f_mapping)
3108 vm_unlock_mapping(vma->vm_file->f_mapping);
3109 }
3110
3111 mutex_unlock(&mm_all_locks_mutex);
3112}
David Howells8feae132009-01-08 12:04:47 +00003113
3114/*
3115 * initialise the VMA slab
3116 */
3117void __init mmap_init(void)
3118{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003119 int ret;
3120
3121 ret = percpu_counter_init(&vm_committed_as, 0);
3122 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003123}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003124
3125/*
3126 * Initialise sysctl_user_reserve_kbytes.
3127 *
3128 * This is intended to prevent a user from starting a single memory hogging
3129 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3130 * mode.
3131 *
3132 * The default value is min(3% of free memory, 128MB)
3133 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3134 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003135static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003136{
3137 unsigned long free_kbytes;
3138
3139 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3140
3141 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3142 return 0;
3143}
3144module_init(init_user_reserve)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003145
3146/*
3147 * Initialise sysctl_admin_reserve_kbytes.
3148 *
3149 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3150 * to log in and kill a memory hogging process.
3151 *
3152 * Systems with more than 256MB will reserve 8MB, enough to recover
3153 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3154 * only reserve 3% of free pages by default.
3155 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003156static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003157{
3158 unsigned long free_kbytes;
3159
3160 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3161
3162 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3163 return 0;
3164}
3165module_init(init_admin_reserve)
Andrew Shewmaker16408792013-04-29 15:08:12 -07003166
3167/*
3168 * Reinititalise user and admin reserves if memory is added or removed.
3169 *
3170 * The default user reserve max is 128MB, and the default max for the
3171 * admin reserve is 8MB. These are usually, but not always, enough to
3172 * enable recovery from a memory hogging process using login/sshd, a shell,
3173 * and tools like top. It may make sense to increase or even disable the
3174 * reserve depending on the existence of swap or variations in the recovery
3175 * tools. So, the admin may have changed them.
3176 *
3177 * If memory is added and the reserves have been eliminated or increased above
3178 * the default max, then we'll trust the admin.
3179 *
3180 * If memory is removed and there isn't enough free memory, then we
3181 * need to reset the reserves.
3182 *
3183 * Otherwise keep the reserve set by the admin.
3184 */
3185static int reserve_mem_notifier(struct notifier_block *nb,
3186 unsigned long action, void *data)
3187{
3188 unsigned long tmp, free_kbytes;
3189
3190 switch (action) {
3191 case MEM_ONLINE:
3192 /* Default max is 128MB. Leave alone if modified by operator. */
3193 tmp = sysctl_user_reserve_kbytes;
3194 if (0 < tmp && tmp < (1UL << 17))
3195 init_user_reserve();
3196
3197 /* Default max is 8MB. Leave alone if modified by operator. */
3198 tmp = sysctl_admin_reserve_kbytes;
3199 if (0 < tmp && tmp < (1UL << 13))
3200 init_admin_reserve();
3201
3202 break;
3203 case MEM_OFFLINE:
3204 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3205
3206 if (sysctl_user_reserve_kbytes > free_kbytes) {
3207 init_user_reserve();
3208 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3209 sysctl_user_reserve_kbytes);
3210 }
3211
3212 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3213 init_admin_reserve();
3214 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3215 sysctl_admin_reserve_kbytes);
3216 }
3217 break;
3218 default:
3219 break;
3220 }
3221 return NOTIFY_OK;
3222}
3223
3224static struct notifier_block reserve_mem_nb = {
3225 .notifier_call = reserve_mem_notifier,
3226};
3227
3228static int __meminit init_reserve_notifier(void)
3229{
3230 if (register_hotmemory_notifier(&reserve_mem_nb))
3231 printk("Failed registering memory add/remove notifier for admin reserve");
3232
3233 return 0;
3234}
3235module_init(init_reserve_notifier)