blob: 87e82b38453c2cbca83f1dd7ad472c02b6a73b77 [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/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700235 * Requires inode->i_mapping->i_mmap_mutex
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;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700263 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700265 mutex_unlock(&mapping->i_mmap_mutex);
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;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700677 mutex_lock(&mapping->i_mmap_mutex);
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)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700684 mutex_unlock(&mapping->i_mmap_mutex);
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) {
779 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800780 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700781 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800782 }
783 }
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 if (file) {
786 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530787 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530789 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530790
791 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530792 uprobe_munmap(next, next->vm_start,
793 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530794 }
795
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700796 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700799 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * are visible to arm/parisc __flush_dcache_page
801 * throughout; but we cannot insert into address
802 * space until vma start or end is updated.
803 */
804 __vma_link_file(insert);
805 }
806 }
807
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800808 vma_adjust_trans_huge(vma, start, end, adjust_next);
809
Michel Lespinassebf181b92012-10-08 16:31:39 -0700810 anon_vma = vma->anon_vma;
811 if (!anon_vma && adjust_next)
812 anon_vma = next->anon_vma;
813 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700814 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
815 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000816 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700817 anon_vma_interval_tree_pre_update_vma(vma);
818 if (adjust_next)
819 anon_vma_interval_tree_pre_update_vma(next);
820 }
Rik van Riel012f18002010-08-09 17:18:40 -0700821
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 if (root) {
823 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700824 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700826 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
828
Michel Lespinassed3737182012-12-11 16:01:38 -0800829 if (start != vma->vm_start) {
830 vma->vm_start = start;
831 start_changed = true;
832 }
833 if (end != vma->vm_end) {
834 vma->vm_end = end;
835 end_changed = true;
836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 vma->vm_pgoff = pgoff;
838 if (adjust_next) {
839 next->vm_start += adjust_next << PAGE_SHIFT;
840 next->vm_pgoff += adjust_next;
841 }
842
843 if (root) {
844 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700845 vma_interval_tree_insert(next, root);
846 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 flush_dcache_mmap_unlock(mapping);
848 }
849
850 if (remove_next) {
851 /*
852 * vma_merge has merged next into vma, and needs
853 * us to remove next before dropping the locks.
854 */
855 __vma_unlink(mm, next, vma);
856 if (file)
857 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 } else if (insert) {
859 /*
860 * split_vma has split insert from vma, and needs
861 * us to insert it before dropping the locks
862 * (it may either follow vma or precede it).
863 */
864 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800865 } else {
866 if (start_changed)
867 vma_gap_update(vma);
868 if (end_changed) {
869 if (!next)
870 mm->highest_vm_end = end;
871 else if (!adjust_next)
872 vma_gap_update(next);
873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 }
875
Michel Lespinassebf181b92012-10-08 16:31:39 -0700876 if (anon_vma) {
877 anon_vma_interval_tree_post_update_vma(vma);
878 if (adjust_next)
879 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800880 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700883 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530885 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100886 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530887
888 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100889 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530890 }
891
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700893 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530894 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700896 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800897 if (next->anon_vma)
898 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700900 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 kmem_cache_free(vm_area_cachep, next);
902 /*
903 * In mprotect's case 6 (see comments on vma_merge),
904 * we must remove another next too. It would clutter
905 * up the code too much to do both in one go.
906 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800907 next = vma->vm_next;
908 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800910 else if (next)
911 vma_gap_update(next);
912 else
913 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530915 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100916 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
918 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800919
920 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921}
922
923/*
924 * If the vma has a ->close operation then the driver probably needs to release
925 * per-vma resources, so we don't attempt to merge those.
926 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927static inline int is_mergeable_vma(struct vm_area_struct *vma,
928 struct file *file, unsigned long vm_flags)
929{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800930 /*
931 * VM_SOFTDIRTY should not prevent from VMA merging, if we
932 * match the flags but dirty bit -- the caller should mark
933 * merged VMA as dirty. If dirty bit won't be excluded from
934 * comparison, we increase pressue on the memory system forcing
935 * the kernel to generate new VMAs when old one could be
936 * extended instead.
937 */
938 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 return 0;
940 if (vma->vm_file != file)
941 return 0;
942 if (vma->vm_ops && vma->vm_ops->close)
943 return 0;
944 return 1;
945}
946
947static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700948 struct anon_vma *anon_vma2,
949 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950{
Shaohua Li965f55d2011-05-24 17:11:20 -0700951 /*
952 * The list_is_singular() test is to avoid merging VMA cloned from
953 * parents. This can improve scalability caused by anon_vma lock.
954 */
955 if ((!anon_vma1 || !anon_vma2) && (!vma ||
956 list_is_singular(&vma->anon_vma_chain)))
957 return 1;
958 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959}
960
961/*
962 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
963 * in front of (at a lower virtual address and file offset than) the vma.
964 *
965 * We cannot merge two vmas if they have differently assigned (non-NULL)
966 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
967 *
968 * We don't check here for the merged mmap wrapping around the end of pagecache
969 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
970 * wrap, nor mmaps which cover the final page at index -1UL.
971 */
972static int
973can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
974 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
975{
976 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700977 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (vma->vm_pgoff == vm_pgoff)
979 return 1;
980 }
981 return 0;
982}
983
984/*
985 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
986 * beyond (at a higher virtual address and file offset than) the vma.
987 *
988 * We cannot merge two vmas if they have differently assigned (non-NULL)
989 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
990 */
991static int
992can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
993 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
994{
995 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700996 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700998 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1000 return 1;
1001 }
1002 return 0;
1003}
1004
1005/*
1006 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1007 * whether that can be merged with its predecessor or its successor.
1008 * Or both (it neatly fills a hole).
1009 *
1010 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1011 * certain not to be mapped by the time vma_merge is called; but when
1012 * called for mprotect, it is certain to be already mapped (either at
1013 * an offset within prev, or at the start of next), and the flags of
1014 * this area are about to be changed to vm_flags - and the no-change
1015 * case has already been eliminated.
1016 *
1017 * The following mprotect cases have to be considered, where AAAA is
1018 * the area passed down from mprotect_fixup, never extending beyond one
1019 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1020 *
1021 * AAAA AAAA AAAA AAAA
1022 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1023 * cannot merge might become might become might become
1024 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1025 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1026 * mremap move: PPPPNNNNNNNN 8
1027 * AAAA
1028 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1029 * might become case 1 below case 2 below case 3 below
1030 *
1031 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1032 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1033 */
1034struct vm_area_struct *vma_merge(struct mm_struct *mm,
1035 struct vm_area_struct *prev, unsigned long addr,
1036 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001037 struct anon_vma *anon_vma, struct file *file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 pgoff_t pgoff, struct mempolicy *policy)
1039{
1040 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1041 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001042 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 /*
1045 * We later require that vma->vm_flags == vm_flags,
1046 * so this tests vma->vm_flags & VM_SPECIAL, too.
1047 */
1048 if (vm_flags & VM_SPECIAL)
1049 return NULL;
1050
1051 if (prev)
1052 next = prev->vm_next;
1053 else
1054 next = mm->mmap;
1055 area = next;
1056 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1057 next = next->vm_next;
1058
1059 /*
1060 * Can it merge with the predecessor?
1061 */
1062 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001063 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 can_vma_merge_after(prev, vm_flags,
1065 anon_vma, file, pgoff)) {
1066 /*
1067 * OK, it can. Can we now merge in the successor as well?
1068 */
1069 if (next && end == next->vm_start &&
1070 mpol_equal(policy, vma_policy(next)) &&
1071 can_vma_merge_before(next, vm_flags,
1072 anon_vma, file, pgoff+pglen) &&
1073 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001074 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001076 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 next->vm_end, prev->vm_pgoff, NULL);
1078 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001079 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001081 if (err)
1082 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001083 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 return prev;
1085 }
1086
1087 /*
1088 * Can this new request be merged in front of next?
1089 */
1090 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001091 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 can_vma_merge_before(next, vm_flags,
1093 anon_vma, file, pgoff+pglen)) {
1094 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001095 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 addr, prev->vm_pgoff, NULL);
1097 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001098 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001100 if (err)
1101 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001102 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 return area;
1104 }
1105
1106 return NULL;
1107}
1108
1109/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001110 * Rough compatbility check to quickly see if it's even worth looking
1111 * at sharing an anon_vma.
1112 *
1113 * They need to have the same vm_file, and the flags can only differ
1114 * in things that mprotect may change.
1115 *
1116 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1117 * we can merge the two vma's. For example, we refuse to merge a vma if
1118 * there is a vm_ops->close() function, because that indicates that the
1119 * driver is doing some kind of reference counting. But that doesn't
1120 * really matter for the anon_vma sharing case.
1121 */
1122static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1123{
1124 return a->vm_end == b->vm_start &&
1125 mpol_equal(vma_policy(a), vma_policy(b)) &&
1126 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001127 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001128 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1129}
1130
1131/*
1132 * Do some basic sanity checking to see if we can re-use the anon_vma
1133 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1134 * the same as 'old', the other will be the new one that is trying
1135 * to share the anon_vma.
1136 *
1137 * NOTE! This runs with mm_sem held for reading, so it is possible that
1138 * the anon_vma of 'old' is concurrently in the process of being set up
1139 * by another page fault trying to merge _that_. But that's ok: if it
1140 * is being set up, that automatically means that it will be a singleton
1141 * acceptable for merging, so we can do all of this optimistically. But
1142 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1143 *
1144 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1145 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1146 * is to return an anon_vma that is "complex" due to having gone through
1147 * a fork).
1148 *
1149 * We also make sure that the two vma's are compatible (adjacent,
1150 * and with the same memory policies). That's all stable, even with just
1151 * a read lock on the mm_sem.
1152 */
1153static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1154{
1155 if (anon_vma_compatible(a, b)) {
1156 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1157
1158 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1159 return anon_vma;
1160 }
1161 return NULL;
1162}
1163
1164/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1166 * neighbouring vmas for a suitable anon_vma, before it goes off
1167 * to allocate a new anon_vma. It checks because a repetitive
1168 * sequence of mprotects and faults may otherwise lead to distinct
1169 * anon_vmas being allocated, preventing vma merge in subsequent
1170 * mprotect.
1171 */
1172struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1173{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001174 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
1177 near = vma->vm_next;
1178 if (!near)
1179 goto try_prev;
1180
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001181 anon_vma = reusable_anon_vma(near, vma, near);
1182 if (anon_vma)
1183 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001185 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 if (!near)
1187 goto none;
1188
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001189 anon_vma = reusable_anon_vma(near, near, vma);
1190 if (anon_vma)
1191 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192none:
1193 /*
1194 * There's no absolute need to look only at touching neighbours:
1195 * we could search further afield for "compatible" anon_vmas.
1196 * But it would probably just be a waste of time searching,
1197 * or lead to too many vmas hanging off the same anon_vma.
1198 * We're trying to allow mprotect remerging later on,
1199 * not trying to minimize memory used for anon_vmas.
1200 */
1201 return NULL;
1202}
1203
1204#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001205void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 struct file *file, long pages)
1207{
1208 const unsigned long stack_flags
1209 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1210
Huang Shijie44de9d02012-07-31 16:41:49 -07001211 mm->total_vm += pages;
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 if (file) {
1214 mm->shared_vm += pages;
1215 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1216 mm->exec_vm += pages;
1217 } else if (flags & stack_flags)
1218 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220#endif /* CONFIG_PROC_FS */
1221
1222/*
Al Viro40401532012-02-13 03:58:52 +00001223 * If a hint addr is less than mmap_min_addr change hint to be as
1224 * low as possible but still greater than mmap_min_addr
1225 */
1226static inline unsigned long round_hint_to_min(unsigned long hint)
1227{
1228 hint &= PAGE_MASK;
1229 if (((void *)hint != NULL) &&
1230 (hint < mmap_min_addr))
1231 return PAGE_ALIGN(mmap_min_addr);
1232 return hint;
1233}
1234
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001235static inline int mlock_future_check(struct mm_struct *mm,
1236 unsigned long flags,
1237 unsigned long len)
1238{
1239 unsigned long locked, lock_limit;
1240
1241 /* mlock MCL_FUTURE? */
1242 if (flags & VM_LOCKED) {
1243 locked = len >> PAGE_SHIFT;
1244 locked += mm->locked_vm;
1245 lock_limit = rlimit(RLIMIT_MEMLOCK);
1246 lock_limit >>= PAGE_SHIFT;
1247 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1248 return -EAGAIN;
1249 }
1250 return 0;
1251}
1252
Al Viro40401532012-02-13 03:58:52 +00001253/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001254 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 */
1256
Al Viroe3fc6292012-05-30 20:08:42 -04001257unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001259 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001260 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
vishnu.pscc71aba2014-10-09 15:26:29 -07001262 struct mm_struct *mm = current->mm;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001263 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Michel Lespinasse41badc12013-02-22 16:32:47 -08001265 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 /*
1268 * Does the application expect PROT_READ to imply PROT_EXEC?
1269 *
1270 * (the exception is when the underlying filesystem is noexec
1271 * mounted, in which case we dont add PROT_EXEC.)
1272 */
1273 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001274 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 prot |= PROT_EXEC;
1276
1277 if (!len)
1278 return -EINVAL;
1279
Eric Paris7cd94142007-11-26 18:47:40 -05001280 if (!(flags & MAP_FIXED))
1281 addr = round_hint_to_min(addr);
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 /* Careful about overflows.. */
1284 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001285 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return -ENOMEM;
1287
1288 /* offset overflow? */
1289 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001290 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
1292 /* Too many mappings? */
1293 if (mm->map_count > sysctl_max_map_count)
1294 return -ENOMEM;
1295
1296 /* Obtain the address to map to. we verify (or select) it and ensure
1297 * that it represents a valid section of the address space.
1298 */
1299 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1300 if (addr & ~PAGE_MASK)
1301 return addr;
1302
1303 /* Do simple checking here so the lower-level routines won't have
1304 * to. we assume access permissions have been handled by the open
1305 * of the memory object, so we don't do any here.
1306 */
1307 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1308 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1309
Huang Shijiecdf7b342009-09-21 17:03:36 -07001310 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 if (!can_do_mlock())
1312 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001313
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001314 if (mlock_future_check(mm, vm_flags, len))
1315 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001318 struct inode *inode = file_inode(file);
1319
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 switch (flags & MAP_TYPE) {
1321 case MAP_SHARED:
1322 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1323 return -EACCES;
1324
1325 /*
1326 * Make sure we don't allow writing to an append-only
1327 * file..
1328 */
1329 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1330 return -EACCES;
1331
1332 /*
1333 * Make sure there are no mandatory locks on the file.
1334 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001335 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 return -EAGAIN;
1337
1338 vm_flags |= VM_SHARED | VM_MAYSHARE;
1339 if (!(file->f_mode & FMODE_WRITE))
1340 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1341
1342 /* fall through */
1343 case MAP_PRIVATE:
1344 if (!(file->f_mode & FMODE_READ))
1345 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001346 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001347 if (vm_flags & VM_EXEC)
1348 return -EPERM;
1349 vm_flags &= ~VM_MAYEXEC;
1350 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001351
Al Viro72c2d532013-09-22 16:27:52 -04001352 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001353 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001354 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1355 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 break;
1357
1358 default:
1359 return -EINVAL;
1360 }
1361 } else {
1362 switch (flags & MAP_TYPE) {
1363 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001364 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1365 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001366 /*
1367 * Ignore pgoff.
1368 */
1369 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 vm_flags |= VM_SHARED | VM_MAYSHARE;
1371 break;
1372 case MAP_PRIVATE:
1373 /*
1374 * Set pgoff according to addr for anon_vma.
1375 */
1376 pgoff = addr >> PAGE_SHIFT;
1377 break;
1378 default:
1379 return -EINVAL;
1380 }
1381 }
1382
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001383 /*
1384 * Set 'VM_NORESERVE' if we should not account for the
1385 * memory use of this mapping.
1386 */
1387 if (flags & MAP_NORESERVE) {
1388 /* We honor MAP_NORESERVE if allowed to overcommit */
1389 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1390 vm_flags |= VM_NORESERVE;
1391
1392 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1393 if (file && is_file_hugepages(file))
1394 vm_flags |= VM_NORESERVE;
1395 }
1396
1397 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001398 if (!IS_ERR_VALUE(addr) &&
1399 ((vm_flags & VM_LOCKED) ||
1400 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001401 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001402 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001403}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001404
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001405SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1406 unsigned long, prot, unsigned long, flags,
1407 unsigned long, fd, unsigned long, pgoff)
1408{
1409 struct file *file = NULL;
1410 unsigned long retval = -EBADF;
1411
1412 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001413 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001414 file = fget(fd);
1415 if (!file)
1416 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001417 if (is_file_hugepages(file))
1418 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001419 retval = -EINVAL;
1420 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1421 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001422 } else if (flags & MAP_HUGETLB) {
1423 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001424 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001425
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001426 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001427 if (!hs)
1428 return -EINVAL;
1429
1430 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001431 /*
1432 * VM_NORESERVE is used because the reservations will be
1433 * taken when vm_ops->mmap() is called
1434 * A dummy user value is used because we are not locking
1435 * memory so no accounting is necessary
1436 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001437 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001438 VM_NORESERVE,
1439 &user, HUGETLB_ANONHUGE_INODE,
1440 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001441 if (IS_ERR(file))
1442 return PTR_ERR(file);
1443 }
1444
1445 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1446
Al Viroeb36c582012-05-30 20:17:35 -04001447 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001448out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001449 if (file)
1450 fput(file);
1451out:
1452 return retval;
1453}
1454
Christoph Hellwiga4679372010-03-10 15:21:15 -08001455#ifdef __ARCH_WANT_SYS_OLD_MMAP
1456struct mmap_arg_struct {
1457 unsigned long addr;
1458 unsigned long len;
1459 unsigned long prot;
1460 unsigned long flags;
1461 unsigned long fd;
1462 unsigned long offset;
1463};
1464
1465SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1466{
1467 struct mmap_arg_struct a;
1468
1469 if (copy_from_user(&a, arg, sizeof(a)))
1470 return -EFAULT;
1471 if (a.offset & ~PAGE_MASK)
1472 return -EINVAL;
1473
1474 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1475 a.offset >> PAGE_SHIFT);
1476}
1477#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1478
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001479/*
1480 * Some shared mappigns will want the pages marked read-only
1481 * to track write events. If so, we'll downgrade vm_page_prot
1482 * to the private version (using protection_map[] without the
1483 * VM_SHARED bit).
1484 */
1485int vma_wants_writenotify(struct vm_area_struct *vma)
1486{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001487 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001488
1489 /* If it was private or non-writable, the write bit is already clear */
1490 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1491 return 0;
1492
1493 /* The backer wishes to know when pages are first written to? */
1494 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1495 return 1;
1496
Peter Feiner64e45502014-10-13 15:55:46 -07001497 /* The open routine did something to the protections that pgprot_modify
1498 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001499 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001500 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001501 return 0;
1502
Peter Feiner64e45502014-10-13 15:55:46 -07001503 /* Do we need to track softdirty? */
1504 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1505 return 1;
1506
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001507 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001508 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001509 return 0;
1510
1511 /* Can the mapping track the dirty pages? */
1512 return vma->vm_file && vma->vm_file->f_mapping &&
1513 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1514}
1515
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001516/*
1517 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001518 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001519 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001520static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001521{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001522 /*
1523 * hugetlb has its own accounting separate from the core VM
1524 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1525 */
1526 if (file && is_file_hugepages(file))
1527 return 0;
1528
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001529 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1530}
1531
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001532unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001533 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001534{
1535 struct mm_struct *mm = current->mm;
1536 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001537 int error;
1538 struct rb_node **rb_link, *rb_parent;
1539 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001540
Cyril Hrubise8420a82013-04-29 15:08:33 -07001541 /* Check against address space limit. */
1542 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1543 unsigned long nr_pages;
1544
1545 /*
1546 * MAP_FIXED may remove pages of mappings that intersects with
1547 * requested mapping. Account for the pages it would unmap.
1548 */
1549 if (!(vm_flags & MAP_FIXED))
1550 return -ENOMEM;
1551
1552 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1553
1554 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1555 return -ENOMEM;
1556 }
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 /* Clear old maps */
1559 error = -ENOMEM;
1560munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001561 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if (do_munmap(mm, addr, len))
1563 return -ENOMEM;
1564 goto munmap_back;
1565 }
1566
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001567 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001568 * Private writable mapping: check memory availability
1569 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001570 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001571 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001572 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001573 return -ENOMEM;
1574 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 }
1576
1577 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001578 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001580 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1581 if (vma)
1582 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
1584 /*
1585 * Determine the object being mapped and call the appropriate
1586 * specific mapper. the address has already been validated, but
1587 * not unmapped, but the maps are removed from the list.
1588 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001589 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 if (!vma) {
1591 error = -ENOMEM;
1592 goto unacct_error;
1593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595 vma->vm_mm = mm;
1596 vma->vm_start = addr;
1597 vma->vm_end = addr + len;
1598 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001599 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001601 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602
1603 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 if (vm_flags & VM_DENYWRITE) {
1605 error = deny_write_access(file);
1606 if (error)
1607 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001609 if (vm_flags & VM_SHARED) {
1610 error = mapping_map_writable(file->f_mapping);
1611 if (error)
1612 goto allow_write_and_free_vma;
1613 }
1614
1615 /* ->mmap() can change vma->vm_file, but must guarantee that
1616 * vma_link() below can deny write-access if VM_DENYWRITE is set
1617 * and map writably if VM_SHARED is set. This usually means the
1618 * new file must not have been exposed to user-space, yet.
1619 */
Al Virocb0942b2012-08-27 14:48:26 -04001620 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 error = file->f_op->mmap(file, vma);
1622 if (error)
1623 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001624
1625 /* Can addr have changed??
1626 *
1627 * Answer: Yes, several device drivers can do it in their
1628 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001629 * Bug: If addr is changed, prev, rb_link, rb_parent should
1630 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001631 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001632 WARN_ON_ONCE(addr != vma->vm_start);
1633
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001634 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001635 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 } else if (vm_flags & VM_SHARED) {
1637 error = shmem_zero_setup(vma);
1638 if (error)
1639 goto free_vma;
1640 }
1641
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001642 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001643 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001644 if (file) {
1645 if (vm_flags & VM_SHARED)
1646 mapping_unmap_writable(file->f_mapping);
1647 if (vm_flags & VM_DENYWRITE)
1648 allow_write_access(file);
1649 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001650 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001651out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001652 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001653
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001654 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001656 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1657 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001658 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001659 else
1660 vma->vm_flags &= ~VM_LOCKED;
1661 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301662
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001663 if (file)
1664 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301665
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001666 /*
1667 * New (or expanded) vma always get soft dirty status.
1668 * Otherwise user-space soft-dirty page tracker won't
1669 * be able to distinguish situation when vma area unmapped,
1670 * then new mapped in-place (which must be aimed as
1671 * a completely new data area).
1672 */
1673 vma->vm_flags |= VM_SOFTDIRTY;
1674
Peter Feiner64e45502014-10-13 15:55:46 -07001675 vma_set_page_prot(vma);
1676
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 return addr;
1678
1679unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 vma->vm_file = NULL;
1681 fput(file);
1682
1683 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001684 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1685 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001686 if (vm_flags & VM_SHARED)
1687 mapping_unmap_writable(file->f_mapping);
1688allow_write_and_free_vma:
1689 if (vm_flags & VM_DENYWRITE)
1690 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691free_vma:
1692 kmem_cache_free(vm_area_cachep, vma);
1693unacct_error:
1694 if (charged)
1695 vm_unacct_memory(charged);
1696 return error;
1697}
1698
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001699unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1700{
1701 /*
1702 * We implement the search by looking for an rbtree node that
1703 * immediately follows a suitable gap. That is,
1704 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1705 * - gap_end = vma->vm_start >= info->low_limit + length;
1706 * - gap_end - gap_start >= length
1707 */
1708
1709 struct mm_struct *mm = current->mm;
1710 struct vm_area_struct *vma;
1711 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1712
1713 /* Adjust search length to account for worst case alignment overhead */
1714 length = info->length + info->align_mask;
1715 if (length < info->length)
1716 return -ENOMEM;
1717
1718 /* Adjust search limits by the desired length */
1719 if (info->high_limit < length)
1720 return -ENOMEM;
1721 high_limit = info->high_limit - length;
1722
1723 if (info->low_limit > high_limit)
1724 return -ENOMEM;
1725 low_limit = info->low_limit + length;
1726
1727 /* Check if rbtree root looks promising */
1728 if (RB_EMPTY_ROOT(&mm->mm_rb))
1729 goto check_highest;
1730 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1731 if (vma->rb_subtree_gap < length)
1732 goto check_highest;
1733
1734 while (true) {
1735 /* Visit left subtree if it looks promising */
1736 gap_end = vma->vm_start;
1737 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1738 struct vm_area_struct *left =
1739 rb_entry(vma->vm_rb.rb_left,
1740 struct vm_area_struct, vm_rb);
1741 if (left->rb_subtree_gap >= length) {
1742 vma = left;
1743 continue;
1744 }
1745 }
1746
1747 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1748check_current:
1749 /* Check if current node has a suitable gap */
1750 if (gap_start > high_limit)
1751 return -ENOMEM;
1752 if (gap_end >= low_limit && gap_end - gap_start >= length)
1753 goto found;
1754
1755 /* Visit right subtree if it looks promising */
1756 if (vma->vm_rb.rb_right) {
1757 struct vm_area_struct *right =
1758 rb_entry(vma->vm_rb.rb_right,
1759 struct vm_area_struct, vm_rb);
1760 if (right->rb_subtree_gap >= length) {
1761 vma = right;
1762 continue;
1763 }
1764 }
1765
1766 /* Go back up the rbtree to find next candidate node */
1767 while (true) {
1768 struct rb_node *prev = &vma->vm_rb;
1769 if (!rb_parent(prev))
1770 goto check_highest;
1771 vma = rb_entry(rb_parent(prev),
1772 struct vm_area_struct, vm_rb);
1773 if (prev == vma->vm_rb.rb_left) {
1774 gap_start = vma->vm_prev->vm_end;
1775 gap_end = vma->vm_start;
1776 goto check_current;
1777 }
1778 }
1779 }
1780
1781check_highest:
1782 /* Check highest gap, which does not precede any rbtree node */
1783 gap_start = mm->highest_vm_end;
1784 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1785 if (gap_start > high_limit)
1786 return -ENOMEM;
1787
1788found:
1789 /* We found a suitable gap. Clip it with the original low_limit. */
1790 if (gap_start < info->low_limit)
1791 gap_start = info->low_limit;
1792
1793 /* Adjust gap address to the desired alignment */
1794 gap_start += (info->align_offset - gap_start) & info->align_mask;
1795
1796 VM_BUG_ON(gap_start + info->length > info->high_limit);
1797 VM_BUG_ON(gap_start + info->length > gap_end);
1798 return gap_start;
1799}
1800
1801unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1802{
1803 struct mm_struct *mm = current->mm;
1804 struct vm_area_struct *vma;
1805 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1806
1807 /* Adjust search length to account for worst case alignment overhead */
1808 length = info->length + info->align_mask;
1809 if (length < info->length)
1810 return -ENOMEM;
1811
1812 /*
1813 * Adjust search limits by the desired length.
1814 * See implementation comment at top of unmapped_area().
1815 */
1816 gap_end = info->high_limit;
1817 if (gap_end < length)
1818 return -ENOMEM;
1819 high_limit = gap_end - length;
1820
1821 if (info->low_limit > high_limit)
1822 return -ENOMEM;
1823 low_limit = info->low_limit + length;
1824
1825 /* Check highest gap, which does not precede any rbtree node */
1826 gap_start = mm->highest_vm_end;
1827 if (gap_start <= high_limit)
1828 goto found_highest;
1829
1830 /* Check if rbtree root looks promising */
1831 if (RB_EMPTY_ROOT(&mm->mm_rb))
1832 return -ENOMEM;
1833 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1834 if (vma->rb_subtree_gap < length)
1835 return -ENOMEM;
1836
1837 while (true) {
1838 /* Visit right subtree if it looks promising */
1839 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1840 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1841 struct vm_area_struct *right =
1842 rb_entry(vma->vm_rb.rb_right,
1843 struct vm_area_struct, vm_rb);
1844 if (right->rb_subtree_gap >= length) {
1845 vma = right;
1846 continue;
1847 }
1848 }
1849
1850check_current:
1851 /* Check if current node has a suitable gap */
1852 gap_end = vma->vm_start;
1853 if (gap_end < low_limit)
1854 return -ENOMEM;
1855 if (gap_start <= high_limit && gap_end - gap_start >= length)
1856 goto found;
1857
1858 /* Visit left subtree if it looks promising */
1859 if (vma->vm_rb.rb_left) {
1860 struct vm_area_struct *left =
1861 rb_entry(vma->vm_rb.rb_left,
1862 struct vm_area_struct, vm_rb);
1863 if (left->rb_subtree_gap >= length) {
1864 vma = left;
1865 continue;
1866 }
1867 }
1868
1869 /* Go back up the rbtree to find next candidate node */
1870 while (true) {
1871 struct rb_node *prev = &vma->vm_rb;
1872 if (!rb_parent(prev))
1873 return -ENOMEM;
1874 vma = rb_entry(rb_parent(prev),
1875 struct vm_area_struct, vm_rb);
1876 if (prev == vma->vm_rb.rb_right) {
1877 gap_start = vma->vm_prev ?
1878 vma->vm_prev->vm_end : 0;
1879 goto check_current;
1880 }
1881 }
1882 }
1883
1884found:
1885 /* We found a suitable gap. Clip it with the original high_limit. */
1886 if (gap_end > info->high_limit)
1887 gap_end = info->high_limit;
1888
1889found_highest:
1890 /* Compute highest gap address at the desired alignment */
1891 gap_end -= info->length;
1892 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1893
1894 VM_BUG_ON(gap_end < info->low_limit);
1895 VM_BUG_ON(gap_end < gap_start);
1896 return gap_end;
1897}
1898
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899/* Get an address range which is currently unmapped.
1900 * For shmat() with addr=0.
1901 *
1902 * Ugly calling convention alert:
1903 * Return value with the low bits set means error value,
1904 * ie
1905 * if (ret & ~PAGE_MASK)
1906 * error = ret;
1907 *
1908 * This function "knows" that -ENOMEM has the bits set.
1909 */
1910#ifndef HAVE_ARCH_UNMAPPED_AREA
1911unsigned long
1912arch_get_unmapped_area(struct file *filp, unsigned long addr,
1913 unsigned long len, unsigned long pgoff, unsigned long flags)
1914{
1915 struct mm_struct *mm = current->mm;
1916 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001917 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001919 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 return -ENOMEM;
1921
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001922 if (flags & MAP_FIXED)
1923 return addr;
1924
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 if (addr) {
1926 addr = PAGE_ALIGN(addr);
1927 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001928 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 (!vma || addr + len <= vma->vm_start))
1930 return addr;
1931 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001933 info.flags = 0;
1934 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001935 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001936 info.high_limit = TASK_SIZE;
1937 info.align_mask = 0;
1938 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939}
vishnu.pscc71aba2014-10-09 15:26:29 -07001940#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942/*
1943 * This mmap-allocator allocates new areas top-down from below the
1944 * stack's low limit (the base):
1945 */
1946#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1947unsigned long
1948arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1949 const unsigned long len, const unsigned long pgoff,
1950 const unsigned long flags)
1951{
1952 struct vm_area_struct *vma;
1953 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001954 unsigned long addr = addr0;
1955 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
1957 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001958 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 return -ENOMEM;
1960
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001961 if (flags & MAP_FIXED)
1962 return addr;
1963
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 /* requesting a specific address */
1965 if (addr) {
1966 addr = PAGE_ALIGN(addr);
1967 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001968 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 (!vma || addr + len <= vma->vm_start))
1970 return addr;
1971 }
1972
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001973 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1974 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001975 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001976 info.high_limit = mm->mmap_base;
1977 info.align_mask = 0;
1978 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 /*
1981 * A failed mmap() very likely causes application failure,
1982 * so fall back to the bottom-up function here. This scenario
1983 * can happen with large stack limits and large mmap()
1984 * allocations.
1985 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001986 if (addr & ~PAGE_MASK) {
1987 VM_BUG_ON(addr != -ENOMEM);
1988 info.flags = 0;
1989 info.low_limit = TASK_UNMAPPED_BASE;
1990 info.high_limit = TASK_SIZE;
1991 addr = vm_unmapped_area(&info);
1992 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
1994 return addr;
1995}
1996#endif
1997
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998unsigned long
1999get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2000 unsigned long pgoff, unsigned long flags)
2001{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002002 unsigned long (*get_area)(struct file *, unsigned long,
2003 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
Al Viro9206de92009-12-03 15:23:11 -05002005 unsigned long error = arch_mmap_check(addr, len, flags);
2006 if (error)
2007 return error;
2008
2009 /* Careful about overflows.. */
2010 if (len > TASK_SIZE)
2011 return -ENOMEM;
2012
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002013 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04002014 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002015 get_area = file->f_op->get_unmapped_area;
2016 addr = get_area(file, addr, len, pgoff, flags);
2017 if (IS_ERR_VALUE(addr))
2018 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002020 if (addr > TASK_SIZE - len)
2021 return -ENOMEM;
2022 if (addr & ~PAGE_MASK)
2023 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002024
Al Viro9ac4ed42012-05-30 17:13:15 -04002025 addr = arch_rebalance_pgtables(addr, len);
2026 error = security_mmap_addr(addr);
2027 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028}
2029
2030EXPORT_SYMBOL(get_unmapped_area);
2031
2032/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002033struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002035 struct rb_node *rb_node;
2036 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002038 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002039 vma = vmacache_find(mm, addr);
2040 if (likely(vma))
2041 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002043 rb_node = mm->mm_rb.rb_node;
2044 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002046 while (rb_node) {
2047 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002049 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002050
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002051 if (tmp->vm_end > addr) {
2052 vma = tmp;
2053 if (tmp->vm_start <= addr)
2054 break;
2055 rb_node = rb_node->rb_left;
2056 } else
2057 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002059
2060 if (vma)
2061 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 return vma;
2063}
2064
2065EXPORT_SYMBOL(find_vma);
2066
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002067/*
2068 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002069 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070struct vm_area_struct *
2071find_vma_prev(struct mm_struct *mm, unsigned long addr,
2072 struct vm_area_struct **pprev)
2073{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002074 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002076 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002077 if (vma) {
2078 *pprev = vma->vm_prev;
2079 } else {
2080 struct rb_node *rb_node = mm->mm_rb.rb_node;
2081 *pprev = NULL;
2082 while (rb_node) {
2083 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2084 rb_node = rb_node->rb_right;
2085 }
2086 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002087 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088}
2089
2090/*
2091 * Verify that the stack growth is acceptable and
2092 * update accounting. This is shared with both the
2093 * grow-up and grow-down cases.
2094 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002095static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096{
2097 struct mm_struct *mm = vma->vm_mm;
2098 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08002099 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
2101 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002102 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 return -ENOMEM;
2104
2105 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08002106 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 return -ENOMEM;
2108
2109 /* mlock limit tests */
2110 if (vma->vm_flags & VM_LOCKED) {
2111 unsigned long locked;
2112 unsigned long limit;
2113 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002114 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2115 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 if (locked > limit && !capable(CAP_IPC_LOCK))
2117 return -ENOMEM;
2118 }
2119
Adam Litke0d59a012007-01-30 14:35:39 -08002120 /* Check to ensure the stack will not grow into a hugetlb-only region */
2121 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2122 vma->vm_end - size;
2123 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2124 return -EFAULT;
2125
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 /*
2127 * Overcommit.. This must be the final test, as it will
2128 * update security statistics.
2129 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002130 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 return -ENOMEM;
2132
2133 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 if (vma->vm_flags & VM_LOCKED)
2135 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002136 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 return 0;
2138}
2139
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002140#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002142 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2143 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002145int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
2147 int error;
2148
2149 if (!(vma->vm_flags & VM_GROWSUP))
2150 return -EFAULT;
2151
2152 /*
2153 * We must make sure the anon_vma is allocated
2154 * so that the anon_vma locking is not a noop.
2155 */
2156 if (unlikely(anon_vma_prepare(vma)))
2157 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002158 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
2160 /*
2161 * vma->vm_start/vm_end cannot change under us because the caller
2162 * is required to hold the mmap_sem in read mode. We need the
2163 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002164 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 */
Helge Deller06b32f32006-12-19 19:28:33 +01002166 if (address < PAGE_ALIGN(address+4))
2167 address = PAGE_ALIGN(address+4);
2168 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002169 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002170 return -ENOMEM;
2171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 error = 0;
2173
2174 /* Somebody else might have raced and expanded it already */
2175 if (address > vma->vm_end) {
2176 unsigned long size, grow;
2177
2178 size = address - vma->vm_start;
2179 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2180
Hugh Dickins42c36f62011-05-09 17:44:42 -07002181 error = -ENOMEM;
2182 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2183 error = acct_stack_growth(vma, size, grow);
2184 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002185 /*
2186 * vma_gap_update() doesn't support concurrent
2187 * updates, but we only hold a shared mmap_sem
2188 * lock here, so we need to protect against
2189 * concurrent vma expansions.
2190 * vma_lock_anon_vma() doesn't help here, as
2191 * we don't guarantee that all growable vmas
2192 * in a mm share the same root anon vma.
2193 * So, we reuse mm->page_table_lock to guard
2194 * against concurrent vma expansions.
2195 */
2196 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002197 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002198 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002199 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002200 if (vma->vm_next)
2201 vma_gap_update(vma->vm_next);
2202 else
2203 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002204 spin_unlock(&vma->vm_mm->page_table_lock);
2205
Hugh Dickins42c36f62011-05-09 17:44:42 -07002206 perf_event_mmap(vma);
2207 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002210 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002211 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002212 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 return error;
2214}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002215#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217/*
2218 * vma is the first one with address < vma->vm_start. Have to extend vma.
2219 */
Michal Hockod05f3162011-05-24 17:11:44 -07002220int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002221 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222{
2223 int error;
2224
2225 /*
2226 * We must make sure the anon_vma is allocated
2227 * so that the anon_vma locking is not a noop.
2228 */
2229 if (unlikely(anon_vma_prepare(vma)))
2230 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002231
2232 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002233 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002234 if (error)
2235 return error;
2236
Rik van Rielbb4a3402010-08-09 17:18:37 -07002237 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238
2239 /*
2240 * vma->vm_start/vm_end cannot change under us because the caller
2241 * is required to hold the mmap_sem in read mode. We need the
2242 * anon_vma lock to serialize against concurrent expand_stacks.
2243 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245 /* Somebody else might have raced and expanded it already */
2246 if (address < vma->vm_start) {
2247 unsigned long size, grow;
2248
2249 size = vma->vm_end - address;
2250 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2251
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002252 error = -ENOMEM;
2253 if (grow <= vma->vm_pgoff) {
2254 error = acct_stack_growth(vma, size, grow);
2255 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002256 /*
2257 * vma_gap_update() doesn't support concurrent
2258 * updates, but we only hold a shared mmap_sem
2259 * lock here, so we need to protect against
2260 * concurrent vma expansions.
2261 * vma_lock_anon_vma() doesn't help here, as
2262 * we don't guarantee that all growable vmas
2263 * in a mm share the same root anon vma.
2264 * So, we reuse mm->page_table_lock to guard
2265 * against concurrent vma expansions.
2266 */
2267 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002268 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002269 vma->vm_start = address;
2270 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002271 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002272 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002273 spin_unlock(&vma->vm_mm->page_table_lock);
2274
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002275 perf_event_mmap(vma);
2276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 }
2278 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002279 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002280 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002281 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 return error;
2283}
2284
Linus Torvalds09884962013-02-27 08:36:04 -08002285/*
2286 * Note how expand_stack() refuses to expand the stack all the way to
2287 * abut the next virtual mapping, *unless* that mapping itself is also
2288 * a stack mapping. We want to leave room for a guard page, after all
2289 * (the guard page itself is not added here, that is done by the
2290 * actual page faulting logic)
2291 *
2292 * This matches the behavior of the guard page logic (see mm/memory.c:
2293 * check_stack_guard_page()), which only allows the guard page to be
2294 * removed under these circumstances.
2295 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002296#ifdef CONFIG_STACK_GROWSUP
2297int expand_stack(struct vm_area_struct *vma, unsigned long address)
2298{
Linus Torvalds09884962013-02-27 08:36:04 -08002299 struct vm_area_struct *next;
2300
2301 address &= PAGE_MASK;
2302 next = vma->vm_next;
2303 if (next && next->vm_start == address + PAGE_SIZE) {
2304 if (!(next->vm_flags & VM_GROWSUP))
2305 return -ENOMEM;
2306 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002307 return expand_upwards(vma, address);
2308}
2309
2310struct vm_area_struct *
2311find_extend_vma(struct mm_struct *mm, unsigned long addr)
2312{
2313 struct vm_area_struct *vma, *prev;
2314
2315 addr &= PAGE_MASK;
2316 vma = find_vma_prev(mm, addr, &prev);
2317 if (vma && (vma->vm_start <= addr))
2318 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002319 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002320 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002321 if (prev->vm_flags & VM_LOCKED)
2322 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002323 return prev;
2324}
2325#else
2326int expand_stack(struct vm_area_struct *vma, unsigned long address)
2327{
Linus Torvalds09884962013-02-27 08:36:04 -08002328 struct vm_area_struct *prev;
2329
2330 address &= PAGE_MASK;
2331 prev = vma->vm_prev;
2332 if (prev && prev->vm_end == address) {
2333 if (!(prev->vm_flags & VM_GROWSDOWN))
2334 return -ENOMEM;
2335 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002336 return expand_downwards(vma, address);
2337}
2338
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002340find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341{
vishnu.pscc71aba2014-10-09 15:26:29 -07002342 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 unsigned long start;
2344
2345 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002346 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 if (!vma)
2348 return NULL;
2349 if (vma->vm_start <= addr)
2350 return vma;
2351 if (!(vma->vm_flags & VM_GROWSDOWN))
2352 return NULL;
2353 start = vma->vm_start;
2354 if (expand_stack(vma, addr))
2355 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002356 if (vma->vm_flags & VM_LOCKED)
2357 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 return vma;
2359}
2360#endif
2361
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002363 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002365 *
2366 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002368static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002370 unsigned long nr_accounted = 0;
2371
Hugh Dickins365e9c872005-10-29 18:16:18 -07002372 /* Update high watermark before we lower total_vm */
2373 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002375 long nrpages = vma_pages(vma);
2376
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002377 if (vma->vm_flags & VM_ACCOUNT)
2378 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002379 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002380 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002381 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002382 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 validate_mm(mm);
2384}
2385
2386/*
2387 * Get rid of page table information in the indicated region.
2388 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002389 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 */
2391static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002392 struct vm_area_struct *vma, struct vm_area_struct *prev,
2393 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394{
vishnu.pscc71aba2014-10-09 15:26:29 -07002395 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002396 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
2398 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002399 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002400 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002401 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002402 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002403 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002404 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405}
2406
2407/*
2408 * Create a list of vma's touched by the unmap, removing them from the mm's
2409 * vma list as we go..
2410 */
2411static void
2412detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2413 struct vm_area_struct *prev, unsigned long end)
2414{
2415 struct vm_area_struct **insertion_point;
2416 struct vm_area_struct *tail_vma = NULL;
2417
2418 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002419 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002421 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 mm->map_count--;
2423 tail_vma = vma;
2424 vma = vma->vm_next;
2425 } while (vma && vma->vm_start < end);
2426 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002427 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002428 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002429 vma_gap_update(vma);
2430 } else
2431 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002433
2434 /* Kill the cache */
2435 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436}
2437
2438/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002439 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2440 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002442static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 unsigned long addr, int new_below)
2444{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08002446 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Andi Kleena5516432008-07-23 21:27:41 -07002448 if (is_vm_hugetlb_page(vma) && (addr &
2449 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 return -EINVAL;
2451
Christoph Lametere94b1762006-12-06 20:33:17 -08002452 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002454 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
2456 /* most fields are the same, copy all, and then fixup */
2457 *new = *vma;
2458
Rik van Riel5beb4932010-03-05 13:42:07 -08002459 INIT_LIST_HEAD(&new->anon_vma_chain);
2460
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 if (new_below)
2462 new->vm_end = addr;
2463 else {
2464 new->vm_start = addr;
2465 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2466 }
2467
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002468 err = vma_dup_policy(vma, new);
2469 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002470 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Rik van Riel5beb4932010-03-05 13:42:07 -08002472 if (anon_vma_clone(new, vma))
2473 goto out_free_mpol;
2474
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002475 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 get_file(new->vm_file);
2477
2478 if (new->vm_ops && new->vm_ops->open)
2479 new->vm_ops->open(new);
2480
2481 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002482 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2484 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002485 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486
Rik van Riel5beb4932010-03-05 13:42:07 -08002487 /* Success. */
2488 if (!err)
2489 return 0;
2490
2491 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002492 if (new->vm_ops && new->vm_ops->close)
2493 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002494 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002495 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002496 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002497 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002498 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002499 out_free_vma:
2500 kmem_cache_free(vm_area_cachep, new);
2501 out_err:
2502 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503}
2504
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002505/*
2506 * Split a vma into two pieces at address 'addr', a new vma is allocated
2507 * either for the first part or the tail.
2508 */
2509int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2510 unsigned long addr, int new_below)
2511{
2512 if (mm->map_count >= sysctl_max_map_count)
2513 return -ENOMEM;
2514
2515 return __split_vma(mm, vma, addr, new_below);
2516}
2517
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518/* Munmap is split into 2 main parts -- this part which finds
2519 * what needs doing, and the areas themselves, which do the
2520 * work. This now handles partial unmappings.
2521 * Jeremy Fitzhardinge <jeremy@goop.org>
2522 */
2523int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2524{
2525 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002526 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
2528 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2529 return -EINVAL;
2530
vishnu.pscc71aba2014-10-09 15:26:29 -07002531 len = PAGE_ALIGN(len);
2532 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 return -EINVAL;
2534
2535 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002536 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002537 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002539 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002540 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
2542 /* if it doesn't overlap, we have nothing.. */
2543 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002544 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 return 0;
2546
2547 /*
2548 * If we need to split any vma, do it now to save pain later.
2549 *
2550 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2551 * unmapped vm_area_struct will remain in use: so lower split_vma
2552 * places tmp vma above, and higher split_vma places tmp vma below.
2553 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002554 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002555 int error;
2556
2557 /*
2558 * Make sure that map_count on return from munmap() will
2559 * not exceed its limit; but let map_count go just above
2560 * its limit temporarily, to help free resources as expected.
2561 */
2562 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2563 return -ENOMEM;
2564
2565 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 if (error)
2567 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002568 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 }
2570
2571 /* Does it split the last one? */
2572 last = find_vma(mm, end);
2573 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002574 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 if (error)
2576 return error;
2577 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002578 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
2580 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002581 * unlock any mlock()ed ranges before detaching vmas
2582 */
2583 if (mm->locked_vm) {
2584 struct vm_area_struct *tmp = vma;
2585 while (tmp && tmp->vm_start < end) {
2586 if (tmp->vm_flags & VM_LOCKED) {
2587 mm->locked_vm -= vma_pages(tmp);
2588 munlock_vma_pages_all(tmp);
2589 }
2590 tmp = tmp->vm_next;
2591 }
2592 }
2593
2594 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 * Remove the vma's, and unmap the actual pages
2596 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002597 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2598 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599
2600 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002601 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
2603 return 0;
2604}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Al Virobfce2812012-04-20 21:57:04 -04002606int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002607{
2608 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002609 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002610
2611 down_write(&mm->mmap_sem);
2612 ret = do_munmap(mm, start, len);
2613 up_write(&mm->mmap_sem);
2614 return ret;
2615}
2616EXPORT_SYMBOL(vm_munmap);
2617
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002618SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002621 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622}
2623
2624static inline void verify_mm_writelocked(struct mm_struct *mm)
2625{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002626#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2628 WARN_ON(1);
2629 up_read(&mm->mmap_sem);
2630 }
2631#endif
2632}
2633
2634/*
2635 * this is really a simplified "do_mmap". it only handles
2636 * anonymous maps. eventually we may be able to do some
2637 * brk-specific accounting here.
2638 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002639static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640{
vishnu.pscc71aba2014-10-09 15:26:29 -07002641 struct mm_struct *mm = current->mm;
2642 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002644 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002646 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
2648 len = PAGE_ALIGN(len);
2649 if (!len)
2650 return addr;
2651
Kirill Korotaev3a459752006-09-07 14:17:04 +04002652 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2653
Al Viro2c6a1012009-12-03 19:40:46 -05002654 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2655 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002656 return error;
2657
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002658 error = mlock_future_check(mm, mm->def_flags, len);
2659 if (error)
2660 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
2662 /*
2663 * mm->mmap_sem is required to protect against another thread
2664 * changing the mappings in case we sleep.
2665 */
2666 verify_mm_writelocked(mm);
2667
2668 /*
2669 * Clear old maps. this also does some error checking for us
2670 */
2671 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002672 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 if (do_munmap(mm, addr, len))
2674 return -ENOMEM;
2675 goto munmap_back;
2676 }
2677
2678 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002679 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 return -ENOMEM;
2681
2682 if (mm->map_count > sysctl_max_map_count)
2683 return -ENOMEM;
2684
Al Viro191c5422012-02-13 03:58:52 +00002685 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 return -ENOMEM;
2687
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002689 vma = vma_merge(mm, prev, addr, addr + len, flags,
2690 NULL, NULL, pgoff, NULL);
2691 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 goto out;
2693
2694 /*
2695 * create a vma struct for an anonymous mapping
2696 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002697 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 if (!vma) {
2699 vm_unacct_memory(len >> PAGE_SHIFT);
2700 return -ENOMEM;
2701 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702
Rik van Riel5beb4932010-03-05 13:42:07 -08002703 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 vma->vm_mm = mm;
2705 vma->vm_start = addr;
2706 vma->vm_end = addr + len;
2707 vma->vm_pgoff = pgoff;
2708 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002709 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 vma_link(mm, vma, prev, rb_link, rb_parent);
2711out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002712 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002714 if (flags & VM_LOCKED)
2715 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002716 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 return addr;
2718}
2719
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002720unsigned long vm_brk(unsigned long addr, unsigned long len)
2721{
2722 struct mm_struct *mm = current->mm;
2723 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002724 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002725
2726 down_write(&mm->mmap_sem);
2727 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002728 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002729 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002730 if (populate)
2731 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002732 return ret;
2733}
2734EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
2736/* Release all mmaps. */
2737void exit_mmap(struct mm_struct *mm)
2738{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002739 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002740 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 unsigned long nr_accounted = 0;
2742
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002743 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002744 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002745
Rik van Rielba470de2008-10-18 20:26:50 -07002746 if (mm->locked_vm) {
2747 vma = mm->mmap;
2748 while (vma) {
2749 if (vma->vm_flags & VM_LOCKED)
2750 munlock_vma_pages_all(vma);
2751 vma = vma->vm_next;
2752 }
2753 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002754
2755 arch_exit_mmap(mm);
2756
Rik van Rielba470de2008-10-18 20:26:50 -07002757 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002758 if (!vma) /* Can happen if dup_mmap() received an OOM */
2759 return;
2760
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002763 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002764 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002765 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002766 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002767
Hugh Dickins6ee86302013-04-29 15:07:44 -07002768 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002769 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002772 * Walk the list again, actually closing and freeing it,
2773 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002775 while (vma) {
2776 if (vma->vm_flags & VM_ACCOUNT)
2777 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002778 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002779 }
2780 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002781
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -08002782 WARN_ON(atomic_long_read(&mm->nr_ptes) >
2783 (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784}
2785
2786/* Insert vm structure into process list sorted by address
2787 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002788 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002790int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791{
Hugh Dickins6597d782012-10-08 16:29:07 -07002792 struct vm_area_struct *prev;
2793 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
2795 /*
2796 * The vm_pgoff of a purely anonymous vma should be irrelevant
2797 * until its first write fault, when page's anon_vma and index
2798 * are set. But now set the vm_pgoff it will almost certainly
2799 * end up with (unless mremap moves it elsewhere before that
2800 * first wfault), so /proc/pid/maps tells a consistent story.
2801 *
2802 * By setting it to reflect the virtual start address of the
2803 * vma, merges and splits can happen in a seamless way, just
2804 * using the existing file pgoff checks and manipulations.
2805 * Similarly in do_mmap_pgoff and in do_brk.
2806 */
2807 if (!vma->vm_file) {
2808 BUG_ON(vma->anon_vma);
2809 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2810 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002811 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2812 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002814 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002815 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002816 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302817
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 vma_link(mm, vma, prev, rb_link, rb_parent);
2819 return 0;
2820}
2821
2822/*
2823 * Copy the vma structure to a new location in the same mm,
2824 * prior to moving page table entries, to effect an mremap move.
2825 */
2826struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002827 unsigned long addr, unsigned long len, pgoff_t pgoff,
2828 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829{
2830 struct vm_area_struct *vma = *vmap;
2831 unsigned long vma_start = vma->vm_start;
2832 struct mm_struct *mm = vma->vm_mm;
2833 struct vm_area_struct *new_vma, *prev;
2834 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002835 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
2837 /*
2838 * If anonymous vma has not yet been faulted, update new pgoff
2839 * to match new location, to increase its chance of merging.
2840 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002841 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002843 faulted_in_anon_vma = false;
2844 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845
Hugh Dickins6597d782012-10-08 16:29:07 -07002846 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2847 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2849 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2850 if (new_vma) {
2851 /*
2852 * Source vma may have been merged into new_vma
2853 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002854 if (unlikely(vma_start >= new_vma->vm_start &&
2855 vma_start < new_vma->vm_end)) {
2856 /*
2857 * The only way we can get a vma_merge with
2858 * self during an mremap is if the vma hasn't
2859 * been faulted in yet and we were allowed to
2860 * reset the dst vma->vm_pgoff to the
2861 * destination address of the mremap to allow
2862 * the merge to happen. mremap must change the
2863 * vm_pgoff linearity between src and dst vmas
2864 * (in turn preventing a vma_merge) to be
2865 * safe. It is only safe to keep the vm_pgoff
2866 * linear if there are no pages mapped yet.
2867 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002868 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002869 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002870 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002871 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002873 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 if (new_vma) {
2875 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 new_vma->vm_start = addr;
2877 new_vma->vm_end = addr + len;
2878 new_vma->vm_pgoff = pgoff;
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002879 if (vma_dup_policy(vma, new_vma))
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002880 goto out_free_vma;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002881 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2882 if (anon_vma_clone(new_vma, vma))
2883 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002884 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 get_file(new_vma->vm_file);
2886 if (new_vma->vm_ops && new_vma->vm_ops->open)
2887 new_vma->vm_ops->open(new_vma);
2888 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002889 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 }
2891 }
2892 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002893
2894 out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002895 mpol_put(vma_policy(new_vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002896 out_free_vma:
2897 kmem_cache_free(vm_area_cachep, new_vma);
2898 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002900
2901/*
2902 * Return true if the calling process may expand its vm space by the passed
2903 * number of pages
2904 */
2905int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2906{
2907 unsigned long cur = mm->total_vm; /* pages */
2908 unsigned long lim;
2909
Jiri Slaby59e99e52010-03-05 13:41:44 -08002910 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002911
2912 if (cur + npages > lim)
2913 return 0;
2914 return 1;
2915}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002916
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002917static int special_mapping_fault(struct vm_area_struct *vma,
2918 struct vm_fault *vmf);
2919
2920/*
2921 * Having a close hook prevents vma merging regardless of flags.
2922 */
2923static void special_mapping_close(struct vm_area_struct *vma)
2924{
2925}
2926
2927static const char *special_mapping_name(struct vm_area_struct *vma)
2928{
2929 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
2930}
2931
2932static const struct vm_operations_struct special_mapping_vmops = {
2933 .close = special_mapping_close,
2934 .fault = special_mapping_fault,
2935 .name = special_mapping_name,
2936};
2937
2938static const struct vm_operations_struct legacy_special_mapping_vmops = {
2939 .close = special_mapping_close,
2940 .fault = special_mapping_fault,
2941};
Roland McGrathfa5dc222007-02-08 14:20:41 -08002942
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002943static int special_mapping_fault(struct vm_area_struct *vma,
2944 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002945{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002946 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002947 struct page **pages;
2948
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002949 /*
2950 * special mappings have no vm_file, and in that case, the mm
2951 * uses vm_pgoff internally. So we have to subtract it from here.
2952 * We are allowed to do this because we are the mm; do not copy
2953 * this code into drivers!
2954 */
2955 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002956
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002957 if (vma->vm_ops == &legacy_special_mapping_vmops)
2958 pages = vma->vm_private_data;
2959 else
2960 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
2961 pages;
2962
2963 for (; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002964 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002965
2966 if (*pages) {
2967 struct page *page = *pages;
2968 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002969 vmf->page = page;
2970 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002971 }
2972
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002973 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002974}
2975
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002976static struct vm_area_struct *__install_special_mapping(
2977 struct mm_struct *mm,
2978 unsigned long addr, unsigned long len,
2979 unsigned long vm_flags, const struct vm_operations_struct *ops,
2980 void *priv)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002981{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002982 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002983 struct vm_area_struct *vma;
2984
2985 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2986 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01002987 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002988
Rik van Riel5beb4932010-03-05 13:42:07 -08002989 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002990 vma->vm_mm = mm;
2991 vma->vm_start = addr;
2992 vma->vm_end = addr + len;
2993
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002994 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07002995 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002996
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002997 vma->vm_ops = ops;
2998 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002999
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003000 ret = insert_vm_struct(mm, vma);
3001 if (ret)
3002 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003003
3004 mm->total_vm += len >> PAGE_SHIFT;
3005
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003006 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003007
Stefani Seibold3935ed62014-03-17 23:22:02 +01003008 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003009
3010out:
3011 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003012 return ERR_PTR(ret);
3013}
3014
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003015/*
3016 * Called with mm->mmap_sem held for writing.
3017 * Insert a new vma covering the given region, with the given flags.
3018 * Its pages are supplied by the given array of struct page *.
3019 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3020 * The region past the last page supplied will always produce SIGBUS.
3021 * The array pointer and the pages it points to are assumed to stay alive
3022 * for as long as this mapping might exist.
3023 */
3024struct vm_area_struct *_install_special_mapping(
3025 struct mm_struct *mm,
3026 unsigned long addr, unsigned long len,
3027 unsigned long vm_flags, const struct vm_special_mapping *spec)
3028{
3029 return __install_special_mapping(mm, addr, len, vm_flags,
3030 &special_mapping_vmops, (void *)spec);
3031}
3032
Stefani Seibold3935ed62014-03-17 23:22:02 +01003033int install_special_mapping(struct mm_struct *mm,
3034 unsigned long addr, unsigned long len,
3035 unsigned long vm_flags, struct page **pages)
3036{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003037 struct vm_area_struct *vma = __install_special_mapping(
3038 mm, addr, len, vm_flags, &legacy_special_mapping_vmops,
3039 (void *)pages);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003040
Duan Jiong14bd5b42014-06-04 16:07:05 -07003041 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003042}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003043
3044static DEFINE_MUTEX(mm_all_locks_mutex);
3045
Peter Zijlstra454ed842008-08-11 09:30:25 +02003046static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003047{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003048 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003049 /*
3050 * The LSB of head.next can't change from under us
3051 * because we hold the mm_all_locks_mutex.
3052 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003053 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003054 /*
3055 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003056 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003057 * the same anon_vma we won't take it again.
3058 *
3059 * No need of atomic instructions here, head.next
3060 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003061 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003062 */
3063 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003064 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003065 BUG();
3066 }
3067}
3068
Peter Zijlstra454ed842008-08-11 09:30:25 +02003069static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003070{
3071 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3072 /*
3073 * AS_MM_ALL_LOCKS can't change from under us because
3074 * we hold the mm_all_locks_mutex.
3075 *
3076 * Operations on ->flags have to be atomic because
3077 * even if AS_MM_ALL_LOCKS is stable thanks to the
3078 * mm_all_locks_mutex, there may be other cpus
3079 * changing other bitflags in parallel to us.
3080 */
3081 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3082 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003083 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003084 }
3085}
3086
3087/*
3088 * This operation locks against the VM for all pte/vma/mm related
3089 * operations that could ever happen on a certain mm. This includes
3090 * vmtruncate, try_to_unmap, and all page faults.
3091 *
3092 * The caller must take the mmap_sem in write mode before calling
3093 * mm_take_all_locks(). The caller isn't allowed to release the
3094 * mmap_sem until mm_drop_all_locks() returns.
3095 *
3096 * mmap_sem in write mode is required in order to block all operations
3097 * that could modify pagetables and free pages without need of
3098 * altering the vma layout (for example populate_range() with
3099 * nonlinear vmas). It's also needed in write mode to avoid new
3100 * anon_vmas to be associated with existing vmas.
3101 *
3102 * A single task can't take more than one mm_take_all_locks() in a row
3103 * or it would deadlock.
3104 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003105 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003106 * mapping->flags avoid to take the same lock twice, if more than one
3107 * vma in this mm is backed by the same anon_vma or address_space.
3108 *
3109 * We can take all the locks in random order because the VM code
Yuanhan Liu631b0cf2013-02-04 14:28:48 -08003110 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003111 * takes more than one of them in a row. Secondly we're protected
3112 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3113 *
3114 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3115 * that may have to take thousand of locks.
3116 *
3117 * mm_take_all_locks() can fail if it's interrupted by signals.
3118 */
3119int mm_take_all_locks(struct mm_struct *mm)
3120{
3121 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003122 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003123
3124 BUG_ON(down_read_trylock(&mm->mmap_sem));
3125
3126 mutex_lock(&mm_all_locks_mutex);
3127
3128 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3129 if (signal_pending(current))
3130 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003131 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003132 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003133 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003134
3135 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3136 if (signal_pending(current))
3137 goto out_unlock;
3138 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003139 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3140 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003141 }
3142
Kautuk Consul584cff52011-10-31 17:08:59 -07003143 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003144
3145out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003146 mm_drop_all_locks(mm);
3147 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003148}
3149
3150static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3151{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003152 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003153 /*
3154 * The LSB of head.next can't change to 0 from under
3155 * us because we hold the mm_all_locks_mutex.
3156 *
3157 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003158 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003159 * never see our bitflag.
3160 *
3161 * No need of atomic instructions here, head.next
3162 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003163 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003164 */
3165 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003166 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003167 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003168 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003169 }
3170}
3171
3172static void vm_unlock_mapping(struct address_space *mapping)
3173{
3174 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3175 /*
3176 * AS_MM_ALL_LOCKS can't change to 0 from under us
3177 * because we hold the mm_all_locks_mutex.
3178 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003179 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003180 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3181 &mapping->flags))
3182 BUG();
3183 }
3184}
3185
3186/*
3187 * The mmap_sem cannot be released by the caller until
3188 * mm_drop_all_locks() returns.
3189 */
3190void mm_drop_all_locks(struct mm_struct *mm)
3191{
3192 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003193 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003194
3195 BUG_ON(down_read_trylock(&mm->mmap_sem));
3196 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3197
3198 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3199 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003200 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3201 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003202 if (vma->vm_file && vma->vm_file->f_mapping)
3203 vm_unlock_mapping(vma->vm_file->f_mapping);
3204 }
3205
3206 mutex_unlock(&mm_all_locks_mutex);
3207}
David Howells8feae132009-01-08 12:04:47 +00003208
3209/*
3210 * initialise the VMA slab
3211 */
3212void __init mmap_init(void)
3213{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003214 int ret;
3215
Tejun Heo908c7f12014-09-08 09:51:29 +09003216 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003217 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003218}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003219
3220/*
3221 * Initialise sysctl_user_reserve_kbytes.
3222 *
3223 * This is intended to prevent a user from starting a single memory hogging
3224 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3225 * mode.
3226 *
3227 * The default value is min(3% of free memory, 128MB)
3228 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3229 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003230static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003231{
3232 unsigned long free_kbytes;
3233
3234 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3235
3236 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3237 return 0;
3238}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003239subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003240
3241/*
3242 * Initialise sysctl_admin_reserve_kbytes.
3243 *
3244 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3245 * to log in and kill a memory hogging process.
3246 *
3247 * Systems with more than 256MB will reserve 8MB, enough to recover
3248 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3249 * only reserve 3% of free pages by default.
3250 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003251static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003252{
3253 unsigned long free_kbytes;
3254
3255 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3256
3257 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3258 return 0;
3259}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003260subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003261
3262/*
3263 * Reinititalise user and admin reserves if memory is added or removed.
3264 *
3265 * The default user reserve max is 128MB, and the default max for the
3266 * admin reserve is 8MB. These are usually, but not always, enough to
3267 * enable recovery from a memory hogging process using login/sshd, a shell,
3268 * and tools like top. It may make sense to increase or even disable the
3269 * reserve depending on the existence of swap or variations in the recovery
3270 * tools. So, the admin may have changed them.
3271 *
3272 * If memory is added and the reserves have been eliminated or increased above
3273 * the default max, then we'll trust the admin.
3274 *
3275 * If memory is removed and there isn't enough free memory, then we
3276 * need to reset the reserves.
3277 *
3278 * Otherwise keep the reserve set by the admin.
3279 */
3280static int reserve_mem_notifier(struct notifier_block *nb,
3281 unsigned long action, void *data)
3282{
3283 unsigned long tmp, free_kbytes;
3284
3285 switch (action) {
3286 case MEM_ONLINE:
3287 /* Default max is 128MB. Leave alone if modified by operator. */
3288 tmp = sysctl_user_reserve_kbytes;
3289 if (0 < tmp && tmp < (1UL << 17))
3290 init_user_reserve();
3291
3292 /* Default max is 8MB. Leave alone if modified by operator. */
3293 tmp = sysctl_admin_reserve_kbytes;
3294 if (0 < tmp && tmp < (1UL << 13))
3295 init_admin_reserve();
3296
3297 break;
3298 case MEM_OFFLINE:
3299 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3300
3301 if (sysctl_user_reserve_kbytes > free_kbytes) {
3302 init_user_reserve();
3303 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3304 sysctl_user_reserve_kbytes);
3305 }
3306
3307 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3308 init_admin_reserve();
3309 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3310 sysctl_admin_reserve_kbytes);
3311 }
3312 break;
3313 default:
3314 break;
3315 }
3316 return NOTIFY_OK;
3317}
3318
3319static struct notifier_block reserve_mem_nb = {
3320 .notifier_call = reserve_mem_notifier,
3321};
3322
3323static int __meminit init_reserve_notifier(void)
3324{
3325 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003326 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003327
3328 return 0;
3329}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003330subsys_initcall(init_reserve_notifier);