blob: 7b36aa7cc89a43c7c5909b7799b77d13106a2929 [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
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07009#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
Cyril Hrubise8420a82013-04-29 15:08:33 -070011#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070013#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/mm.h>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070015#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/shm.h>
17#include <linux/mman.h>
18#include <linux/pagemap.h>
19#include <linux/swap.h>
20#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080021#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/init.h>
23#include <linux/file.h>
24#include <linux/fs.h>
25#include <linux/personality.h>
26#include <linux/security.h>
27#include <linux/hugetlb.h>
28#include <linux/profile.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040029#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/mount.h>
31#include <linux/mempolicy.h>
32#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070033#include <linux/mmu_notifier.h>
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -070034#include <linux/mmdebug.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020035#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040036#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080037#include <linux/khugepaged.h>
Srikar Dronamraju2b144492012-02-09 14:56:42 +053038#include <linux/uprobes.h>
Michel Lespinassed3737182012-12-11 16:01:38 -080039#include <linux/rbtree_augmented.h>
Clark Williamscf4aebc22013-02-07 09:46:59 -060040#include <linux/sched/sysctl.h>
Andrew Shewmaker16408792013-04-29 15:08:12 -070041#include <linux/notifier.h>
42#include <linux/memory.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070043#include <linux/printk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#include <asm/uaccess.h>
46#include <asm/cacheflush.h>
47#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020048#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Jan Beulich42b77722008-07-23 21:27:10 -070050#include "internal.h"
51
Kirill Korotaev3a459752006-09-07 14:17:04 +040052#ifndef arch_mmap_check
53#define arch_mmap_check(addr, len, flags) (0)
54#endif
55
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080056#ifndef arch_rebalance_pgtables
57#define arch_rebalance_pgtables(addr, len) (addr)
58#endif
59
Hugh Dickinse0da3822005-04-19 13:29:15 -070060static void unmap_region(struct mm_struct *mm,
61 struct vm_area_struct *vma, struct vm_area_struct *prev,
62 unsigned long start, unsigned long end);
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/* description of effects of mapping type and prot in current implementation.
65 * this is due to the limited x86 page protection hardware. The expected
66 * behavior is in parens:
67 *
68 * map_type prot
69 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
70 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (yes) yes w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070073 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
75 * w: (no) no w: (no) no w: (copy) copy w: (no) no
76 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
77 *
78 */
79pgprot_t protection_map[16] = {
80 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
81 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
82};
83
Hugh Dickins804af2c2006-07-26 21:39:49 +010084pgprot_t vm_get_page_prot(unsigned long vm_flags)
85{
Dave Kleikampb845f312008-07-08 00:28:51 +100086 return __pgprot(pgprot_val(protection_map[vm_flags &
87 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
88 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010089}
90EXPORT_SYMBOL(vm_get_page_prot);
91
Peter Feiner64e45502014-10-13 15:55:46 -070092static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
93{
94 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
95}
96
97/* Update vma->vm_page_prot to reflect vma->vm_flags. */
98void vma_set_page_prot(struct vm_area_struct *vma)
99{
100 unsigned long vm_flags = vma->vm_flags;
101
102 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
103 if (vma_wants_writenotify(vma)) {
104 vm_flags &= ~VM_SHARED;
105 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
106 vm_flags);
107 }
108}
109
110
Shaohua Li34679d72011-05-24 17:11:18 -0700111int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
112int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800113unsigned long sysctl_overcommit_kbytes __read_mostly;
Christoph Lameterc3d8c142005-09-06 15:16:33 -0700114int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700115unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700116unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -0700117/*
118 * Make sure vm_committed_as in one cacheline and not cacheline shared with
119 * other variables. It can be updated by several CPUs frequently.
120 */
121struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -0800124 * The global memory commitment made in the system can be a metric
125 * that can be used to drive ballooning decisions when Linux is hosted
126 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
127 * balancing memory across competing virtual machines that are hosted.
128 * Several metrics drive this policy engine including the guest reported
129 * memory commitment.
130 */
131unsigned long vm_memory_committed(void)
132{
133 return percpu_counter_read_positive(&vm_committed_as);
134}
135EXPORT_SYMBOL_GPL(vm_memory_committed);
136
137/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 * Check that a process has enough memory to allocate a new virtual
139 * mapping. 0 means there is enough memory for the allocation to
140 * succeed and -ENOMEM implies there is not.
141 *
142 * We currently support three overcommit policies, which are set via the
143 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
144 *
145 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
146 * Additional code 2002 Jul 20 by Robert Love.
147 *
148 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
149 *
150 * Note this is a helper function intended to be used by LSMs which
151 * wish to use this logic.
152 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700153int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700155 unsigned long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -0700157 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
158 -(s64)vm_committed_as_batch * num_online_cpus(),
159 "memory commitment underflow");
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 vm_acct_memory(pages);
162
163 /*
164 * Sometimes we want to use more memory than we have
165 */
166 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
167 return 0;
168
169 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700170 free = global_page_state(NR_FREE_PAGES);
171 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Dmitry Finkc15bef32011-07-25 17:12:19 -0700173 /*
174 * shmem pages shouldn't be counted as free in this
175 * case, they can't be purged, only swapped out, and
176 * that won't affect the overall amount of available
177 * memory in the system.
178 */
179 free -= global_page_state(NR_SHMEM);
180
Shaohua Liec8acf22013-02-22 16:34:38 -0800181 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183 /*
184 * Any slabs which are created with the
185 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
186 * which are reclaimable, under pressure. The dentry
187 * cache and most inode caches should fall into this
188 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700189 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
191 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700192 * Leave reserved pages. The pages are not for anonymous pages.
193 */
194 if (free <= totalreserve_pages)
195 goto error;
196 else
197 free -= totalreserve_pages;
198
199 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700200 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 */
202 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700203 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205 if (free > pages)
206 return 0;
207
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700208 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 }
210
Jerome Marchand00619bc2013-11-12 15:08:31 -0800211 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700213 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 */
215 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700216 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700218 /*
219 * Don't let a single process grow so big a user can't recover
220 */
221 if (mm) {
222 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
223 allowed -= min(mm->total_vm / 32, reserve);
224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700226 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700228error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 vm_unacct_memory(pages);
230
231 return -ENOMEM;
232}
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800235 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 */
237static void __remove_shared_vm_struct(struct vm_area_struct *vma,
238 struct file *file, struct address_space *mapping)
239{
240 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500241 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700243 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 flush_dcache_mmap_lock(mapping);
246 if (unlikely(vma->vm_flags & VM_NONLINEAR))
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700247 list_del_init(&vma->shared.nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700249 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 flush_dcache_mmap_unlock(mapping);
251}
252
253/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700254 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700255 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700257void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
259 struct file *file = vma->vm_file;
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 if (file) {
262 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800263 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800265 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700267}
268
269/*
270 * Close a vm structure and free it, returning the next.
271 */
272static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
273{
274 struct vm_area_struct *next = vma->vm_next;
275
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700276 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 if (vma->vm_ops && vma->vm_ops->close)
278 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700279 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700280 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700281 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700283 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700286static unsigned long do_brk(unsigned long addr, unsigned long len);
287
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100288SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700290 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 unsigned long newbrk, oldbrk;
292 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700293 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800294 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
296 down_write(&mm->mmap_sem);
297
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700298#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800299 /*
300 * CONFIG_COMPAT_BRK can still be overridden by setting
301 * randomize_va_space to 2, which will still cause mm->start_brk
302 * to be arbitrarily shifted
303 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700304 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800305 min_brk = mm->start_brk;
306 else
307 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700308#else
309 min_brk = mm->start_brk;
310#endif
311 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700313
314 /*
315 * Check against rlimit here. If this check is done later after the test
316 * of oldbrk with newbrk then it can escape the test and let the data
317 * segment grow beyond its set limit the in case where the limit is
318 * not page aligned -Ram Gupta
319 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700320 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
321 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700322 goto out;
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 newbrk = PAGE_ALIGN(brk);
325 oldbrk = PAGE_ALIGN(mm->brk);
326 if (oldbrk == newbrk)
327 goto set_brk;
328
329 /* Always allow shrinking brk. */
330 if (brk <= mm->brk) {
331 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
332 goto set_brk;
333 goto out;
334 }
335
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 /* Check against existing mmap mappings. */
337 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
338 goto out;
339
340 /* Ok, looks good - let it rip. */
341 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
342 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344set_brk:
345 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800346 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
347 up_write(&mm->mmap_sem);
348 if (populate)
349 mm_populate(oldbrk, newbrk - oldbrk);
350 return brk;
351
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352out:
353 retval = mm->brk;
354 up_write(&mm->mmap_sem);
355 return retval;
356}
357
Michel Lespinassed3737182012-12-11 16:01:38 -0800358static long vma_compute_subtree_gap(struct vm_area_struct *vma)
359{
360 unsigned long max, subtree_gap;
361 max = vma->vm_start;
362 if (vma->vm_prev)
363 max -= vma->vm_prev->vm_end;
364 if (vma->vm_rb.rb_left) {
365 subtree_gap = rb_entry(vma->vm_rb.rb_left,
366 struct vm_area_struct, vm_rb)->rb_subtree_gap;
367 if (subtree_gap > max)
368 max = subtree_gap;
369 }
370 if (vma->vm_rb.rb_right) {
371 subtree_gap = rb_entry(vma->vm_rb.rb_right,
372 struct vm_area_struct, vm_rb)->rb_subtree_gap;
373 if (subtree_gap > max)
374 max = subtree_gap;
375 }
376 return max;
377}
378
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700379#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380static int browse_rb(struct rb_root *root)
381{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800382 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 struct rb_node *nd, *pn = NULL;
384 unsigned long prev = 0, pend = 0;
385
386 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
387 struct vm_area_struct *vma;
388 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800389 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700390 pr_emerg("vm_start %lx < prev %lx\n",
391 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800392 bug = 1;
393 }
394 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700395 pr_emerg("vm_start %lx < pend %lx\n",
396 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800397 bug = 1;
398 }
399 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700400 pr_emerg("vm_start %lx > vm_end %lx\n",
401 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800402 bug = 1;
403 }
404 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700405 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800406 vma->rb_subtree_gap,
407 vma_compute_subtree_gap(vma));
408 bug = 1;
409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 i++;
411 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800412 prev = vma->vm_start;
413 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 }
415 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800416 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800418 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700419 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800420 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800422 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
Michel Lespinassed3737182012-12-11 16:01:38 -0800425static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
426{
427 struct rb_node *nd;
428
429 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
430 struct vm_area_struct *vma;
431 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700432 VM_BUG_ON_VMA(vma != ignore &&
433 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
434 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700438static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
440 int bug = 0;
441 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800442 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700443 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700444
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700445 while (vma) {
446 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700447
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800448 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700449 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
450 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800451 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800452 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700453 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 i++;
455 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800456 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700457 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800458 bug = 1;
459 }
460 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700461 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700462 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800463 bug = 1;
464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800466 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700467 if (i != -1)
468 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800469 bug = 1;
470 }
Sasha Levin96dad672014-10-09 15:28:39 -0700471 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472}
473#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800474#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475#define validate_mm(mm) do { } while (0)
476#endif
477
Michel Lespinassed3737182012-12-11 16:01:38 -0800478RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
479 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
480
481/*
482 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
483 * vma->vm_prev->vm_end values changed, without modifying the vma's position
484 * in the rbtree.
485 */
486static void vma_gap_update(struct vm_area_struct *vma)
487{
488 /*
489 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
490 * function that does exacltly what we want.
491 */
492 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
493}
494
495static inline void vma_rb_insert(struct vm_area_struct *vma,
496 struct rb_root *root)
497{
498 /* All rb_subtree_gap values must be consistent prior to insertion */
499 validate_mm_rb(root, NULL);
500
501 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
502}
503
504static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
505{
506 /*
507 * All rb_subtree_gap values must be consistent prior to erase,
508 * with the possible exception of the vma being erased.
509 */
510 validate_mm_rb(root, vma);
511
512 /*
513 * Note rb_erase_augmented is a fairly large inline function,
514 * so make sure we instantiate it only once with our desired
515 * augmented rbtree callbacks.
516 */
517 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
518}
519
Michel Lespinassebf181b92012-10-08 16:31:39 -0700520/*
521 * vma has some anon_vma assigned, and is already inserted on that
522 * anon_vma's interval trees.
523 *
524 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
525 * vma must be removed from the anon_vma's interval trees using
526 * anon_vma_interval_tree_pre_update_vma().
527 *
528 * After the update, the vma will be reinserted using
529 * anon_vma_interval_tree_post_update_vma().
530 *
531 * The entire update must be protected by exclusive mmap_sem and by
532 * the root anon_vma's mutex.
533 */
534static inline void
535anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
536{
537 struct anon_vma_chain *avc;
538
539 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
540 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
541}
542
543static inline void
544anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
545{
546 struct anon_vma_chain *avc;
547
548 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
549 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
550}
551
Hugh Dickins6597d782012-10-08 16:29:07 -0700552static int find_vma_links(struct mm_struct *mm, unsigned long addr,
553 unsigned long end, struct vm_area_struct **pprev,
554 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
Hugh Dickins6597d782012-10-08 16:29:07 -0700556 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 __rb_link = &mm->mm_rb.rb_node;
559 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 while (*__rb_link) {
562 struct vm_area_struct *vma_tmp;
563
564 __rb_parent = *__rb_link;
565 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
566
567 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700568 /* Fail if an existing vma overlaps the area */
569 if (vma_tmp->vm_start < end)
570 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 __rb_link = &__rb_parent->rb_left;
572 } else {
573 rb_prev = __rb_parent;
574 __rb_link = &__rb_parent->rb_right;
575 }
576 }
577
578 *pprev = NULL;
579 if (rb_prev)
580 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
581 *rb_link = __rb_link;
582 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700583 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
Cyril Hrubise8420a82013-04-29 15:08:33 -0700586static unsigned long count_vma_pages_range(struct mm_struct *mm,
587 unsigned long addr, unsigned long end)
588{
589 unsigned long nr_pages = 0;
590 struct vm_area_struct *vma;
591
592 /* Find first overlaping mapping */
593 vma = find_vma_intersection(mm, addr, end);
594 if (!vma)
595 return 0;
596
597 nr_pages = (min(end, vma->vm_end) -
598 max(addr, vma->vm_start)) >> PAGE_SHIFT;
599
600 /* Iterate over the rest of the overlaps */
601 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
602 unsigned long overlap_len;
603
604 if (vma->vm_start > end)
605 break;
606
607 overlap_len = min(end, vma->vm_end) - vma->vm_start;
608 nr_pages += overlap_len >> PAGE_SHIFT;
609 }
610
611 return nr_pages;
612}
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
615 struct rb_node **rb_link, struct rb_node *rb_parent)
616{
Michel Lespinassed3737182012-12-11 16:01:38 -0800617 /* Update tracking information for the gap following the new vma. */
618 if (vma->vm_next)
619 vma_gap_update(vma->vm_next);
620 else
621 mm->highest_vm_end = vma->vm_end;
622
623 /*
624 * vma->vm_prev wasn't known when we followed the rbtree to find the
625 * correct insertion point for that vma. As a result, we could not
626 * update the vma vm_rb parents rb_subtree_gap values on the way down.
627 * So, we first insert the vma with a zero rb_subtree_gap value
628 * (to be consistent with what we did on the way down), and then
629 * immediately update the gap to the correct value. Finally we
630 * rebalance the rbtree after all augmented values have been set.
631 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800633 vma->rb_subtree_gap = 0;
634 vma_gap_update(vma);
635 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636}
637
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700638static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
ZhenwenXu48aae422009-01-06 14:40:21 -0800640 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642 file = vma->vm_file;
643 if (file) {
644 struct address_space *mapping = file->f_mapping;
645
646 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500647 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700649 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
651 flush_dcache_mmap_lock(mapping);
652 if (unlikely(vma->vm_flags & VM_NONLINEAR))
653 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
654 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700655 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 flush_dcache_mmap_unlock(mapping);
657 }
658}
659
660static void
661__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
662 struct vm_area_struct *prev, struct rb_node **rb_link,
663 struct rb_node *rb_parent)
664{
665 __vma_link_list(mm, vma, prev, rb_parent);
666 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
669static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
670 struct vm_area_struct *prev, struct rb_node **rb_link,
671 struct rb_node *rb_parent)
672{
673 struct address_space *mapping = NULL;
674
Huang Shijie64ac4942014-06-04 16:07:33 -0700675 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800677 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700678 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
680 __vma_link(mm, vma, prev, rb_link, rb_parent);
681 __vma_link_file(vma);
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800684 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
686 mm->map_count++;
687 validate_mm(mm);
688}
689
690/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700691 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700692 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800694static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
Hugh Dickins6597d782012-10-08 16:29:07 -0700696 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800697 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Hugh Dickins6597d782012-10-08 16:29:07 -0700699 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
700 &prev, &rb_link, &rb_parent))
701 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 __vma_link(mm, vma, prev, rb_link, rb_parent);
703 mm->map_count++;
704}
705
706static inline void
707__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
708 struct vm_area_struct *prev)
709{
Michel Lespinassed3737182012-12-11 16:01:38 -0800710 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700711
Michel Lespinassed3737182012-12-11 16:01:38 -0800712 vma_rb_erase(vma, &mm->mm_rb);
713 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700714 if (next)
715 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700716
717 /* Kill the cache */
718 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719}
720
721/*
722 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
723 * is already present in an i_mmap tree without adjusting the tree.
724 * The following helper function should be used when such adjustments
725 * are necessary. The "insert" vma (if any) is to be inserted
726 * before we drop the necessary locks.
727 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800728int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
730{
731 struct mm_struct *mm = vma->vm_mm;
732 struct vm_area_struct *next = vma->vm_next;
733 struct vm_area_struct *importer = NULL;
734 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700735 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700736 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800738 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 long adjust_next = 0;
740 int remove_next = 0;
741
742 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700743 struct vm_area_struct *exporter = NULL;
744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (end >= next->vm_end) {
746 /*
747 * vma expands, overlapping all the next, and
748 * perhaps the one after too (mprotect case 6).
749 */
750again: remove_next = 1 + (end > next->vm_end);
751 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700752 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 importer = vma;
754 } else if (end > next->vm_start) {
755 /*
756 * vma expands, overlapping part of the next:
757 * mprotect case 5 shifting the boundary up.
758 */
759 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700760 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 importer = vma;
762 } else if (end < vma->vm_end) {
763 /*
764 * vma shrinks, and !insert tells it's not
765 * split_vma inserting another: so it must be
766 * mprotect case 4 shifting the boundary down.
767 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700768 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700769 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 importer = next;
771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Rik van Riel5beb4932010-03-05 13:42:07 -0800773 /*
774 * Easily overlooked: when mprotect shifts the boundary,
775 * make sure the expanding vma has anon_vma set if the
776 * shrinking vma had, to cover any anon pages imported.
777 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700778 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800779 int error;
780
781 error = anon_vma_clone(importer, exporter);
782 if (error)
783 return error;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700784 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800785 }
786 }
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 if (file) {
789 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530790 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530792 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530793
794 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530795 uprobe_munmap(next, next->vm_start,
796 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530797 }
798
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800799 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700802 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 * are visible to arm/parisc __flush_dcache_page
804 * throughout; but we cannot insert into address
805 * space until vma start or end is updated.
806 */
807 __vma_link_file(insert);
808 }
809 }
810
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800811 vma_adjust_trans_huge(vma, start, end, adjust_next);
812
Michel Lespinassebf181b92012-10-08 16:31:39 -0700813 anon_vma = vma->anon_vma;
814 if (!anon_vma && adjust_next)
815 anon_vma = next->anon_vma;
816 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700817 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
818 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000819 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700820 anon_vma_interval_tree_pre_update_vma(vma);
821 if (adjust_next)
822 anon_vma_interval_tree_pre_update_vma(next);
823 }
Rik van Riel012f18002010-08-09 17:18:40 -0700824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 if (root) {
826 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700827 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700829 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
831
Michel Lespinassed3737182012-12-11 16:01:38 -0800832 if (start != vma->vm_start) {
833 vma->vm_start = start;
834 start_changed = true;
835 }
836 if (end != vma->vm_end) {
837 vma->vm_end = end;
838 end_changed = true;
839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 vma->vm_pgoff = pgoff;
841 if (adjust_next) {
842 next->vm_start += adjust_next << PAGE_SHIFT;
843 next->vm_pgoff += adjust_next;
844 }
845
846 if (root) {
847 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700848 vma_interval_tree_insert(next, root);
849 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 flush_dcache_mmap_unlock(mapping);
851 }
852
853 if (remove_next) {
854 /*
855 * vma_merge has merged next into vma, and needs
856 * us to remove next before dropping the locks.
857 */
858 __vma_unlink(mm, next, vma);
859 if (file)
860 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 } else if (insert) {
862 /*
863 * split_vma has split insert from vma, and needs
864 * us to insert it before dropping the locks
865 * (it may either follow vma or precede it).
866 */
867 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800868 } else {
869 if (start_changed)
870 vma_gap_update(vma);
871 if (end_changed) {
872 if (!next)
873 mm->highest_vm_end = end;
874 else if (!adjust_next)
875 vma_gap_update(next);
876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
878
Michel Lespinassebf181b92012-10-08 16:31:39 -0700879 if (anon_vma) {
880 anon_vma_interval_tree_post_update_vma(vma);
881 if (adjust_next)
882 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800883 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800886 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530888 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100889 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530890
891 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100892 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530893 }
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700896 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530897 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700899 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800900 if (next->anon_vma)
901 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700903 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 kmem_cache_free(vm_area_cachep, next);
905 /*
906 * In mprotect's case 6 (see comments on vma_merge),
907 * we must remove another next too. It would clutter
908 * up the code too much to do both in one go.
909 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800910 next = vma->vm_next;
911 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800913 else if (next)
914 vma_gap_update(next);
915 else
916 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530918 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100919 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800922
923 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924}
925
926/*
927 * If the vma has a ->close operation then the driver probably needs to release
928 * per-vma resources, so we don't attempt to merge those.
929 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930static inline int is_mergeable_vma(struct vm_area_struct *vma,
931 struct file *file, unsigned long vm_flags)
932{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800933 /*
934 * VM_SOFTDIRTY should not prevent from VMA merging, if we
935 * match the flags but dirty bit -- the caller should mark
936 * merged VMA as dirty. If dirty bit won't be excluded from
937 * comparison, we increase pressue on the memory system forcing
938 * the kernel to generate new VMAs when old one could be
939 * extended instead.
940 */
941 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 return 0;
943 if (vma->vm_file != file)
944 return 0;
945 if (vma->vm_ops && vma->vm_ops->close)
946 return 0;
947 return 1;
948}
949
950static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700951 struct anon_vma *anon_vma2,
952 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
Shaohua Li965f55d2011-05-24 17:11:20 -0700954 /*
955 * The list_is_singular() test is to avoid merging VMA cloned from
956 * parents. This can improve scalability caused by anon_vma lock.
957 */
958 if ((!anon_vma1 || !anon_vma2) && (!vma ||
959 list_is_singular(&vma->anon_vma_chain)))
960 return 1;
961 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
963
964/*
965 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
966 * in front of (at a lower virtual address and file offset than) the vma.
967 *
968 * We cannot merge two vmas if they have differently assigned (non-NULL)
969 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
970 *
971 * We don't check here for the merged mmap wrapping around the end of pagecache
972 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
973 * wrap, nor mmaps which cover the final page at index -1UL.
974 */
975static int
976can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
977 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
978{
979 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700980 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 if (vma->vm_pgoff == vm_pgoff)
982 return 1;
983 }
984 return 0;
985}
986
987/*
988 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
989 * beyond (at a higher virtual address and file offset than) the vma.
990 *
991 * We cannot merge two vmas if they have differently assigned (non-NULL)
992 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
993 */
994static int
995can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
996 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
997{
998 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700999 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001001 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1003 return 1;
1004 }
1005 return 0;
1006}
1007
1008/*
1009 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1010 * whether that can be merged with its predecessor or its successor.
1011 * Or both (it neatly fills a hole).
1012 *
1013 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1014 * certain not to be mapped by the time vma_merge is called; but when
1015 * called for mprotect, it is certain to be already mapped (either at
1016 * an offset within prev, or at the start of next), and the flags of
1017 * this area are about to be changed to vm_flags - and the no-change
1018 * case has already been eliminated.
1019 *
1020 * The following mprotect cases have to be considered, where AAAA is
1021 * the area passed down from mprotect_fixup, never extending beyond one
1022 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1023 *
1024 * AAAA AAAA AAAA AAAA
1025 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1026 * cannot merge might become might become might become
1027 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1028 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1029 * mremap move: PPPPNNNNNNNN 8
1030 * AAAA
1031 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1032 * might become case 1 below case 2 below case 3 below
1033 *
1034 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1035 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1036 */
1037struct vm_area_struct *vma_merge(struct mm_struct *mm,
1038 struct vm_area_struct *prev, unsigned long addr,
1039 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001040 struct anon_vma *anon_vma, struct file *file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 pgoff_t pgoff, struct mempolicy *policy)
1042{
1043 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1044 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001045 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
1047 /*
1048 * We later require that vma->vm_flags == vm_flags,
1049 * so this tests vma->vm_flags & VM_SPECIAL, too.
1050 */
1051 if (vm_flags & VM_SPECIAL)
1052 return NULL;
1053
1054 if (prev)
1055 next = prev->vm_next;
1056 else
1057 next = mm->mmap;
1058 area = next;
1059 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1060 next = next->vm_next;
1061
1062 /*
1063 * Can it merge with the predecessor?
1064 */
1065 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001066 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 can_vma_merge_after(prev, vm_flags,
1068 anon_vma, file, pgoff)) {
1069 /*
1070 * OK, it can. Can we now merge in the successor as well?
1071 */
1072 if (next && end == next->vm_start &&
1073 mpol_equal(policy, vma_policy(next)) &&
1074 can_vma_merge_before(next, vm_flags,
1075 anon_vma, file, pgoff+pglen) &&
1076 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001077 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001079 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 next->vm_end, prev->vm_pgoff, NULL);
1081 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001082 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001084 if (err)
1085 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001086 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 return prev;
1088 }
1089
1090 /*
1091 * Can this new request be merged in front of next?
1092 */
1093 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001094 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 can_vma_merge_before(next, vm_flags,
1096 anon_vma, file, pgoff+pglen)) {
1097 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001098 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 addr, prev->vm_pgoff, NULL);
1100 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001101 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001103 if (err)
1104 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001105 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 return area;
1107 }
1108
1109 return NULL;
1110}
1111
1112/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001113 * Rough compatbility check to quickly see if it's even worth looking
1114 * at sharing an anon_vma.
1115 *
1116 * They need to have the same vm_file, and the flags can only differ
1117 * in things that mprotect may change.
1118 *
1119 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1120 * we can merge the two vma's. For example, we refuse to merge a vma if
1121 * there is a vm_ops->close() function, because that indicates that the
1122 * driver is doing some kind of reference counting. But that doesn't
1123 * really matter for the anon_vma sharing case.
1124 */
1125static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1126{
1127 return a->vm_end == b->vm_start &&
1128 mpol_equal(vma_policy(a), vma_policy(b)) &&
1129 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001130 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001131 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1132}
1133
1134/*
1135 * Do some basic sanity checking to see if we can re-use the anon_vma
1136 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1137 * the same as 'old', the other will be the new one that is trying
1138 * to share the anon_vma.
1139 *
1140 * NOTE! This runs with mm_sem held for reading, so it is possible that
1141 * the anon_vma of 'old' is concurrently in the process of being set up
1142 * by another page fault trying to merge _that_. But that's ok: if it
1143 * is being set up, that automatically means that it will be a singleton
1144 * acceptable for merging, so we can do all of this optimistically. But
1145 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1146 *
1147 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1148 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1149 * is to return an anon_vma that is "complex" due to having gone through
1150 * a fork).
1151 *
1152 * We also make sure that the two vma's are compatible (adjacent,
1153 * and with the same memory policies). That's all stable, even with just
1154 * a read lock on the mm_sem.
1155 */
1156static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1157{
1158 if (anon_vma_compatible(a, b)) {
1159 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1160
1161 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1162 return anon_vma;
1163 }
1164 return NULL;
1165}
1166
1167/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1169 * neighbouring vmas for a suitable anon_vma, before it goes off
1170 * to allocate a new anon_vma. It checks because a repetitive
1171 * sequence of mprotects and faults may otherwise lead to distinct
1172 * anon_vmas being allocated, preventing vma merge in subsequent
1173 * mprotect.
1174 */
1175struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1176{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001177 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 near = vma->vm_next;
1181 if (!near)
1182 goto try_prev;
1183
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001184 anon_vma = reusable_anon_vma(near, vma, near);
1185 if (anon_vma)
1186 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001188 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 if (!near)
1190 goto none;
1191
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001192 anon_vma = reusable_anon_vma(near, near, vma);
1193 if (anon_vma)
1194 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195none:
1196 /*
1197 * There's no absolute need to look only at touching neighbours:
1198 * we could search further afield for "compatible" anon_vmas.
1199 * But it would probably just be a waste of time searching,
1200 * or lead to too many vmas hanging off the same anon_vma.
1201 * We're trying to allow mprotect remerging later on,
1202 * not trying to minimize memory used for anon_vmas.
1203 */
1204 return NULL;
1205}
1206
1207#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001208void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 struct file *file, long pages)
1210{
1211 const unsigned long stack_flags
1212 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1213
Huang Shijie44de9d02012-07-31 16:41:49 -07001214 mm->total_vm += pages;
1215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (file) {
1217 mm->shared_vm += pages;
1218 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1219 mm->exec_vm += pages;
1220 } else if (flags & stack_flags)
1221 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223#endif /* CONFIG_PROC_FS */
1224
1225/*
Al Viro40401532012-02-13 03:58:52 +00001226 * If a hint addr is less than mmap_min_addr change hint to be as
1227 * low as possible but still greater than mmap_min_addr
1228 */
1229static inline unsigned long round_hint_to_min(unsigned long hint)
1230{
1231 hint &= PAGE_MASK;
1232 if (((void *)hint != NULL) &&
1233 (hint < mmap_min_addr))
1234 return PAGE_ALIGN(mmap_min_addr);
1235 return hint;
1236}
1237
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001238static inline int mlock_future_check(struct mm_struct *mm,
1239 unsigned long flags,
1240 unsigned long len)
1241{
1242 unsigned long locked, lock_limit;
1243
1244 /* mlock MCL_FUTURE? */
1245 if (flags & VM_LOCKED) {
1246 locked = len >> PAGE_SHIFT;
1247 locked += mm->locked_vm;
1248 lock_limit = rlimit(RLIMIT_MEMLOCK);
1249 lock_limit >>= PAGE_SHIFT;
1250 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1251 return -EAGAIN;
1252 }
1253 return 0;
1254}
1255
Al Viro40401532012-02-13 03:58:52 +00001256/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001257 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 */
1259
Al Viroe3fc6292012-05-30 20:08:42 -04001260unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001262 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001263 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264{
vishnu.pscc71aba2014-10-09 15:26:29 -07001265 struct mm_struct *mm = current->mm;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001266 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Michel Lespinasse41badc12013-02-22 16:32:47 -08001268 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 /*
1271 * Does the application expect PROT_READ to imply PROT_EXEC?
1272 *
1273 * (the exception is when the underlying filesystem is noexec
1274 * mounted, in which case we dont add PROT_EXEC.)
1275 */
1276 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001277 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 prot |= PROT_EXEC;
1279
1280 if (!len)
1281 return -EINVAL;
1282
Eric Paris7cd94142007-11-26 18:47:40 -05001283 if (!(flags & MAP_FIXED))
1284 addr = round_hint_to_min(addr);
1285
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 /* Careful about overflows.. */
1287 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001288 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 return -ENOMEM;
1290
1291 /* offset overflow? */
1292 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001293 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
1295 /* Too many mappings? */
1296 if (mm->map_count > sysctl_max_map_count)
1297 return -ENOMEM;
1298
1299 /* Obtain the address to map to. we verify (or select) it and ensure
1300 * that it represents a valid section of the address space.
1301 */
1302 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1303 if (addr & ~PAGE_MASK)
1304 return addr;
1305
1306 /* Do simple checking here so the lower-level routines won't have
1307 * to. we assume access permissions have been handled by the open
1308 * of the memory object, so we don't do any here.
1309 */
1310 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1311 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1312
Huang Shijiecdf7b342009-09-21 17:03:36 -07001313 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (!can_do_mlock())
1315 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001316
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001317 if (mlock_future_check(mm, vm_flags, len))
1318 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001321 struct inode *inode = file_inode(file);
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 switch (flags & MAP_TYPE) {
1324 case MAP_SHARED:
1325 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1326 return -EACCES;
1327
1328 /*
1329 * Make sure we don't allow writing to an append-only
1330 * file..
1331 */
1332 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1333 return -EACCES;
1334
1335 /*
1336 * Make sure there are no mandatory locks on the file.
1337 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001338 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 return -EAGAIN;
1340
1341 vm_flags |= VM_SHARED | VM_MAYSHARE;
1342 if (!(file->f_mode & FMODE_WRITE))
1343 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1344
1345 /* fall through */
1346 case MAP_PRIVATE:
1347 if (!(file->f_mode & FMODE_READ))
1348 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001349 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001350 if (vm_flags & VM_EXEC)
1351 return -EPERM;
1352 vm_flags &= ~VM_MAYEXEC;
1353 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001354
Al Viro72c2d532013-09-22 16:27:52 -04001355 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001356 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001357 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1358 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 break;
1360
1361 default:
1362 return -EINVAL;
1363 }
1364 } else {
1365 switch (flags & MAP_TYPE) {
1366 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001367 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1368 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001369 /*
1370 * Ignore pgoff.
1371 */
1372 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 vm_flags |= VM_SHARED | VM_MAYSHARE;
1374 break;
1375 case MAP_PRIVATE:
1376 /*
1377 * Set pgoff according to addr for anon_vma.
1378 */
1379 pgoff = addr >> PAGE_SHIFT;
1380 break;
1381 default:
1382 return -EINVAL;
1383 }
1384 }
1385
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001386 /*
1387 * Set 'VM_NORESERVE' if we should not account for the
1388 * memory use of this mapping.
1389 */
1390 if (flags & MAP_NORESERVE) {
1391 /* We honor MAP_NORESERVE if allowed to overcommit */
1392 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1393 vm_flags |= VM_NORESERVE;
1394
1395 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1396 if (file && is_file_hugepages(file))
1397 vm_flags |= VM_NORESERVE;
1398 }
1399
1400 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001401 if (!IS_ERR_VALUE(addr) &&
1402 ((vm_flags & VM_LOCKED) ||
1403 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001404 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001405 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001406}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001407
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001408SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1409 unsigned long, prot, unsigned long, flags,
1410 unsigned long, fd, unsigned long, pgoff)
1411{
1412 struct file *file = NULL;
1413 unsigned long retval = -EBADF;
1414
1415 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001416 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001417 file = fget(fd);
1418 if (!file)
1419 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001420 if (is_file_hugepages(file))
1421 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001422 retval = -EINVAL;
1423 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1424 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001425 } else if (flags & MAP_HUGETLB) {
1426 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001427 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001428
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001429 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001430 if (!hs)
1431 return -EINVAL;
1432
1433 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001434 /*
1435 * VM_NORESERVE is used because the reservations will be
1436 * taken when vm_ops->mmap() is called
1437 * A dummy user value is used because we are not locking
1438 * memory so no accounting is necessary
1439 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001440 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001441 VM_NORESERVE,
1442 &user, HUGETLB_ANONHUGE_INODE,
1443 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001444 if (IS_ERR(file))
1445 return PTR_ERR(file);
1446 }
1447
1448 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1449
Al Viroeb36c582012-05-30 20:17:35 -04001450 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001451out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001452 if (file)
1453 fput(file);
1454out:
1455 return retval;
1456}
1457
Christoph Hellwiga4679372010-03-10 15:21:15 -08001458#ifdef __ARCH_WANT_SYS_OLD_MMAP
1459struct mmap_arg_struct {
1460 unsigned long addr;
1461 unsigned long len;
1462 unsigned long prot;
1463 unsigned long flags;
1464 unsigned long fd;
1465 unsigned long offset;
1466};
1467
1468SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1469{
1470 struct mmap_arg_struct a;
1471
1472 if (copy_from_user(&a, arg, sizeof(a)))
1473 return -EFAULT;
1474 if (a.offset & ~PAGE_MASK)
1475 return -EINVAL;
1476
1477 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1478 a.offset >> PAGE_SHIFT);
1479}
1480#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1481
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001482/*
1483 * Some shared mappigns will want the pages marked read-only
1484 * to track write events. If so, we'll downgrade vm_page_prot
1485 * to the private version (using protection_map[] without the
1486 * VM_SHARED bit).
1487 */
1488int vma_wants_writenotify(struct vm_area_struct *vma)
1489{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001490 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001491
1492 /* If it was private or non-writable, the write bit is already clear */
1493 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1494 return 0;
1495
1496 /* The backer wishes to know when pages are first written to? */
1497 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1498 return 1;
1499
Peter Feiner64e45502014-10-13 15:55:46 -07001500 /* The open routine did something to the protections that pgprot_modify
1501 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001502 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001503 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001504 return 0;
1505
Peter Feiner64e45502014-10-13 15:55:46 -07001506 /* Do we need to track softdirty? */
1507 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1508 return 1;
1509
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001510 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001511 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001512 return 0;
1513
1514 /* Can the mapping track the dirty pages? */
1515 return vma->vm_file && vma->vm_file->f_mapping &&
1516 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1517}
1518
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001519/*
1520 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001521 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001522 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001523static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001524{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001525 /*
1526 * hugetlb has its own accounting separate from the core VM
1527 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1528 */
1529 if (file && is_file_hugepages(file))
1530 return 0;
1531
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001532 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1533}
1534
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001535unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001536 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001537{
1538 struct mm_struct *mm = current->mm;
1539 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001540 int error;
1541 struct rb_node **rb_link, *rb_parent;
1542 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001543
Cyril Hrubise8420a82013-04-29 15:08:33 -07001544 /* Check against address space limit. */
1545 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1546 unsigned long nr_pages;
1547
1548 /*
1549 * MAP_FIXED may remove pages of mappings that intersects with
1550 * requested mapping. Account for the pages it would unmap.
1551 */
1552 if (!(vm_flags & MAP_FIXED))
1553 return -ENOMEM;
1554
1555 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1556
1557 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1558 return -ENOMEM;
1559 }
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 /* Clear old maps */
1562 error = -ENOMEM;
1563munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001564 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 if (do_munmap(mm, addr, len))
1566 return -ENOMEM;
1567 goto munmap_back;
1568 }
1569
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001570 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001571 * Private writable mapping: check memory availability
1572 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001573 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001574 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001575 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001576 return -ENOMEM;
1577 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 }
1579
1580 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001581 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001583 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1584 if (vma)
1585 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 /*
1588 * Determine the object being mapped and call the appropriate
1589 * specific mapper. the address has already been validated, but
1590 * not unmapped, but the maps are removed from the list.
1591 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001592 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 if (!vma) {
1594 error = -ENOMEM;
1595 goto unacct_error;
1596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
1598 vma->vm_mm = mm;
1599 vma->vm_start = addr;
1600 vma->vm_end = addr + len;
1601 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001602 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001604 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
1606 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 if (vm_flags & VM_DENYWRITE) {
1608 error = deny_write_access(file);
1609 if (error)
1610 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001612 if (vm_flags & VM_SHARED) {
1613 error = mapping_map_writable(file->f_mapping);
1614 if (error)
1615 goto allow_write_and_free_vma;
1616 }
1617
1618 /* ->mmap() can change vma->vm_file, but must guarantee that
1619 * vma_link() below can deny write-access if VM_DENYWRITE is set
1620 * and map writably if VM_SHARED is set. This usually means the
1621 * new file must not have been exposed to user-space, yet.
1622 */
Al Virocb0942b2012-08-27 14:48:26 -04001623 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 error = file->f_op->mmap(file, vma);
1625 if (error)
1626 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001627
1628 /* Can addr have changed??
1629 *
1630 * Answer: Yes, several device drivers can do it in their
1631 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001632 * Bug: If addr is changed, prev, rb_link, rb_parent should
1633 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001634 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001635 WARN_ON_ONCE(addr != vma->vm_start);
1636
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001637 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001638 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 } else if (vm_flags & VM_SHARED) {
1640 error = shmem_zero_setup(vma);
1641 if (error)
1642 goto free_vma;
1643 }
1644
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001645 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001646 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001647 if (file) {
1648 if (vm_flags & VM_SHARED)
1649 mapping_unmap_writable(file->f_mapping);
1650 if (vm_flags & VM_DENYWRITE)
1651 allow_write_access(file);
1652 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001653 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001654out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001655 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001656
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001657 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001659 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1660 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001661 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001662 else
1663 vma->vm_flags &= ~VM_LOCKED;
1664 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301665
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001666 if (file)
1667 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301668
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001669 /*
1670 * New (or expanded) vma always get soft dirty status.
1671 * Otherwise user-space soft-dirty page tracker won't
1672 * be able to distinguish situation when vma area unmapped,
1673 * then new mapped in-place (which must be aimed as
1674 * a completely new data area).
1675 */
1676 vma->vm_flags |= VM_SOFTDIRTY;
1677
Peter Feiner64e45502014-10-13 15:55:46 -07001678 vma_set_page_prot(vma);
1679
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 return addr;
1681
1682unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 vma->vm_file = NULL;
1684 fput(file);
1685
1686 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001687 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1688 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001689 if (vm_flags & VM_SHARED)
1690 mapping_unmap_writable(file->f_mapping);
1691allow_write_and_free_vma:
1692 if (vm_flags & VM_DENYWRITE)
1693 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694free_vma:
1695 kmem_cache_free(vm_area_cachep, vma);
1696unacct_error:
1697 if (charged)
1698 vm_unacct_memory(charged);
1699 return error;
1700}
1701
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001702unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1703{
1704 /*
1705 * We implement the search by looking for an rbtree node that
1706 * immediately follows a suitable gap. That is,
1707 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1708 * - gap_end = vma->vm_start >= info->low_limit + length;
1709 * - gap_end - gap_start >= length
1710 */
1711
1712 struct mm_struct *mm = current->mm;
1713 struct vm_area_struct *vma;
1714 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1715
1716 /* Adjust search length to account for worst case alignment overhead */
1717 length = info->length + info->align_mask;
1718 if (length < info->length)
1719 return -ENOMEM;
1720
1721 /* Adjust search limits by the desired length */
1722 if (info->high_limit < length)
1723 return -ENOMEM;
1724 high_limit = info->high_limit - length;
1725
1726 if (info->low_limit > high_limit)
1727 return -ENOMEM;
1728 low_limit = info->low_limit + length;
1729
1730 /* Check if rbtree root looks promising */
1731 if (RB_EMPTY_ROOT(&mm->mm_rb))
1732 goto check_highest;
1733 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1734 if (vma->rb_subtree_gap < length)
1735 goto check_highest;
1736
1737 while (true) {
1738 /* Visit left subtree if it looks promising */
1739 gap_end = vma->vm_start;
1740 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1741 struct vm_area_struct *left =
1742 rb_entry(vma->vm_rb.rb_left,
1743 struct vm_area_struct, vm_rb);
1744 if (left->rb_subtree_gap >= length) {
1745 vma = left;
1746 continue;
1747 }
1748 }
1749
1750 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1751check_current:
1752 /* Check if current node has a suitable gap */
1753 if (gap_start > high_limit)
1754 return -ENOMEM;
1755 if (gap_end >= low_limit && gap_end - gap_start >= length)
1756 goto found;
1757
1758 /* Visit right subtree if it looks promising */
1759 if (vma->vm_rb.rb_right) {
1760 struct vm_area_struct *right =
1761 rb_entry(vma->vm_rb.rb_right,
1762 struct vm_area_struct, vm_rb);
1763 if (right->rb_subtree_gap >= length) {
1764 vma = right;
1765 continue;
1766 }
1767 }
1768
1769 /* Go back up the rbtree to find next candidate node */
1770 while (true) {
1771 struct rb_node *prev = &vma->vm_rb;
1772 if (!rb_parent(prev))
1773 goto check_highest;
1774 vma = rb_entry(rb_parent(prev),
1775 struct vm_area_struct, vm_rb);
1776 if (prev == vma->vm_rb.rb_left) {
1777 gap_start = vma->vm_prev->vm_end;
1778 gap_end = vma->vm_start;
1779 goto check_current;
1780 }
1781 }
1782 }
1783
1784check_highest:
1785 /* Check highest gap, which does not precede any rbtree node */
1786 gap_start = mm->highest_vm_end;
1787 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1788 if (gap_start > high_limit)
1789 return -ENOMEM;
1790
1791found:
1792 /* We found a suitable gap. Clip it with the original low_limit. */
1793 if (gap_start < info->low_limit)
1794 gap_start = info->low_limit;
1795
1796 /* Adjust gap address to the desired alignment */
1797 gap_start += (info->align_offset - gap_start) & info->align_mask;
1798
1799 VM_BUG_ON(gap_start + info->length > info->high_limit);
1800 VM_BUG_ON(gap_start + info->length > gap_end);
1801 return gap_start;
1802}
1803
1804unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1805{
1806 struct mm_struct *mm = current->mm;
1807 struct vm_area_struct *vma;
1808 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1809
1810 /* Adjust search length to account for worst case alignment overhead */
1811 length = info->length + info->align_mask;
1812 if (length < info->length)
1813 return -ENOMEM;
1814
1815 /*
1816 * Adjust search limits by the desired length.
1817 * See implementation comment at top of unmapped_area().
1818 */
1819 gap_end = info->high_limit;
1820 if (gap_end < length)
1821 return -ENOMEM;
1822 high_limit = gap_end - length;
1823
1824 if (info->low_limit > high_limit)
1825 return -ENOMEM;
1826 low_limit = info->low_limit + length;
1827
1828 /* Check highest gap, which does not precede any rbtree node */
1829 gap_start = mm->highest_vm_end;
1830 if (gap_start <= high_limit)
1831 goto found_highest;
1832
1833 /* Check if rbtree root looks promising */
1834 if (RB_EMPTY_ROOT(&mm->mm_rb))
1835 return -ENOMEM;
1836 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1837 if (vma->rb_subtree_gap < length)
1838 return -ENOMEM;
1839
1840 while (true) {
1841 /* Visit right subtree if it looks promising */
1842 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1843 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1844 struct vm_area_struct *right =
1845 rb_entry(vma->vm_rb.rb_right,
1846 struct vm_area_struct, vm_rb);
1847 if (right->rb_subtree_gap >= length) {
1848 vma = right;
1849 continue;
1850 }
1851 }
1852
1853check_current:
1854 /* Check if current node has a suitable gap */
1855 gap_end = vma->vm_start;
1856 if (gap_end < low_limit)
1857 return -ENOMEM;
1858 if (gap_start <= high_limit && gap_end - gap_start >= length)
1859 goto found;
1860
1861 /* Visit left subtree if it looks promising */
1862 if (vma->vm_rb.rb_left) {
1863 struct vm_area_struct *left =
1864 rb_entry(vma->vm_rb.rb_left,
1865 struct vm_area_struct, vm_rb);
1866 if (left->rb_subtree_gap >= length) {
1867 vma = left;
1868 continue;
1869 }
1870 }
1871
1872 /* Go back up the rbtree to find next candidate node */
1873 while (true) {
1874 struct rb_node *prev = &vma->vm_rb;
1875 if (!rb_parent(prev))
1876 return -ENOMEM;
1877 vma = rb_entry(rb_parent(prev),
1878 struct vm_area_struct, vm_rb);
1879 if (prev == vma->vm_rb.rb_right) {
1880 gap_start = vma->vm_prev ?
1881 vma->vm_prev->vm_end : 0;
1882 goto check_current;
1883 }
1884 }
1885 }
1886
1887found:
1888 /* We found a suitable gap. Clip it with the original high_limit. */
1889 if (gap_end > info->high_limit)
1890 gap_end = info->high_limit;
1891
1892found_highest:
1893 /* Compute highest gap address at the desired alignment */
1894 gap_end -= info->length;
1895 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1896
1897 VM_BUG_ON(gap_end < info->low_limit);
1898 VM_BUG_ON(gap_end < gap_start);
1899 return gap_end;
1900}
1901
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902/* Get an address range which is currently unmapped.
1903 * For shmat() with addr=0.
1904 *
1905 * Ugly calling convention alert:
1906 * Return value with the low bits set means error value,
1907 * ie
1908 * if (ret & ~PAGE_MASK)
1909 * error = ret;
1910 *
1911 * This function "knows" that -ENOMEM has the bits set.
1912 */
1913#ifndef HAVE_ARCH_UNMAPPED_AREA
1914unsigned long
1915arch_get_unmapped_area(struct file *filp, unsigned long addr,
1916 unsigned long len, unsigned long pgoff, unsigned long flags)
1917{
1918 struct mm_struct *mm = current->mm;
1919 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001920 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001922 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 return -ENOMEM;
1924
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001925 if (flags & MAP_FIXED)
1926 return addr;
1927
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 if (addr) {
1929 addr = PAGE_ALIGN(addr);
1930 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001931 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 (!vma || addr + len <= vma->vm_start))
1933 return addr;
1934 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001936 info.flags = 0;
1937 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001938 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001939 info.high_limit = TASK_SIZE;
1940 info.align_mask = 0;
1941 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942}
vishnu.pscc71aba2014-10-09 15:26:29 -07001943#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945/*
1946 * This mmap-allocator allocates new areas top-down from below the
1947 * stack's low limit (the base):
1948 */
1949#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1950unsigned long
1951arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1952 const unsigned long len, const unsigned long pgoff,
1953 const unsigned long flags)
1954{
1955 struct vm_area_struct *vma;
1956 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001957 unsigned long addr = addr0;
1958 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001961 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 return -ENOMEM;
1963
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001964 if (flags & MAP_FIXED)
1965 return addr;
1966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 /* requesting a specific address */
1968 if (addr) {
1969 addr = PAGE_ALIGN(addr);
1970 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001971 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 (!vma || addr + len <= vma->vm_start))
1973 return addr;
1974 }
1975
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001976 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1977 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001978 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001979 info.high_limit = mm->mmap_base;
1980 info.align_mask = 0;
1981 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 /*
1984 * A failed mmap() very likely causes application failure,
1985 * so fall back to the bottom-up function here. This scenario
1986 * can happen with large stack limits and large mmap()
1987 * allocations.
1988 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001989 if (addr & ~PAGE_MASK) {
1990 VM_BUG_ON(addr != -ENOMEM);
1991 info.flags = 0;
1992 info.low_limit = TASK_UNMAPPED_BASE;
1993 info.high_limit = TASK_SIZE;
1994 addr = vm_unmapped_area(&info);
1995 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
1997 return addr;
1998}
1999#endif
2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001unsigned long
2002get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2003 unsigned long pgoff, unsigned long flags)
2004{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002005 unsigned long (*get_area)(struct file *, unsigned long,
2006 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
Al Viro9206de92009-12-03 15:23:11 -05002008 unsigned long error = arch_mmap_check(addr, len, flags);
2009 if (error)
2010 return error;
2011
2012 /* Careful about overflows.. */
2013 if (len > TASK_SIZE)
2014 return -ENOMEM;
2015
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002016 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04002017 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002018 get_area = file->f_op->get_unmapped_area;
2019 addr = get_area(file, addr, len, pgoff, flags);
2020 if (IS_ERR_VALUE(addr))
2021 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002023 if (addr > TASK_SIZE - len)
2024 return -ENOMEM;
2025 if (addr & ~PAGE_MASK)
2026 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002027
Al Viro9ac4ed42012-05-30 17:13:15 -04002028 addr = arch_rebalance_pgtables(addr, len);
2029 error = security_mmap_addr(addr);
2030 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031}
2032
2033EXPORT_SYMBOL(get_unmapped_area);
2034
2035/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002036struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002038 struct rb_node *rb_node;
2039 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002041 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002042 vma = vmacache_find(mm, addr);
2043 if (likely(vma))
2044 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002046 rb_node = mm->mm_rb.rb_node;
2047 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002049 while (rb_node) {
2050 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002052 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002053
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002054 if (tmp->vm_end > addr) {
2055 vma = tmp;
2056 if (tmp->vm_start <= addr)
2057 break;
2058 rb_node = rb_node->rb_left;
2059 } else
2060 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002062
2063 if (vma)
2064 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 return vma;
2066}
2067
2068EXPORT_SYMBOL(find_vma);
2069
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002070/*
2071 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002072 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073struct vm_area_struct *
2074find_vma_prev(struct mm_struct *mm, unsigned long addr,
2075 struct vm_area_struct **pprev)
2076{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002077 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002079 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002080 if (vma) {
2081 *pprev = vma->vm_prev;
2082 } else {
2083 struct rb_node *rb_node = mm->mm_rb.rb_node;
2084 *pprev = NULL;
2085 while (rb_node) {
2086 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2087 rb_node = rb_node->rb_right;
2088 }
2089 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002090 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091}
2092
2093/*
2094 * Verify that the stack growth is acceptable and
2095 * update accounting. This is shared with both the
2096 * grow-up and grow-down cases.
2097 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002098static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099{
2100 struct mm_struct *mm = vma->vm_mm;
2101 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08002102 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
2104 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002105 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return -ENOMEM;
2107
2108 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08002109 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 return -ENOMEM;
2111
2112 /* mlock limit tests */
2113 if (vma->vm_flags & VM_LOCKED) {
2114 unsigned long locked;
2115 unsigned long limit;
2116 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002117 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2118 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 if (locked > limit && !capable(CAP_IPC_LOCK))
2120 return -ENOMEM;
2121 }
2122
Adam Litke0d59a012007-01-30 14:35:39 -08002123 /* Check to ensure the stack will not grow into a hugetlb-only region */
2124 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2125 vma->vm_end - size;
2126 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2127 return -EFAULT;
2128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 /*
2130 * Overcommit.. This must be the final test, as it will
2131 * update security statistics.
2132 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002133 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 return -ENOMEM;
2135
2136 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 if (vma->vm_flags & VM_LOCKED)
2138 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002139 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 return 0;
2141}
2142
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002143#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002145 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2146 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002148int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149{
2150 int error;
2151
2152 if (!(vma->vm_flags & VM_GROWSUP))
2153 return -EFAULT;
2154
2155 /*
2156 * We must make sure the anon_vma is allocated
2157 * so that the anon_vma locking is not a noop.
2158 */
2159 if (unlikely(anon_vma_prepare(vma)))
2160 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002161 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
2163 /*
2164 * vma->vm_start/vm_end cannot change under us because the caller
2165 * is required to hold the mmap_sem in read mode. We need the
2166 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002167 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 */
Helge Deller06b32f32006-12-19 19:28:33 +01002169 if (address < PAGE_ALIGN(address+4))
2170 address = PAGE_ALIGN(address+4);
2171 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002172 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002173 return -ENOMEM;
2174 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 error = 0;
2176
2177 /* Somebody else might have raced and expanded it already */
2178 if (address > vma->vm_end) {
2179 unsigned long size, grow;
2180
2181 size = address - vma->vm_start;
2182 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2183
Hugh Dickins42c36f62011-05-09 17:44:42 -07002184 error = -ENOMEM;
2185 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2186 error = acct_stack_growth(vma, size, grow);
2187 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002188 /*
2189 * vma_gap_update() doesn't support concurrent
2190 * updates, but we only hold a shared mmap_sem
2191 * lock here, so we need to protect against
2192 * concurrent vma expansions.
2193 * vma_lock_anon_vma() doesn't help here, as
2194 * we don't guarantee that all growable vmas
2195 * in a mm share the same root anon vma.
2196 * So, we reuse mm->page_table_lock to guard
2197 * against concurrent vma expansions.
2198 */
2199 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002200 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002201 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002202 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002203 if (vma->vm_next)
2204 vma_gap_update(vma->vm_next);
2205 else
2206 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002207 spin_unlock(&vma->vm_mm->page_table_lock);
2208
Hugh Dickins42c36f62011-05-09 17:44:42 -07002209 perf_event_mmap(vma);
2210 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002213 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002214 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002215 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 return error;
2217}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002218#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220/*
2221 * vma is the first one with address < vma->vm_start. Have to extend vma.
2222 */
Michal Hockod05f3162011-05-24 17:11:44 -07002223int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002224 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225{
2226 int error;
2227
2228 /*
2229 * We must make sure the anon_vma is allocated
2230 * so that the anon_vma locking is not a noop.
2231 */
2232 if (unlikely(anon_vma_prepare(vma)))
2233 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002234
2235 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002236 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002237 if (error)
2238 return error;
2239
Rik van Rielbb4a3402010-08-09 17:18:37 -07002240 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241
2242 /*
2243 * vma->vm_start/vm_end cannot change under us because the caller
2244 * is required to hold the mmap_sem in read mode. We need the
2245 * anon_vma lock to serialize against concurrent expand_stacks.
2246 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 /* Somebody else might have raced and expanded it already */
2249 if (address < vma->vm_start) {
2250 unsigned long size, grow;
2251
2252 size = vma->vm_end - address;
2253 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2254
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002255 error = -ENOMEM;
2256 if (grow <= vma->vm_pgoff) {
2257 error = acct_stack_growth(vma, size, grow);
2258 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002259 /*
2260 * vma_gap_update() doesn't support concurrent
2261 * updates, but we only hold a shared mmap_sem
2262 * lock here, so we need to protect against
2263 * concurrent vma expansions.
2264 * vma_lock_anon_vma() doesn't help here, as
2265 * we don't guarantee that all growable vmas
2266 * in a mm share the same root anon vma.
2267 * So, we reuse mm->page_table_lock to guard
2268 * against concurrent vma expansions.
2269 */
2270 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002271 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002272 vma->vm_start = address;
2273 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002274 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002275 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002276 spin_unlock(&vma->vm_mm->page_table_lock);
2277
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002278 perf_event_mmap(vma);
2279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 }
2281 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002282 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002283 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002284 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 return error;
2286}
2287
Linus Torvalds09884962013-02-27 08:36:04 -08002288/*
2289 * Note how expand_stack() refuses to expand the stack all the way to
2290 * abut the next virtual mapping, *unless* that mapping itself is also
2291 * a stack mapping. We want to leave room for a guard page, after all
2292 * (the guard page itself is not added here, that is done by the
2293 * actual page faulting logic)
2294 *
2295 * This matches the behavior of the guard page logic (see mm/memory.c:
2296 * check_stack_guard_page()), which only allows the guard page to be
2297 * removed under these circumstances.
2298 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002299#ifdef CONFIG_STACK_GROWSUP
2300int expand_stack(struct vm_area_struct *vma, unsigned long address)
2301{
Linus Torvalds09884962013-02-27 08:36:04 -08002302 struct vm_area_struct *next;
2303
2304 address &= PAGE_MASK;
2305 next = vma->vm_next;
2306 if (next && next->vm_start == address + PAGE_SIZE) {
2307 if (!(next->vm_flags & VM_GROWSUP))
2308 return -ENOMEM;
2309 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002310 return expand_upwards(vma, address);
2311}
2312
2313struct vm_area_struct *
2314find_extend_vma(struct mm_struct *mm, unsigned long addr)
2315{
2316 struct vm_area_struct *vma, *prev;
2317
2318 addr &= PAGE_MASK;
2319 vma = find_vma_prev(mm, addr, &prev);
2320 if (vma && (vma->vm_start <= addr))
2321 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002322 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002323 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002324 if (prev->vm_flags & VM_LOCKED)
2325 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002326 return prev;
2327}
2328#else
2329int expand_stack(struct vm_area_struct *vma, unsigned long address)
2330{
Linus Torvalds09884962013-02-27 08:36:04 -08002331 struct vm_area_struct *prev;
2332
2333 address &= PAGE_MASK;
2334 prev = vma->vm_prev;
2335 if (prev && prev->vm_end == address) {
2336 if (!(prev->vm_flags & VM_GROWSDOWN))
2337 return -ENOMEM;
2338 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002339 return expand_downwards(vma, address);
2340}
2341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002343find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344{
vishnu.pscc71aba2014-10-09 15:26:29 -07002345 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 unsigned long start;
2347
2348 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002349 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 if (!vma)
2351 return NULL;
2352 if (vma->vm_start <= addr)
2353 return vma;
2354 if (!(vma->vm_flags & VM_GROWSDOWN))
2355 return NULL;
2356 start = vma->vm_start;
2357 if (expand_stack(vma, addr))
2358 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002359 if (vma->vm_flags & VM_LOCKED)
2360 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 return vma;
2362}
2363#endif
2364
Jesse Barnese1d6d012014-12-12 16:55:27 -08002365EXPORT_SYMBOL_GPL(find_extend_vma);
2366
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002368 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002370 *
2371 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002373static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002375 unsigned long nr_accounted = 0;
2376
Hugh Dickins365e9c872005-10-29 18:16:18 -07002377 /* Update high watermark before we lower total_vm */
2378 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002380 long nrpages = vma_pages(vma);
2381
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002382 if (vma->vm_flags & VM_ACCOUNT)
2383 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002384 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002385 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002386 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002387 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 validate_mm(mm);
2389}
2390
2391/*
2392 * Get rid of page table information in the indicated region.
2393 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002394 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 */
2396static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002397 struct vm_area_struct *vma, struct vm_area_struct *prev,
2398 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399{
vishnu.pscc71aba2014-10-09 15:26:29 -07002400 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002401 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
2403 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002404 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002405 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002406 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002407 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002408 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002409 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410}
2411
2412/*
2413 * Create a list of vma's touched by the unmap, removing them from the mm's
2414 * vma list as we go..
2415 */
2416static void
2417detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2418 struct vm_area_struct *prev, unsigned long end)
2419{
2420 struct vm_area_struct **insertion_point;
2421 struct vm_area_struct *tail_vma = NULL;
2422
2423 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002424 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002426 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 mm->map_count--;
2428 tail_vma = vma;
2429 vma = vma->vm_next;
2430 } while (vma && vma->vm_start < end);
2431 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002432 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002433 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002434 vma_gap_update(vma);
2435 } else
2436 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002438
2439 /* Kill the cache */
2440 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441}
2442
2443/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002444 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2445 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002447static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 unsigned long addr, int new_below)
2449{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08002451 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
Andi Kleena5516432008-07-23 21:27:41 -07002453 if (is_vm_hugetlb_page(vma) && (addr &
2454 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 return -EINVAL;
2456
Christoph Lametere94b1762006-12-06 20:33:17 -08002457 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002459 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
2461 /* most fields are the same, copy all, and then fixup */
2462 *new = *vma;
2463
Rik van Riel5beb4932010-03-05 13:42:07 -08002464 INIT_LIST_HEAD(&new->anon_vma_chain);
2465
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 if (new_below)
2467 new->vm_end = addr;
2468 else {
2469 new->vm_start = addr;
2470 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2471 }
2472
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002473 err = vma_dup_policy(vma, new);
2474 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002475 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002477 err = anon_vma_clone(new, vma);
2478 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002479 goto out_free_mpol;
2480
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002481 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 get_file(new->vm_file);
2483
2484 if (new->vm_ops && new->vm_ops->open)
2485 new->vm_ops->open(new);
2486
2487 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002488 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2490 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002491 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Rik van Riel5beb4932010-03-05 13:42:07 -08002493 /* Success. */
2494 if (!err)
2495 return 0;
2496
2497 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002498 if (new->vm_ops && new->vm_ops->close)
2499 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002500 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002501 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002502 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002503 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002504 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002505 out_free_vma:
2506 kmem_cache_free(vm_area_cachep, new);
2507 out_err:
2508 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509}
2510
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002511/*
2512 * Split a vma into two pieces at address 'addr', a new vma is allocated
2513 * either for the first part or the tail.
2514 */
2515int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2516 unsigned long addr, int new_below)
2517{
2518 if (mm->map_count >= sysctl_max_map_count)
2519 return -ENOMEM;
2520
2521 return __split_vma(mm, vma, addr, new_below);
2522}
2523
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524/* Munmap is split into 2 main parts -- this part which finds
2525 * what needs doing, and the areas themselves, which do the
2526 * work. This now handles partial unmappings.
2527 * Jeremy Fitzhardinge <jeremy@goop.org>
2528 */
2529int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2530{
2531 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002532 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
2534 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2535 return -EINVAL;
2536
vishnu.pscc71aba2014-10-09 15:26:29 -07002537 len = PAGE_ALIGN(len);
2538 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 return -EINVAL;
2540
2541 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002542 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002543 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002545 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002546 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
2548 /* if it doesn't overlap, we have nothing.. */
2549 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002550 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 return 0;
2552
2553 /*
2554 * If we need to split any vma, do it now to save pain later.
2555 *
2556 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2557 * unmapped vm_area_struct will remain in use: so lower split_vma
2558 * places tmp vma above, and higher split_vma places tmp vma below.
2559 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002560 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002561 int error;
2562
2563 /*
2564 * Make sure that map_count on return from munmap() will
2565 * not exceed its limit; but let map_count go just above
2566 * its limit temporarily, to help free resources as expected.
2567 */
2568 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2569 return -ENOMEM;
2570
2571 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 if (error)
2573 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002574 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 }
2576
2577 /* Does it split the last one? */
2578 last = find_vma(mm, end);
2579 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002580 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 if (error)
2582 return error;
2583 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002584 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
2586 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002587 * unlock any mlock()ed ranges before detaching vmas
2588 */
2589 if (mm->locked_vm) {
2590 struct vm_area_struct *tmp = vma;
2591 while (tmp && tmp->vm_start < end) {
2592 if (tmp->vm_flags & VM_LOCKED) {
2593 mm->locked_vm -= vma_pages(tmp);
2594 munlock_vma_pages_all(tmp);
2595 }
2596 tmp = tmp->vm_next;
2597 }
2598 }
2599
2600 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 * Remove the vma's, and unmap the actual pages
2602 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002603 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2604 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Dave Hansen1de4fa12014-11-14 07:18:31 -08002606 arch_unmap(mm, vma, start, end);
2607
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002609 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
2611 return 0;
2612}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Al Virobfce2812012-04-20 21:57:04 -04002614int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002615{
2616 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002617 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002618
2619 down_write(&mm->mmap_sem);
2620 ret = do_munmap(mm, start, len);
2621 up_write(&mm->mmap_sem);
2622 return ret;
2623}
2624EXPORT_SYMBOL(vm_munmap);
2625
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002626SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002629 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630}
2631
2632static inline void verify_mm_writelocked(struct mm_struct *mm)
2633{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002634#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2636 WARN_ON(1);
2637 up_read(&mm->mmap_sem);
2638 }
2639#endif
2640}
2641
2642/*
2643 * this is really a simplified "do_mmap". it only handles
2644 * anonymous maps. eventually we may be able to do some
2645 * brk-specific accounting here.
2646 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002647static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648{
vishnu.pscc71aba2014-10-09 15:26:29 -07002649 struct mm_struct *mm = current->mm;
2650 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002652 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002654 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
2656 len = PAGE_ALIGN(len);
2657 if (!len)
2658 return addr;
2659
Kirill Korotaev3a459752006-09-07 14:17:04 +04002660 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2661
Al Viro2c6a1012009-12-03 19:40:46 -05002662 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2663 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002664 return error;
2665
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002666 error = mlock_future_check(mm, mm->def_flags, len);
2667 if (error)
2668 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
2670 /*
2671 * mm->mmap_sem is required to protect against another thread
2672 * changing the mappings in case we sleep.
2673 */
2674 verify_mm_writelocked(mm);
2675
2676 /*
2677 * Clear old maps. this also does some error checking for us
2678 */
2679 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002680 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 if (do_munmap(mm, addr, len))
2682 return -ENOMEM;
2683 goto munmap_back;
2684 }
2685
2686 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002687 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 return -ENOMEM;
2689
2690 if (mm->map_count > sysctl_max_map_count)
2691 return -ENOMEM;
2692
Al Viro191c5422012-02-13 03:58:52 +00002693 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 return -ENOMEM;
2695
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002697 vma = vma_merge(mm, prev, addr, addr + len, flags,
2698 NULL, NULL, pgoff, NULL);
2699 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 goto out;
2701
2702 /*
2703 * create a vma struct for an anonymous mapping
2704 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002705 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 if (!vma) {
2707 vm_unacct_memory(len >> PAGE_SHIFT);
2708 return -ENOMEM;
2709 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
Rik van Riel5beb4932010-03-05 13:42:07 -08002711 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 vma->vm_mm = mm;
2713 vma->vm_start = addr;
2714 vma->vm_end = addr + len;
2715 vma->vm_pgoff = pgoff;
2716 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002717 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 vma_link(mm, vma, prev, rb_link, rb_parent);
2719out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002720 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002722 if (flags & VM_LOCKED)
2723 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002724 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 return addr;
2726}
2727
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002728unsigned long vm_brk(unsigned long addr, unsigned long len)
2729{
2730 struct mm_struct *mm = current->mm;
2731 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002732 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002733
2734 down_write(&mm->mmap_sem);
2735 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002736 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002737 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002738 if (populate)
2739 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002740 return ret;
2741}
2742EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743
2744/* Release all mmaps. */
2745void exit_mmap(struct mm_struct *mm)
2746{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002747 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002748 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 unsigned long nr_accounted = 0;
2750
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002751 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002752 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002753
Rik van Rielba470de2008-10-18 20:26:50 -07002754 if (mm->locked_vm) {
2755 vma = mm->mmap;
2756 while (vma) {
2757 if (vma->vm_flags & VM_LOCKED)
2758 munlock_vma_pages_all(vma);
2759 vma = vma->vm_next;
2760 }
2761 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002762
2763 arch_exit_mmap(mm);
2764
Rik van Rielba470de2008-10-18 20:26:50 -07002765 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002766 if (!vma) /* Can happen if dup_mmap() received an OOM */
2767 return;
2768
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002771 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002772 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002773 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002774 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002775
Hugh Dickins6ee86302013-04-29 15:07:44 -07002776 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002777 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002780 * Walk the list again, actually closing and freeing it,
2781 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002783 while (vma) {
2784 if (vma->vm_flags & VM_ACCOUNT)
2785 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002786 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002787 }
2788 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002789
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -08002790 WARN_ON(atomic_long_read(&mm->nr_ptes) >
2791 (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792}
2793
2794/* Insert vm structure into process list sorted by address
2795 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002796 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002798int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799{
Hugh Dickins6597d782012-10-08 16:29:07 -07002800 struct vm_area_struct *prev;
2801 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802
2803 /*
2804 * The vm_pgoff of a purely anonymous vma should be irrelevant
2805 * until its first write fault, when page's anon_vma and index
2806 * are set. But now set the vm_pgoff it will almost certainly
2807 * end up with (unless mremap moves it elsewhere before that
2808 * first wfault), so /proc/pid/maps tells a consistent story.
2809 *
2810 * By setting it to reflect the virtual start address of the
2811 * vma, merges and splits can happen in a seamless way, just
2812 * using the existing file pgoff checks and manipulations.
2813 * Similarly in do_mmap_pgoff and in do_brk.
2814 */
2815 if (!vma->vm_file) {
2816 BUG_ON(vma->anon_vma);
2817 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2818 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002819 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2820 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002822 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002823 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002824 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302825
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 vma_link(mm, vma, prev, rb_link, rb_parent);
2827 return 0;
2828}
2829
2830/*
2831 * Copy the vma structure to a new location in the same mm,
2832 * prior to moving page table entries, to effect an mremap move.
2833 */
2834struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002835 unsigned long addr, unsigned long len, pgoff_t pgoff,
2836 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837{
2838 struct vm_area_struct *vma = *vmap;
2839 unsigned long vma_start = vma->vm_start;
2840 struct mm_struct *mm = vma->vm_mm;
2841 struct vm_area_struct *new_vma, *prev;
2842 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002843 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
2845 /*
2846 * If anonymous vma has not yet been faulted, update new pgoff
2847 * to match new location, to increase its chance of merging.
2848 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002849 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002851 faulted_in_anon_vma = false;
2852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853
Hugh Dickins6597d782012-10-08 16:29:07 -07002854 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2855 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2857 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2858 if (new_vma) {
2859 /*
2860 * Source vma may have been merged into new_vma
2861 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002862 if (unlikely(vma_start >= new_vma->vm_start &&
2863 vma_start < new_vma->vm_end)) {
2864 /*
2865 * The only way we can get a vma_merge with
2866 * self during an mremap is if the vma hasn't
2867 * been faulted in yet and we were allowed to
2868 * reset the dst vma->vm_pgoff to the
2869 * destination address of the mremap to allow
2870 * the merge to happen. mremap must change the
2871 * vm_pgoff linearity between src and dst vmas
2872 * (in turn preventing a vma_merge) to be
2873 * safe. It is only safe to keep the vm_pgoff
2874 * linear if there are no pages mapped yet.
2875 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002876 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002877 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002878 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002879 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002881 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 if (new_vma) {
2883 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 new_vma->vm_start = addr;
2885 new_vma->vm_end = addr + len;
2886 new_vma->vm_pgoff = pgoff;
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002887 if (vma_dup_policy(vma, new_vma))
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002888 goto out_free_vma;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002889 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2890 if (anon_vma_clone(new_vma, vma))
2891 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002892 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 get_file(new_vma->vm_file);
2894 if (new_vma->vm_ops && new_vma->vm_ops->open)
2895 new_vma->vm_ops->open(new_vma);
2896 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002897 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 }
2899 }
2900 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002901
2902 out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002903 mpol_put(vma_policy(new_vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002904 out_free_vma:
2905 kmem_cache_free(vm_area_cachep, new_vma);
2906 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002908
2909/*
2910 * Return true if the calling process may expand its vm space by the passed
2911 * number of pages
2912 */
2913int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2914{
2915 unsigned long cur = mm->total_vm; /* pages */
2916 unsigned long lim;
2917
Jiri Slaby59e99e52010-03-05 13:41:44 -08002918 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002919
2920 if (cur + npages > lim)
2921 return 0;
2922 return 1;
2923}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002924
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002925static int special_mapping_fault(struct vm_area_struct *vma,
2926 struct vm_fault *vmf);
2927
2928/*
2929 * Having a close hook prevents vma merging regardless of flags.
2930 */
2931static void special_mapping_close(struct vm_area_struct *vma)
2932{
2933}
2934
2935static const char *special_mapping_name(struct vm_area_struct *vma)
2936{
2937 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
2938}
2939
2940static const struct vm_operations_struct special_mapping_vmops = {
2941 .close = special_mapping_close,
2942 .fault = special_mapping_fault,
2943 .name = special_mapping_name,
2944};
2945
2946static const struct vm_operations_struct legacy_special_mapping_vmops = {
2947 .close = special_mapping_close,
2948 .fault = special_mapping_fault,
2949};
Roland McGrathfa5dc222007-02-08 14:20:41 -08002950
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002951static int special_mapping_fault(struct vm_area_struct *vma,
2952 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002953{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002954 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002955 struct page **pages;
2956
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002957 /*
2958 * special mappings have no vm_file, and in that case, the mm
2959 * uses vm_pgoff internally. So we have to subtract it from here.
2960 * We are allowed to do this because we are the mm; do not copy
2961 * this code into drivers!
2962 */
2963 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002964
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002965 if (vma->vm_ops == &legacy_special_mapping_vmops)
2966 pages = vma->vm_private_data;
2967 else
2968 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
2969 pages;
2970
2971 for (; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002972 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002973
2974 if (*pages) {
2975 struct page *page = *pages;
2976 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002977 vmf->page = page;
2978 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002979 }
2980
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002981 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002982}
2983
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002984static struct vm_area_struct *__install_special_mapping(
2985 struct mm_struct *mm,
2986 unsigned long addr, unsigned long len,
2987 unsigned long vm_flags, const struct vm_operations_struct *ops,
2988 void *priv)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002989{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002990 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002991 struct vm_area_struct *vma;
2992
2993 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2994 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01002995 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002996
Rik van Riel5beb4932010-03-05 13:42:07 -08002997 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002998 vma->vm_mm = mm;
2999 vma->vm_start = addr;
3000 vma->vm_end = addr + len;
3001
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003002 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003003 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003004
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003005 vma->vm_ops = ops;
3006 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003007
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003008 ret = insert_vm_struct(mm, vma);
3009 if (ret)
3010 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003011
3012 mm->total_vm += len >> PAGE_SHIFT;
3013
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003014 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003015
Stefani Seibold3935ed62014-03-17 23:22:02 +01003016 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003017
3018out:
3019 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003020 return ERR_PTR(ret);
3021}
3022
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003023/*
3024 * Called with mm->mmap_sem held for writing.
3025 * Insert a new vma covering the given region, with the given flags.
3026 * Its pages are supplied by the given array of struct page *.
3027 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3028 * The region past the last page supplied will always produce SIGBUS.
3029 * The array pointer and the pages it points to are assumed to stay alive
3030 * for as long as this mapping might exist.
3031 */
3032struct vm_area_struct *_install_special_mapping(
3033 struct mm_struct *mm,
3034 unsigned long addr, unsigned long len,
3035 unsigned long vm_flags, const struct vm_special_mapping *spec)
3036{
3037 return __install_special_mapping(mm, addr, len, vm_flags,
3038 &special_mapping_vmops, (void *)spec);
3039}
3040
Stefani Seibold3935ed62014-03-17 23:22:02 +01003041int install_special_mapping(struct mm_struct *mm,
3042 unsigned long addr, unsigned long len,
3043 unsigned long vm_flags, struct page **pages)
3044{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003045 struct vm_area_struct *vma = __install_special_mapping(
3046 mm, addr, len, vm_flags, &legacy_special_mapping_vmops,
3047 (void *)pages);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003048
Duan Jiong14bd5b42014-06-04 16:07:05 -07003049 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003050}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003051
3052static DEFINE_MUTEX(mm_all_locks_mutex);
3053
Peter Zijlstra454ed842008-08-11 09:30:25 +02003054static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003055{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003056 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003057 /*
3058 * The LSB of head.next can't change from under us
3059 * because we hold the mm_all_locks_mutex.
3060 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003061 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003062 /*
3063 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003064 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003065 * the same anon_vma we won't take it again.
3066 *
3067 * No need of atomic instructions here, head.next
3068 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003069 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003070 */
3071 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003072 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003073 BUG();
3074 }
3075}
3076
Peter Zijlstra454ed842008-08-11 09:30:25 +02003077static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003078{
3079 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3080 /*
3081 * AS_MM_ALL_LOCKS can't change from under us because
3082 * we hold the mm_all_locks_mutex.
3083 *
3084 * Operations on ->flags have to be atomic because
3085 * even if AS_MM_ALL_LOCKS is stable thanks to the
3086 * mm_all_locks_mutex, there may be other cpus
3087 * changing other bitflags in parallel to us.
3088 */
3089 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3090 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003091 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003092 }
3093}
3094
3095/*
3096 * This operation locks against the VM for all pte/vma/mm related
3097 * operations that could ever happen on a certain mm. This includes
3098 * vmtruncate, try_to_unmap, and all page faults.
3099 *
3100 * The caller must take the mmap_sem in write mode before calling
3101 * mm_take_all_locks(). The caller isn't allowed to release the
3102 * mmap_sem until mm_drop_all_locks() returns.
3103 *
3104 * mmap_sem in write mode is required in order to block all operations
3105 * that could modify pagetables and free pages without need of
3106 * altering the vma layout (for example populate_range() with
3107 * nonlinear vmas). It's also needed in write mode to avoid new
3108 * anon_vmas to be associated with existing vmas.
3109 *
3110 * A single task can't take more than one mm_take_all_locks() in a row
3111 * or it would deadlock.
3112 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003113 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003114 * mapping->flags avoid to take the same lock twice, if more than one
3115 * vma in this mm is backed by the same anon_vma or address_space.
3116 *
3117 * We can take all the locks in random order because the VM code
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003118 * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003119 * takes more than one of them in a row. Secondly we're protected
3120 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3121 *
3122 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3123 * that may have to take thousand of locks.
3124 *
3125 * mm_take_all_locks() can fail if it's interrupted by signals.
3126 */
3127int mm_take_all_locks(struct mm_struct *mm)
3128{
3129 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003130 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003131
3132 BUG_ON(down_read_trylock(&mm->mmap_sem));
3133
3134 mutex_lock(&mm_all_locks_mutex);
3135
3136 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3137 if (signal_pending(current))
3138 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003139 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003140 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003141 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003142
3143 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3144 if (signal_pending(current))
3145 goto out_unlock;
3146 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003147 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3148 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003149 }
3150
Kautuk Consul584cff52011-10-31 17:08:59 -07003151 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003152
3153out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003154 mm_drop_all_locks(mm);
3155 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003156}
3157
3158static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3159{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003160 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003161 /*
3162 * The LSB of head.next can't change to 0 from under
3163 * us because we hold the mm_all_locks_mutex.
3164 *
3165 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003166 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003167 * never see our bitflag.
3168 *
3169 * No need of atomic instructions here, head.next
3170 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003171 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003172 */
3173 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003174 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003175 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003176 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003177 }
3178}
3179
3180static void vm_unlock_mapping(struct address_space *mapping)
3181{
3182 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3183 /*
3184 * AS_MM_ALL_LOCKS can't change to 0 from under us
3185 * because we hold the mm_all_locks_mutex.
3186 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003187 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003188 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3189 &mapping->flags))
3190 BUG();
3191 }
3192}
3193
3194/*
3195 * The mmap_sem cannot be released by the caller until
3196 * mm_drop_all_locks() returns.
3197 */
3198void mm_drop_all_locks(struct mm_struct *mm)
3199{
3200 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003201 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003202
3203 BUG_ON(down_read_trylock(&mm->mmap_sem));
3204 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3205
3206 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3207 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003208 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3209 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003210 if (vma->vm_file && vma->vm_file->f_mapping)
3211 vm_unlock_mapping(vma->vm_file->f_mapping);
3212 }
3213
3214 mutex_unlock(&mm_all_locks_mutex);
3215}
David Howells8feae132009-01-08 12:04:47 +00003216
3217/*
3218 * initialise the VMA slab
3219 */
3220void __init mmap_init(void)
3221{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003222 int ret;
3223
Tejun Heo908c7f12014-09-08 09:51:29 +09003224 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003225 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003226}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003227
3228/*
3229 * Initialise sysctl_user_reserve_kbytes.
3230 *
3231 * This is intended to prevent a user from starting a single memory hogging
3232 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3233 * mode.
3234 *
3235 * The default value is min(3% of free memory, 128MB)
3236 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3237 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003238static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003239{
3240 unsigned long free_kbytes;
3241
3242 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3243
3244 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3245 return 0;
3246}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003247subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003248
3249/*
3250 * Initialise sysctl_admin_reserve_kbytes.
3251 *
3252 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3253 * to log in and kill a memory hogging process.
3254 *
3255 * Systems with more than 256MB will reserve 8MB, enough to recover
3256 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3257 * only reserve 3% of free pages by default.
3258 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003259static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003260{
3261 unsigned long free_kbytes;
3262
3263 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3264
3265 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3266 return 0;
3267}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003268subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003269
3270/*
3271 * Reinititalise user and admin reserves if memory is added or removed.
3272 *
3273 * The default user reserve max is 128MB, and the default max for the
3274 * admin reserve is 8MB. These are usually, but not always, enough to
3275 * enable recovery from a memory hogging process using login/sshd, a shell,
3276 * and tools like top. It may make sense to increase or even disable the
3277 * reserve depending on the existence of swap or variations in the recovery
3278 * tools. So, the admin may have changed them.
3279 *
3280 * If memory is added and the reserves have been eliminated or increased above
3281 * the default max, then we'll trust the admin.
3282 *
3283 * If memory is removed and there isn't enough free memory, then we
3284 * need to reset the reserves.
3285 *
3286 * Otherwise keep the reserve set by the admin.
3287 */
3288static int reserve_mem_notifier(struct notifier_block *nb,
3289 unsigned long action, void *data)
3290{
3291 unsigned long tmp, free_kbytes;
3292
3293 switch (action) {
3294 case MEM_ONLINE:
3295 /* Default max is 128MB. Leave alone if modified by operator. */
3296 tmp = sysctl_user_reserve_kbytes;
3297 if (0 < tmp && tmp < (1UL << 17))
3298 init_user_reserve();
3299
3300 /* Default max is 8MB. Leave alone if modified by operator. */
3301 tmp = sysctl_admin_reserve_kbytes;
3302 if (0 < tmp && tmp < (1UL << 13))
3303 init_admin_reserve();
3304
3305 break;
3306 case MEM_OFFLINE:
3307 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3308
3309 if (sysctl_user_reserve_kbytes > free_kbytes) {
3310 init_user_reserve();
3311 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3312 sysctl_user_reserve_kbytes);
3313 }
3314
3315 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3316 init_admin_reserve();
3317 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3318 sysctl_admin_reserve_kbytes);
3319 }
3320 break;
3321 default:
3322 break;
3323 }
3324 return NOTIFY_OK;
3325}
3326
3327static struct notifier_block reserve_mem_nb = {
3328 .notifier_call = reserve_mem_notifier,
3329};
3330
3331static int __meminit init_reserve_notifier(void)
3332{
3333 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003334 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003335
3336 return 0;
3337}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003338subsys_initcall(init_reserve_notifier);