blob: da9990acc08b2d8014e342771a52650b47ee05fc [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{
Roman Gushchin5703b082015-02-11 15:28:39 -0800155 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);
Roman Gushchin5703b082015-02-11 15:28:39 -0800223 allowed -= min_t(long, mm->total_vm / 32, reserve);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700226 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700228error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 vm_unacct_memory(pages);
230
231 return -ENOMEM;
232}
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800235 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 */
237static void __remove_shared_vm_struct(struct vm_area_struct *vma,
238 struct file *file, struct address_space *mapping)
239{
240 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500241 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700243 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800246 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 flush_dcache_mmap_unlock(mapping);
248}
249
250/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700251 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700252 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700254void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 struct file *file = vma->vm_file;
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 if (file) {
259 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800260 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800262 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700264}
265
266/*
267 * Close a vm structure and free it, returning the next.
268 */
269static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
270{
271 struct vm_area_struct *next = vma->vm_next;
272
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700273 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 if (vma->vm_ops && vma->vm_ops->close)
275 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700276 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700277 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700278 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700280 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281}
282
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700283static unsigned long do_brk(unsigned long addr, unsigned long len);
284
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100285SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700287 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 unsigned long newbrk, oldbrk;
289 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700290 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800291 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293 down_write(&mm->mmap_sem);
294
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700295#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800296 /*
297 * CONFIG_COMPAT_BRK can still be overridden by setting
298 * randomize_va_space to 2, which will still cause mm->start_brk
299 * to be arbitrarily shifted
300 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700301 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800302 min_brk = mm->start_brk;
303 else
304 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700305#else
306 min_brk = mm->start_brk;
307#endif
308 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700310
311 /*
312 * Check against rlimit here. If this check is done later after the test
313 * of oldbrk with newbrk then it can escape the test and let the data
314 * segment grow beyond its set limit the in case where the limit is
315 * not page aligned -Ram Gupta
316 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700317 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
318 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700319 goto out;
320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 newbrk = PAGE_ALIGN(brk);
322 oldbrk = PAGE_ALIGN(mm->brk);
323 if (oldbrk == newbrk)
324 goto set_brk;
325
326 /* Always allow shrinking brk. */
327 if (brk <= mm->brk) {
328 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
329 goto set_brk;
330 goto out;
331 }
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 /* Check against existing mmap mappings. */
334 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
335 goto out;
336
337 /* Ok, looks good - let it rip. */
338 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
339 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341set_brk:
342 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800343 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
344 up_write(&mm->mmap_sem);
345 if (populate)
346 mm_populate(oldbrk, newbrk - oldbrk);
347 return brk;
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349out:
350 retval = mm->brk;
351 up_write(&mm->mmap_sem);
352 return retval;
353}
354
Michel Lespinassed3737182012-12-11 16:01:38 -0800355static long vma_compute_subtree_gap(struct vm_area_struct *vma)
356{
357 unsigned long max, subtree_gap;
358 max = vma->vm_start;
359 if (vma->vm_prev)
360 max -= vma->vm_prev->vm_end;
361 if (vma->vm_rb.rb_left) {
362 subtree_gap = rb_entry(vma->vm_rb.rb_left,
363 struct vm_area_struct, vm_rb)->rb_subtree_gap;
364 if (subtree_gap > max)
365 max = subtree_gap;
366 }
367 if (vma->vm_rb.rb_right) {
368 subtree_gap = rb_entry(vma->vm_rb.rb_right,
369 struct vm_area_struct, vm_rb)->rb_subtree_gap;
370 if (subtree_gap > max)
371 max = subtree_gap;
372 }
373 return max;
374}
375
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700376#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377static int browse_rb(struct rb_root *root)
378{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800379 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 struct rb_node *nd, *pn = NULL;
381 unsigned long prev = 0, pend = 0;
382
383 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
384 struct vm_area_struct *vma;
385 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800386 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700387 pr_emerg("vm_start %lx < prev %lx\n",
388 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800389 bug = 1;
390 }
391 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700392 pr_emerg("vm_start %lx < pend %lx\n",
393 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800394 bug = 1;
395 }
396 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700397 pr_emerg("vm_start %lx > vm_end %lx\n",
398 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800399 bug = 1;
400 }
401 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700402 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800403 vma->rb_subtree_gap,
404 vma_compute_subtree_gap(vma));
405 bug = 1;
406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 i++;
408 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800409 prev = vma->vm_start;
410 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 }
412 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800413 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800415 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700416 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800417 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800419 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
Michel Lespinassed3737182012-12-11 16:01:38 -0800422static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
423{
424 struct rb_node *nd;
425
426 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
427 struct vm_area_struct *vma;
428 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700429 VM_BUG_ON_VMA(vma != ignore &&
430 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
431 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700435static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
437 int bug = 0;
438 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800439 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700440 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700441
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700442 while (vma) {
443 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700444
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800445 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700446 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
447 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800448 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800449 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700450 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 i++;
452 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800453 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700454 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800455 bug = 1;
456 }
457 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700458 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700459 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800460 bug = 1;
461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800463 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700464 if (i != -1)
465 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800466 bug = 1;
467 }
Sasha Levin96dad672014-10-09 15:28:39 -0700468 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800471#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472#define validate_mm(mm) do { } while (0)
473#endif
474
Michel Lespinassed3737182012-12-11 16:01:38 -0800475RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
476 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
477
478/*
479 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
480 * vma->vm_prev->vm_end values changed, without modifying the vma's position
481 * in the rbtree.
482 */
483static void vma_gap_update(struct vm_area_struct *vma)
484{
485 /*
486 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
487 * function that does exacltly what we want.
488 */
489 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
490}
491
492static inline void vma_rb_insert(struct vm_area_struct *vma,
493 struct rb_root *root)
494{
495 /* All rb_subtree_gap values must be consistent prior to insertion */
496 validate_mm_rb(root, NULL);
497
498 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
499}
500
501static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
502{
503 /*
504 * All rb_subtree_gap values must be consistent prior to erase,
505 * with the possible exception of the vma being erased.
506 */
507 validate_mm_rb(root, vma);
508
509 /*
510 * Note rb_erase_augmented is a fairly large inline function,
511 * so make sure we instantiate it only once with our desired
512 * augmented rbtree callbacks.
513 */
514 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
515}
516
Michel Lespinassebf181b92012-10-08 16:31:39 -0700517/*
518 * vma has some anon_vma assigned, and is already inserted on that
519 * anon_vma's interval trees.
520 *
521 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
522 * vma must be removed from the anon_vma's interval trees using
523 * anon_vma_interval_tree_pre_update_vma().
524 *
525 * After the update, the vma will be reinserted using
526 * anon_vma_interval_tree_post_update_vma().
527 *
528 * The entire update must be protected by exclusive mmap_sem and by
529 * the root anon_vma's mutex.
530 */
531static inline void
532anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
533{
534 struct anon_vma_chain *avc;
535
536 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
537 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
538}
539
540static inline void
541anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
542{
543 struct anon_vma_chain *avc;
544
545 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
546 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
547}
548
Hugh Dickins6597d782012-10-08 16:29:07 -0700549static int find_vma_links(struct mm_struct *mm, unsigned long addr,
550 unsigned long end, struct vm_area_struct **pprev,
551 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
Hugh Dickins6597d782012-10-08 16:29:07 -0700553 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 __rb_link = &mm->mm_rb.rb_node;
556 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 while (*__rb_link) {
559 struct vm_area_struct *vma_tmp;
560
561 __rb_parent = *__rb_link;
562 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
563
564 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700565 /* Fail if an existing vma overlaps the area */
566 if (vma_tmp->vm_start < end)
567 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 __rb_link = &__rb_parent->rb_left;
569 } else {
570 rb_prev = __rb_parent;
571 __rb_link = &__rb_parent->rb_right;
572 }
573 }
574
575 *pprev = NULL;
576 if (rb_prev)
577 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
578 *rb_link = __rb_link;
579 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700580 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581}
582
Cyril Hrubise8420a82013-04-29 15:08:33 -0700583static unsigned long count_vma_pages_range(struct mm_struct *mm,
584 unsigned long addr, unsigned long end)
585{
586 unsigned long nr_pages = 0;
587 struct vm_area_struct *vma;
588
589 /* Find first overlaping mapping */
590 vma = find_vma_intersection(mm, addr, end);
591 if (!vma)
592 return 0;
593
594 nr_pages = (min(end, vma->vm_end) -
595 max(addr, vma->vm_start)) >> PAGE_SHIFT;
596
597 /* Iterate over the rest of the overlaps */
598 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
599 unsigned long overlap_len;
600
601 if (vma->vm_start > end)
602 break;
603
604 overlap_len = min(end, vma->vm_end) - vma->vm_start;
605 nr_pages += overlap_len >> PAGE_SHIFT;
606 }
607
608 return nr_pages;
609}
610
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
612 struct rb_node **rb_link, struct rb_node *rb_parent)
613{
Michel Lespinassed3737182012-12-11 16:01:38 -0800614 /* Update tracking information for the gap following the new vma. */
615 if (vma->vm_next)
616 vma_gap_update(vma->vm_next);
617 else
618 mm->highest_vm_end = vma->vm_end;
619
620 /*
621 * vma->vm_prev wasn't known when we followed the rbtree to find the
622 * correct insertion point for that vma. As a result, we could not
623 * update the vma vm_rb parents rb_subtree_gap values on the way down.
624 * So, we first insert the vma with a zero rb_subtree_gap value
625 * (to be consistent with what we did on the way down), and then
626 * immediately update the gap to the correct value. Finally we
627 * rebalance the rbtree after all augmented values have been set.
628 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800630 vma->rb_subtree_gap = 0;
631 vma_gap_update(vma);
632 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
634
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700635static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
ZhenwenXu48aae422009-01-06 14:40:21 -0800637 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639 file = vma->vm_file;
640 if (file) {
641 struct address_space *mapping = file->f_mapping;
642
643 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500644 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700646 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
648 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800649 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 flush_dcache_mmap_unlock(mapping);
651 }
652}
653
654static void
655__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
656 struct vm_area_struct *prev, struct rb_node **rb_link,
657 struct rb_node *rb_parent)
658{
659 __vma_link_list(mm, vma, prev, rb_parent);
660 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661}
662
663static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
664 struct vm_area_struct *prev, struct rb_node **rb_link,
665 struct rb_node *rb_parent)
666{
667 struct address_space *mapping = NULL;
668
Huang Shijie64ac4942014-06-04 16:07:33 -0700669 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800671 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
674 __vma_link(mm, vma, prev, rb_link, rb_parent);
675 __vma_link_file(vma);
676
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800678 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
680 mm->map_count++;
681 validate_mm(mm);
682}
683
684/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700685 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700686 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800688static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
Hugh Dickins6597d782012-10-08 16:29:07 -0700690 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800691 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Hugh Dickins6597d782012-10-08 16:29:07 -0700693 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
694 &prev, &rb_link, &rb_parent))
695 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 __vma_link(mm, vma, prev, rb_link, rb_parent);
697 mm->map_count++;
698}
699
700static inline void
701__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
702 struct vm_area_struct *prev)
703{
Michel Lespinassed3737182012-12-11 16:01:38 -0800704 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700705
Michel Lespinassed3737182012-12-11 16:01:38 -0800706 vma_rb_erase(vma, &mm->mm_rb);
707 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700708 if (next)
709 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700710
711 /* Kill the cache */
712 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713}
714
715/*
716 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
717 * is already present in an i_mmap tree without adjusting the tree.
718 * The following helper function should be used when such adjustments
719 * are necessary. The "insert" vma (if any) is to be inserted
720 * before we drop the necessary locks.
721 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800722int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
724{
725 struct mm_struct *mm = vma->vm_mm;
726 struct vm_area_struct *next = vma->vm_next;
727 struct vm_area_struct *importer = NULL;
728 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700729 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700730 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800732 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 long adjust_next = 0;
734 int remove_next = 0;
735
736 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700737 struct vm_area_struct *exporter = NULL;
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (end >= next->vm_end) {
740 /*
741 * vma expands, overlapping all the next, and
742 * perhaps the one after too (mprotect case 6).
743 */
744again: remove_next = 1 + (end > next->vm_end);
745 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700746 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 importer = vma;
748 } else if (end > next->vm_start) {
749 /*
750 * vma expands, overlapping part of the next:
751 * mprotect case 5 shifting the boundary up.
752 */
753 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700754 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 importer = vma;
756 } else if (end < vma->vm_end) {
757 /*
758 * vma shrinks, and !insert tells it's not
759 * split_vma inserting another: so it must be
760 * mprotect case 4 shifting the boundary down.
761 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700762 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700763 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 importer = next;
765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
Rik van Riel5beb4932010-03-05 13:42:07 -0800767 /*
768 * Easily overlooked: when mprotect shifts the boundary,
769 * make sure the expanding vma has anon_vma set if the
770 * shrinking vma had, to cover any anon pages imported.
771 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700772 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800773 int error;
774
Linus Torvalds287d97a2010-04-10 15:22:30 -0700775 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300776 error = anon_vma_clone(importer, exporter);
777 if (error) {
778 importer->anon_vma = NULL;
779 return error;
780 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800781 }
782 }
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 if (file) {
785 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800786 root = &mapping->i_mmap;
787 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530788
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800789 if (adjust_next)
790 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530791
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800792 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700795 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 * are visible to arm/parisc __flush_dcache_page
797 * throughout; but we cannot insert into address
798 * space until vma start or end is updated.
799 */
800 __vma_link_file(insert);
801 }
802 }
803
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800804 vma_adjust_trans_huge(vma, start, end, adjust_next);
805
Michel Lespinassebf181b92012-10-08 16:31:39 -0700806 anon_vma = vma->anon_vma;
807 if (!anon_vma && adjust_next)
808 anon_vma = next->anon_vma;
809 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700810 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
811 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000812 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700813 anon_vma_interval_tree_pre_update_vma(vma);
814 if (adjust_next)
815 anon_vma_interval_tree_pre_update_vma(next);
816 }
Rik van Riel012f18002010-08-09 17:18:40 -0700817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (root) {
819 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700820 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700822 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
824
Michel Lespinassed3737182012-12-11 16:01:38 -0800825 if (start != vma->vm_start) {
826 vma->vm_start = start;
827 start_changed = true;
828 }
829 if (end != vma->vm_end) {
830 vma->vm_end = end;
831 end_changed = true;
832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 vma->vm_pgoff = pgoff;
834 if (adjust_next) {
835 next->vm_start += adjust_next << PAGE_SHIFT;
836 next->vm_pgoff += adjust_next;
837 }
838
839 if (root) {
840 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700841 vma_interval_tree_insert(next, root);
842 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 flush_dcache_mmap_unlock(mapping);
844 }
845
846 if (remove_next) {
847 /*
848 * vma_merge has merged next into vma, and needs
849 * us to remove next before dropping the locks.
850 */
851 __vma_unlink(mm, next, vma);
852 if (file)
853 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 } else if (insert) {
855 /*
856 * split_vma has split insert from vma, and needs
857 * us to insert it before dropping the locks
858 * (it may either follow vma or precede it).
859 */
860 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800861 } else {
862 if (start_changed)
863 vma_gap_update(vma);
864 if (end_changed) {
865 if (!next)
866 mm->highest_vm_end = end;
867 else if (!adjust_next)
868 vma_gap_update(next);
869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 }
871
Michel Lespinassebf181b92012-10-08 16:31:39 -0700872 if (anon_vma) {
873 anon_vma_interval_tree_post_update_vma(vma);
874 if (adjust_next)
875 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800876 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800879 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530881 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100882 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530883
884 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100885 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530886 }
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700889 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530890 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700892 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800893 if (next->anon_vma)
894 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700896 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 kmem_cache_free(vm_area_cachep, next);
898 /*
899 * In mprotect's case 6 (see comments on vma_merge),
900 * we must remove another next too. It would clutter
901 * up the code too much to do both in one go.
902 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800903 next = vma->vm_next;
904 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800906 else if (next)
907 vma_gap_update(next);
908 else
909 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530911 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100912 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
914 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800915
916 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917}
918
919/*
920 * If the vma has a ->close operation then the driver probably needs to release
921 * per-vma resources, so we don't attempt to merge those.
922 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923static inline int is_mergeable_vma(struct vm_area_struct *vma,
924 struct file *file, unsigned long vm_flags)
925{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800926 /*
927 * VM_SOFTDIRTY should not prevent from VMA merging, if we
928 * match the flags but dirty bit -- the caller should mark
929 * merged VMA as dirty. If dirty bit won't be excluded from
930 * comparison, we increase pressue on the memory system forcing
931 * the kernel to generate new VMAs when old one could be
932 * extended instead.
933 */
934 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 return 0;
936 if (vma->vm_file != file)
937 return 0;
938 if (vma->vm_ops && vma->vm_ops->close)
939 return 0;
940 return 1;
941}
942
943static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700944 struct anon_vma *anon_vma2,
945 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946{
Shaohua Li965f55d2011-05-24 17:11:20 -0700947 /*
948 * The list_is_singular() test is to avoid merging VMA cloned from
949 * parents. This can improve scalability caused by anon_vma lock.
950 */
951 if ((!anon_vma1 || !anon_vma2) && (!vma ||
952 list_is_singular(&vma->anon_vma_chain)))
953 return 1;
954 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
956
957/*
958 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
959 * in front of (at a lower virtual address and file offset than) the vma.
960 *
961 * We cannot merge two vmas if they have differently assigned (non-NULL)
962 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
963 *
964 * We don't check here for the merged mmap wrapping around the end of pagecache
965 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
966 * wrap, nor mmaps which cover the final page at index -1UL.
967 */
968static int
969can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
970 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
971{
972 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700973 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (vma->vm_pgoff == vm_pgoff)
975 return 1;
976 }
977 return 0;
978}
979
980/*
981 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
982 * beyond (at a higher virtual address and file offset than) the vma.
983 *
984 * We cannot merge two vmas if they have differently assigned (non-NULL)
985 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
986 */
987static int
988can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
989 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
990{
991 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700992 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700994 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
996 return 1;
997 }
998 return 0;
999}
1000
1001/*
1002 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1003 * whether that can be merged with its predecessor or its successor.
1004 * Or both (it neatly fills a hole).
1005 *
1006 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1007 * certain not to be mapped by the time vma_merge is called; but when
1008 * called for mprotect, it is certain to be already mapped (either at
1009 * an offset within prev, or at the start of next), and the flags of
1010 * this area are about to be changed to vm_flags - and the no-change
1011 * case has already been eliminated.
1012 *
1013 * The following mprotect cases have to be considered, where AAAA is
1014 * the area passed down from mprotect_fixup, never extending beyond one
1015 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1016 *
1017 * AAAA AAAA AAAA AAAA
1018 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1019 * cannot merge might become might become might become
1020 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1021 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1022 * mremap move: PPPPNNNNNNNN 8
1023 * AAAA
1024 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1025 * might become case 1 below case 2 below case 3 below
1026 *
1027 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1028 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1029 */
1030struct vm_area_struct *vma_merge(struct mm_struct *mm,
1031 struct vm_area_struct *prev, unsigned long addr,
1032 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001033 struct anon_vma *anon_vma, struct file *file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 pgoff_t pgoff, struct mempolicy *policy)
1035{
1036 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1037 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001038 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 /*
1041 * We later require that vma->vm_flags == vm_flags,
1042 * so this tests vma->vm_flags & VM_SPECIAL, too.
1043 */
1044 if (vm_flags & VM_SPECIAL)
1045 return NULL;
1046
1047 if (prev)
1048 next = prev->vm_next;
1049 else
1050 next = mm->mmap;
1051 area = next;
1052 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1053 next = next->vm_next;
1054
1055 /*
1056 * Can it merge with the predecessor?
1057 */
1058 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001059 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 can_vma_merge_after(prev, vm_flags,
1061 anon_vma, file, pgoff)) {
1062 /*
1063 * OK, it can. Can we now merge in the successor as well?
1064 */
1065 if (next && end == next->vm_start &&
1066 mpol_equal(policy, vma_policy(next)) &&
1067 can_vma_merge_before(next, vm_flags,
1068 anon_vma, file, pgoff+pglen) &&
1069 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001070 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001072 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 next->vm_end, prev->vm_pgoff, NULL);
1074 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001075 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001077 if (err)
1078 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001079 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 return prev;
1081 }
1082
1083 /*
1084 * Can this new request be merged in front of next?
1085 */
1086 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001087 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 can_vma_merge_before(next, vm_flags,
1089 anon_vma, file, pgoff+pglen)) {
1090 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001091 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 addr, prev->vm_pgoff, NULL);
1093 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001094 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001096 if (err)
1097 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001098 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 return area;
1100 }
1101
1102 return NULL;
1103}
1104
1105/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001106 * Rough compatbility check to quickly see if it's even worth looking
1107 * at sharing an anon_vma.
1108 *
1109 * They need to have the same vm_file, and the flags can only differ
1110 * in things that mprotect may change.
1111 *
1112 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1113 * we can merge the two vma's. For example, we refuse to merge a vma if
1114 * there is a vm_ops->close() function, because that indicates that the
1115 * driver is doing some kind of reference counting. But that doesn't
1116 * really matter for the anon_vma sharing case.
1117 */
1118static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1119{
1120 return a->vm_end == b->vm_start &&
1121 mpol_equal(vma_policy(a), vma_policy(b)) &&
1122 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001123 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001124 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1125}
1126
1127/*
1128 * Do some basic sanity checking to see if we can re-use the anon_vma
1129 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1130 * the same as 'old', the other will be the new one that is trying
1131 * to share the anon_vma.
1132 *
1133 * NOTE! This runs with mm_sem held for reading, so it is possible that
1134 * the anon_vma of 'old' is concurrently in the process of being set up
1135 * by another page fault trying to merge _that_. But that's ok: if it
1136 * is being set up, that automatically means that it will be a singleton
1137 * acceptable for merging, so we can do all of this optimistically. But
1138 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1139 *
1140 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1141 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1142 * is to return an anon_vma that is "complex" due to having gone through
1143 * a fork).
1144 *
1145 * We also make sure that the two vma's are compatible (adjacent,
1146 * and with the same memory policies). That's all stable, even with just
1147 * a read lock on the mm_sem.
1148 */
1149static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1150{
1151 if (anon_vma_compatible(a, b)) {
1152 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1153
1154 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1155 return anon_vma;
1156 }
1157 return NULL;
1158}
1159
1160/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1162 * neighbouring vmas for a suitable anon_vma, before it goes off
1163 * to allocate a new anon_vma. It checks because a repetitive
1164 * sequence of mprotects and faults may otherwise lead to distinct
1165 * anon_vmas being allocated, preventing vma merge in subsequent
1166 * mprotect.
1167 */
1168struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1169{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001170 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173 near = vma->vm_next;
1174 if (!near)
1175 goto try_prev;
1176
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001177 anon_vma = reusable_anon_vma(near, vma, near);
1178 if (anon_vma)
1179 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001181 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (!near)
1183 goto none;
1184
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001185 anon_vma = reusable_anon_vma(near, near, vma);
1186 if (anon_vma)
1187 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188none:
1189 /*
1190 * There's no absolute need to look only at touching neighbours:
1191 * we could search further afield for "compatible" anon_vmas.
1192 * But it would probably just be a waste of time searching,
1193 * or lead to too many vmas hanging off the same anon_vma.
1194 * We're trying to allow mprotect remerging later on,
1195 * not trying to minimize memory used for anon_vmas.
1196 */
1197 return NULL;
1198}
1199
1200#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001201void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 struct file *file, long pages)
1203{
1204 const unsigned long stack_flags
1205 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1206
Huang Shijie44de9d02012-07-31 16:41:49 -07001207 mm->total_vm += pages;
1208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (file) {
1210 mm->shared_vm += pages;
1211 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1212 mm->exec_vm += pages;
1213 } else if (flags & stack_flags)
1214 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215}
1216#endif /* CONFIG_PROC_FS */
1217
1218/*
Al Viro40401532012-02-13 03:58:52 +00001219 * If a hint addr is less than mmap_min_addr change hint to be as
1220 * low as possible but still greater than mmap_min_addr
1221 */
1222static inline unsigned long round_hint_to_min(unsigned long hint)
1223{
1224 hint &= PAGE_MASK;
1225 if (((void *)hint != NULL) &&
1226 (hint < mmap_min_addr))
1227 return PAGE_ALIGN(mmap_min_addr);
1228 return hint;
1229}
1230
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001231static inline int mlock_future_check(struct mm_struct *mm,
1232 unsigned long flags,
1233 unsigned long len)
1234{
1235 unsigned long locked, lock_limit;
1236
1237 /* mlock MCL_FUTURE? */
1238 if (flags & VM_LOCKED) {
1239 locked = len >> PAGE_SHIFT;
1240 locked += mm->locked_vm;
1241 lock_limit = rlimit(RLIMIT_MEMLOCK);
1242 lock_limit >>= PAGE_SHIFT;
1243 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1244 return -EAGAIN;
1245 }
1246 return 0;
1247}
1248
Al Viro40401532012-02-13 03:58:52 +00001249/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001250 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 */
1252
Al Viroe3fc6292012-05-30 20:08:42 -04001253unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001255 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001256 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257{
vishnu.pscc71aba2014-10-09 15:26:29 -07001258 struct mm_struct *mm = current->mm;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001259 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Michel Lespinasse41badc12013-02-22 16:32:47 -08001261 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 /*
1264 * Does the application expect PROT_READ to imply PROT_EXEC?
1265 *
1266 * (the exception is when the underlying filesystem is noexec
1267 * mounted, in which case we dont add PROT_EXEC.)
1268 */
1269 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001270 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 prot |= PROT_EXEC;
1272
1273 if (!len)
1274 return -EINVAL;
1275
Eric Paris7cd94142007-11-26 18:47:40 -05001276 if (!(flags & MAP_FIXED))
1277 addr = round_hint_to_min(addr);
1278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 /* Careful about overflows.. */
1280 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001281 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 return -ENOMEM;
1283
1284 /* offset overflow? */
1285 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001286 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288 /* Too many mappings? */
1289 if (mm->map_count > sysctl_max_map_count)
1290 return -ENOMEM;
1291
1292 /* Obtain the address to map to. we verify (or select) it and ensure
1293 * that it represents a valid section of the address space.
1294 */
1295 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1296 if (addr & ~PAGE_MASK)
1297 return addr;
1298
1299 /* Do simple checking here so the lower-level routines won't have
1300 * to. we assume access permissions have been handled by the open
1301 * of the memory object, so we don't do any here.
1302 */
1303 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1304 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1305
Huang Shijiecdf7b342009-09-21 17:03:36 -07001306 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 if (!can_do_mlock())
1308 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001309
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001310 if (mlock_future_check(mm, vm_flags, len))
1311 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001314 struct inode *inode = file_inode(file);
1315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 switch (flags & MAP_TYPE) {
1317 case MAP_SHARED:
1318 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1319 return -EACCES;
1320
1321 /*
1322 * Make sure we don't allow writing to an append-only
1323 * file..
1324 */
1325 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1326 return -EACCES;
1327
1328 /*
1329 * Make sure there are no mandatory locks on the file.
1330 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001331 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 return -EAGAIN;
1333
1334 vm_flags |= VM_SHARED | VM_MAYSHARE;
1335 if (!(file->f_mode & FMODE_WRITE))
1336 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1337
1338 /* fall through */
1339 case MAP_PRIVATE:
1340 if (!(file->f_mode & FMODE_READ))
1341 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001342 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001343 if (vm_flags & VM_EXEC)
1344 return -EPERM;
1345 vm_flags &= ~VM_MAYEXEC;
1346 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001347
Al Viro72c2d532013-09-22 16:27:52 -04001348 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001349 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001350 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1351 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 break;
1353
1354 default:
1355 return -EINVAL;
1356 }
1357 } else {
1358 switch (flags & MAP_TYPE) {
1359 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001360 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1361 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001362 /*
1363 * Ignore pgoff.
1364 */
1365 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 vm_flags |= VM_SHARED | VM_MAYSHARE;
1367 break;
1368 case MAP_PRIVATE:
1369 /*
1370 * Set pgoff according to addr for anon_vma.
1371 */
1372 pgoff = addr >> PAGE_SHIFT;
1373 break;
1374 default:
1375 return -EINVAL;
1376 }
1377 }
1378
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001379 /*
1380 * Set 'VM_NORESERVE' if we should not account for the
1381 * memory use of this mapping.
1382 */
1383 if (flags & MAP_NORESERVE) {
1384 /* We honor MAP_NORESERVE if allowed to overcommit */
1385 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1386 vm_flags |= VM_NORESERVE;
1387
1388 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1389 if (file && is_file_hugepages(file))
1390 vm_flags |= VM_NORESERVE;
1391 }
1392
1393 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001394 if (!IS_ERR_VALUE(addr) &&
1395 ((vm_flags & VM_LOCKED) ||
1396 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001397 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001398 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001399}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001400
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001401SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1402 unsigned long, prot, unsigned long, flags,
1403 unsigned long, fd, unsigned long, pgoff)
1404{
1405 struct file *file = NULL;
1406 unsigned long retval = -EBADF;
1407
1408 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001409 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001410 file = fget(fd);
1411 if (!file)
1412 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001413 if (is_file_hugepages(file))
1414 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001415 retval = -EINVAL;
1416 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1417 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001418 } else if (flags & MAP_HUGETLB) {
1419 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001420 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001421
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001422 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001423 if (!hs)
1424 return -EINVAL;
1425
1426 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001427 /*
1428 * VM_NORESERVE is used because the reservations will be
1429 * taken when vm_ops->mmap() is called
1430 * A dummy user value is used because we are not locking
1431 * memory so no accounting is necessary
1432 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001433 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001434 VM_NORESERVE,
1435 &user, HUGETLB_ANONHUGE_INODE,
1436 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001437 if (IS_ERR(file))
1438 return PTR_ERR(file);
1439 }
1440
1441 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1442
Al Viroeb36c582012-05-30 20:17:35 -04001443 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001444out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001445 if (file)
1446 fput(file);
1447out:
1448 return retval;
1449}
1450
Christoph Hellwiga4679372010-03-10 15:21:15 -08001451#ifdef __ARCH_WANT_SYS_OLD_MMAP
1452struct mmap_arg_struct {
1453 unsigned long addr;
1454 unsigned long len;
1455 unsigned long prot;
1456 unsigned long flags;
1457 unsigned long fd;
1458 unsigned long offset;
1459};
1460
1461SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1462{
1463 struct mmap_arg_struct a;
1464
1465 if (copy_from_user(&a, arg, sizeof(a)))
1466 return -EFAULT;
1467 if (a.offset & ~PAGE_MASK)
1468 return -EINVAL;
1469
1470 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1471 a.offset >> PAGE_SHIFT);
1472}
1473#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1474
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001475/*
1476 * Some shared mappigns will want the pages marked read-only
1477 * to track write events. If so, we'll downgrade vm_page_prot
1478 * to the private version (using protection_map[] without the
1479 * VM_SHARED bit).
1480 */
1481int vma_wants_writenotify(struct vm_area_struct *vma)
1482{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001483 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001484
1485 /* If it was private or non-writable, the write bit is already clear */
1486 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1487 return 0;
1488
1489 /* The backer wishes to know when pages are first written to? */
1490 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1491 return 1;
1492
Peter Feiner64e45502014-10-13 15:55:46 -07001493 /* The open routine did something to the protections that pgprot_modify
1494 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001495 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001496 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001497 return 0;
1498
Peter Feiner64e45502014-10-13 15:55:46 -07001499 /* Do we need to track softdirty? */
1500 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1501 return 1;
1502
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001503 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001504 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001505 return 0;
1506
1507 /* Can the mapping track the dirty pages? */
1508 return vma->vm_file && vma->vm_file->f_mapping &&
1509 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1510}
1511
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001512/*
1513 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001514 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001515 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001516static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001517{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001518 /*
1519 * hugetlb has its own accounting separate from the core VM
1520 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1521 */
1522 if (file && is_file_hugepages(file))
1523 return 0;
1524
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001525 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1526}
1527
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001528unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001529 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001530{
1531 struct mm_struct *mm = current->mm;
1532 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001533 int error;
1534 struct rb_node **rb_link, *rb_parent;
1535 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001536
Cyril Hrubise8420a82013-04-29 15:08:33 -07001537 /* Check against address space limit. */
1538 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1539 unsigned long nr_pages;
1540
1541 /*
1542 * MAP_FIXED may remove pages of mappings that intersects with
1543 * requested mapping. Account for the pages it would unmap.
1544 */
1545 if (!(vm_flags & MAP_FIXED))
1546 return -ENOMEM;
1547
1548 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1549
1550 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1551 return -ENOMEM;
1552 }
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 /* Clear old maps */
1555 error = -ENOMEM;
1556munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001557 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (do_munmap(mm, addr, len))
1559 return -ENOMEM;
1560 goto munmap_back;
1561 }
1562
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001563 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001564 * Private writable mapping: check memory availability
1565 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001566 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001567 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001568 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001569 return -ENOMEM;
1570 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
1572
1573 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001574 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001576 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1577 if (vma)
1578 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 /*
1581 * Determine the object being mapped and call the appropriate
1582 * specific mapper. the address has already been validated, but
1583 * not unmapped, but the maps are removed from the list.
1584 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001585 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 if (!vma) {
1587 error = -ENOMEM;
1588 goto unacct_error;
1589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 vma->vm_mm = mm;
1592 vma->vm_start = addr;
1593 vma->vm_end = addr + len;
1594 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001595 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001597 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
1599 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if (vm_flags & VM_DENYWRITE) {
1601 error = deny_write_access(file);
1602 if (error)
1603 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001605 if (vm_flags & VM_SHARED) {
1606 error = mapping_map_writable(file->f_mapping);
1607 if (error)
1608 goto allow_write_and_free_vma;
1609 }
1610
1611 /* ->mmap() can change vma->vm_file, but must guarantee that
1612 * vma_link() below can deny write-access if VM_DENYWRITE is set
1613 * and map writably if VM_SHARED is set. This usually means the
1614 * new file must not have been exposed to user-space, yet.
1615 */
Al Virocb0942b2012-08-27 14:48:26 -04001616 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 error = file->f_op->mmap(file, vma);
1618 if (error)
1619 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001620
1621 /* Can addr have changed??
1622 *
1623 * Answer: Yes, several device drivers can do it in their
1624 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001625 * Bug: If addr is changed, prev, rb_link, rb_parent should
1626 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001627 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001628 WARN_ON_ONCE(addr != vma->vm_start);
1629
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001630 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001631 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 } else if (vm_flags & VM_SHARED) {
1633 error = shmem_zero_setup(vma);
1634 if (error)
1635 goto free_vma;
1636 }
1637
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001638 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001639 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001640 if (file) {
1641 if (vm_flags & VM_SHARED)
1642 mapping_unmap_writable(file->f_mapping);
1643 if (vm_flags & VM_DENYWRITE)
1644 allow_write_access(file);
1645 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001646 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001647out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001648 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001649
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001650 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001652 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1653 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001654 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001655 else
1656 vma->vm_flags &= ~VM_LOCKED;
1657 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301658
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001659 if (file)
1660 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301661
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001662 /*
1663 * New (or expanded) vma always get soft dirty status.
1664 * Otherwise user-space soft-dirty page tracker won't
1665 * be able to distinguish situation when vma area unmapped,
1666 * then new mapped in-place (which must be aimed as
1667 * a completely new data area).
1668 */
1669 vma->vm_flags |= VM_SOFTDIRTY;
1670
Peter Feiner64e45502014-10-13 15:55:46 -07001671 vma_set_page_prot(vma);
1672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return addr;
1674
1675unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 vma->vm_file = NULL;
1677 fput(file);
1678
1679 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001680 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1681 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001682 if (vm_flags & VM_SHARED)
1683 mapping_unmap_writable(file->f_mapping);
1684allow_write_and_free_vma:
1685 if (vm_flags & VM_DENYWRITE)
1686 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687free_vma:
1688 kmem_cache_free(vm_area_cachep, vma);
1689unacct_error:
1690 if (charged)
1691 vm_unacct_memory(charged);
1692 return error;
1693}
1694
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001695unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1696{
1697 /*
1698 * We implement the search by looking for an rbtree node that
1699 * immediately follows a suitable gap. That is,
1700 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1701 * - gap_end = vma->vm_start >= info->low_limit + length;
1702 * - gap_end - gap_start >= length
1703 */
1704
1705 struct mm_struct *mm = current->mm;
1706 struct vm_area_struct *vma;
1707 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1708
1709 /* Adjust search length to account for worst case alignment overhead */
1710 length = info->length + info->align_mask;
1711 if (length < info->length)
1712 return -ENOMEM;
1713
1714 /* Adjust search limits by the desired length */
1715 if (info->high_limit < length)
1716 return -ENOMEM;
1717 high_limit = info->high_limit - length;
1718
1719 if (info->low_limit > high_limit)
1720 return -ENOMEM;
1721 low_limit = info->low_limit + length;
1722
1723 /* Check if rbtree root looks promising */
1724 if (RB_EMPTY_ROOT(&mm->mm_rb))
1725 goto check_highest;
1726 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1727 if (vma->rb_subtree_gap < length)
1728 goto check_highest;
1729
1730 while (true) {
1731 /* Visit left subtree if it looks promising */
1732 gap_end = vma->vm_start;
1733 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1734 struct vm_area_struct *left =
1735 rb_entry(vma->vm_rb.rb_left,
1736 struct vm_area_struct, vm_rb);
1737 if (left->rb_subtree_gap >= length) {
1738 vma = left;
1739 continue;
1740 }
1741 }
1742
1743 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1744check_current:
1745 /* Check if current node has a suitable gap */
1746 if (gap_start > high_limit)
1747 return -ENOMEM;
1748 if (gap_end >= low_limit && gap_end - gap_start >= length)
1749 goto found;
1750
1751 /* Visit right subtree if it looks promising */
1752 if (vma->vm_rb.rb_right) {
1753 struct vm_area_struct *right =
1754 rb_entry(vma->vm_rb.rb_right,
1755 struct vm_area_struct, vm_rb);
1756 if (right->rb_subtree_gap >= length) {
1757 vma = right;
1758 continue;
1759 }
1760 }
1761
1762 /* Go back up the rbtree to find next candidate node */
1763 while (true) {
1764 struct rb_node *prev = &vma->vm_rb;
1765 if (!rb_parent(prev))
1766 goto check_highest;
1767 vma = rb_entry(rb_parent(prev),
1768 struct vm_area_struct, vm_rb);
1769 if (prev == vma->vm_rb.rb_left) {
1770 gap_start = vma->vm_prev->vm_end;
1771 gap_end = vma->vm_start;
1772 goto check_current;
1773 }
1774 }
1775 }
1776
1777check_highest:
1778 /* Check highest gap, which does not precede any rbtree node */
1779 gap_start = mm->highest_vm_end;
1780 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1781 if (gap_start > high_limit)
1782 return -ENOMEM;
1783
1784found:
1785 /* We found a suitable gap. Clip it with the original low_limit. */
1786 if (gap_start < info->low_limit)
1787 gap_start = info->low_limit;
1788
1789 /* Adjust gap address to the desired alignment */
1790 gap_start += (info->align_offset - gap_start) & info->align_mask;
1791
1792 VM_BUG_ON(gap_start + info->length > info->high_limit);
1793 VM_BUG_ON(gap_start + info->length > gap_end);
1794 return gap_start;
1795}
1796
1797unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1798{
1799 struct mm_struct *mm = current->mm;
1800 struct vm_area_struct *vma;
1801 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1802
1803 /* Adjust search length to account for worst case alignment overhead */
1804 length = info->length + info->align_mask;
1805 if (length < info->length)
1806 return -ENOMEM;
1807
1808 /*
1809 * Adjust search limits by the desired length.
1810 * See implementation comment at top of unmapped_area().
1811 */
1812 gap_end = info->high_limit;
1813 if (gap_end < length)
1814 return -ENOMEM;
1815 high_limit = gap_end - length;
1816
1817 if (info->low_limit > high_limit)
1818 return -ENOMEM;
1819 low_limit = info->low_limit + length;
1820
1821 /* Check highest gap, which does not precede any rbtree node */
1822 gap_start = mm->highest_vm_end;
1823 if (gap_start <= high_limit)
1824 goto found_highest;
1825
1826 /* Check if rbtree root looks promising */
1827 if (RB_EMPTY_ROOT(&mm->mm_rb))
1828 return -ENOMEM;
1829 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1830 if (vma->rb_subtree_gap < length)
1831 return -ENOMEM;
1832
1833 while (true) {
1834 /* Visit right subtree if it looks promising */
1835 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1836 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1837 struct vm_area_struct *right =
1838 rb_entry(vma->vm_rb.rb_right,
1839 struct vm_area_struct, vm_rb);
1840 if (right->rb_subtree_gap >= length) {
1841 vma = right;
1842 continue;
1843 }
1844 }
1845
1846check_current:
1847 /* Check if current node has a suitable gap */
1848 gap_end = vma->vm_start;
1849 if (gap_end < low_limit)
1850 return -ENOMEM;
1851 if (gap_start <= high_limit && gap_end - gap_start >= length)
1852 goto found;
1853
1854 /* Visit left subtree if it looks promising */
1855 if (vma->vm_rb.rb_left) {
1856 struct vm_area_struct *left =
1857 rb_entry(vma->vm_rb.rb_left,
1858 struct vm_area_struct, vm_rb);
1859 if (left->rb_subtree_gap >= length) {
1860 vma = left;
1861 continue;
1862 }
1863 }
1864
1865 /* Go back up the rbtree to find next candidate node */
1866 while (true) {
1867 struct rb_node *prev = &vma->vm_rb;
1868 if (!rb_parent(prev))
1869 return -ENOMEM;
1870 vma = rb_entry(rb_parent(prev),
1871 struct vm_area_struct, vm_rb);
1872 if (prev == vma->vm_rb.rb_right) {
1873 gap_start = vma->vm_prev ?
1874 vma->vm_prev->vm_end : 0;
1875 goto check_current;
1876 }
1877 }
1878 }
1879
1880found:
1881 /* We found a suitable gap. Clip it with the original high_limit. */
1882 if (gap_end > info->high_limit)
1883 gap_end = info->high_limit;
1884
1885found_highest:
1886 /* Compute highest gap address at the desired alignment */
1887 gap_end -= info->length;
1888 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1889
1890 VM_BUG_ON(gap_end < info->low_limit);
1891 VM_BUG_ON(gap_end < gap_start);
1892 return gap_end;
1893}
1894
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895/* Get an address range which is currently unmapped.
1896 * For shmat() with addr=0.
1897 *
1898 * Ugly calling convention alert:
1899 * Return value with the low bits set means error value,
1900 * ie
1901 * if (ret & ~PAGE_MASK)
1902 * error = ret;
1903 *
1904 * This function "knows" that -ENOMEM has the bits set.
1905 */
1906#ifndef HAVE_ARCH_UNMAPPED_AREA
1907unsigned long
1908arch_get_unmapped_area(struct file *filp, unsigned long addr,
1909 unsigned long len, unsigned long pgoff, unsigned long flags)
1910{
1911 struct mm_struct *mm = current->mm;
1912 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001913 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001915 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 return -ENOMEM;
1917
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001918 if (flags & MAP_FIXED)
1919 return addr;
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 if (addr) {
1922 addr = PAGE_ALIGN(addr);
1923 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001924 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 (!vma || addr + len <= vma->vm_start))
1926 return addr;
1927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001929 info.flags = 0;
1930 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001931 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001932 info.high_limit = TASK_SIZE;
1933 info.align_mask = 0;
1934 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935}
vishnu.pscc71aba2014-10-09 15:26:29 -07001936#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938/*
1939 * This mmap-allocator allocates new areas top-down from below the
1940 * stack's low limit (the base):
1941 */
1942#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1943unsigned long
1944arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1945 const unsigned long len, const unsigned long pgoff,
1946 const unsigned long flags)
1947{
1948 struct vm_area_struct *vma;
1949 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001950 unsigned long addr = addr0;
1951 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001954 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return -ENOMEM;
1956
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001957 if (flags & MAP_FIXED)
1958 return addr;
1959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 /* requesting a specific address */
1961 if (addr) {
1962 addr = PAGE_ALIGN(addr);
1963 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001964 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 (!vma || addr + len <= vma->vm_start))
1966 return addr;
1967 }
1968
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001969 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1970 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001971 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001972 info.high_limit = mm->mmap_base;
1973 info.align_mask = 0;
1974 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001975
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 /*
1977 * A failed mmap() very likely causes application failure,
1978 * so fall back to the bottom-up function here. This scenario
1979 * can happen with large stack limits and large mmap()
1980 * allocations.
1981 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001982 if (addr & ~PAGE_MASK) {
1983 VM_BUG_ON(addr != -ENOMEM);
1984 info.flags = 0;
1985 info.low_limit = TASK_UNMAPPED_BASE;
1986 info.high_limit = TASK_SIZE;
1987 addr = vm_unmapped_area(&info);
1988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
1990 return addr;
1991}
1992#endif
1993
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994unsigned long
1995get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1996 unsigned long pgoff, unsigned long flags)
1997{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001998 unsigned long (*get_area)(struct file *, unsigned long,
1999 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
Al Viro9206de92009-12-03 15:23:11 -05002001 unsigned long error = arch_mmap_check(addr, len, flags);
2002 if (error)
2003 return error;
2004
2005 /* Careful about overflows.. */
2006 if (len > TASK_SIZE)
2007 return -ENOMEM;
2008
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002009 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04002010 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002011 get_area = file->f_op->get_unmapped_area;
2012 addr = get_area(file, addr, len, pgoff, flags);
2013 if (IS_ERR_VALUE(addr))
2014 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002016 if (addr > TASK_SIZE - len)
2017 return -ENOMEM;
2018 if (addr & ~PAGE_MASK)
2019 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002020
Al Viro9ac4ed42012-05-30 17:13:15 -04002021 addr = arch_rebalance_pgtables(addr, len);
2022 error = security_mmap_addr(addr);
2023 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024}
2025
2026EXPORT_SYMBOL(get_unmapped_area);
2027
2028/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002029struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002031 struct rb_node *rb_node;
2032 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002034 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002035 vma = vmacache_find(mm, addr);
2036 if (likely(vma))
2037 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002039 rb_node = mm->mm_rb.rb_node;
2040 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002042 while (rb_node) {
2043 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002045 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002046
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002047 if (tmp->vm_end > addr) {
2048 vma = tmp;
2049 if (tmp->vm_start <= addr)
2050 break;
2051 rb_node = rb_node->rb_left;
2052 } else
2053 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002055
2056 if (vma)
2057 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 return vma;
2059}
2060
2061EXPORT_SYMBOL(find_vma);
2062
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002063/*
2064 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002065 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066struct vm_area_struct *
2067find_vma_prev(struct mm_struct *mm, unsigned long addr,
2068 struct vm_area_struct **pprev)
2069{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002070 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002072 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002073 if (vma) {
2074 *pprev = vma->vm_prev;
2075 } else {
2076 struct rb_node *rb_node = mm->mm_rb.rb_node;
2077 *pprev = NULL;
2078 while (rb_node) {
2079 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2080 rb_node = rb_node->rb_right;
2081 }
2082 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002083 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084}
2085
2086/*
2087 * Verify that the stack growth is acceptable and
2088 * update accounting. This is shared with both the
2089 * grow-up and grow-down cases.
2090 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002091static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092{
2093 struct mm_struct *mm = vma->vm_mm;
2094 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08002095 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
2097 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002098 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 return -ENOMEM;
2100
2101 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08002102 actual_size = size;
2103 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2104 actual_size -= PAGE_SIZE;
2105 if (actual_size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return -ENOMEM;
2107
2108 /* mlock limit tests */
2109 if (vma->vm_flags & VM_LOCKED) {
2110 unsigned long locked;
2111 unsigned long limit;
2112 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002113 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2114 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 if (locked > limit && !capable(CAP_IPC_LOCK))
2116 return -ENOMEM;
2117 }
2118
Adam Litke0d59a012007-01-30 14:35:39 -08002119 /* Check to ensure the stack will not grow into a hugetlb-only region */
2120 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2121 vma->vm_end - size;
2122 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2123 return -EFAULT;
2124
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 /*
2126 * Overcommit.. This must be the final test, as it will
2127 * update security statistics.
2128 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002129 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 return -ENOMEM;
2131
2132 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 if (vma->vm_flags & VM_LOCKED)
2134 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002135 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 return 0;
2137}
2138
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002139#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002141 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2142 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002144int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145{
2146 int error;
2147
2148 if (!(vma->vm_flags & VM_GROWSUP))
2149 return -EFAULT;
2150
2151 /*
2152 * We must make sure the anon_vma is allocated
2153 * so that the anon_vma locking is not a noop.
2154 */
2155 if (unlikely(anon_vma_prepare(vma)))
2156 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002157 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
2159 /*
2160 * vma->vm_start/vm_end cannot change under us because the caller
2161 * is required to hold the mmap_sem in read mode. We need the
2162 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002163 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 */
Helge Deller06b32f32006-12-19 19:28:33 +01002165 if (address < PAGE_ALIGN(address+4))
2166 address = PAGE_ALIGN(address+4);
2167 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002168 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002169 return -ENOMEM;
2170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 error = 0;
2172
2173 /* Somebody else might have raced and expanded it already */
2174 if (address > vma->vm_end) {
2175 unsigned long size, grow;
2176
2177 size = address - vma->vm_start;
2178 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2179
Hugh Dickins42c36f62011-05-09 17:44:42 -07002180 error = -ENOMEM;
2181 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2182 error = acct_stack_growth(vma, size, grow);
2183 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002184 /*
2185 * vma_gap_update() doesn't support concurrent
2186 * updates, but we only hold a shared mmap_sem
2187 * lock here, so we need to protect against
2188 * concurrent vma expansions.
2189 * vma_lock_anon_vma() doesn't help here, as
2190 * we don't guarantee that all growable vmas
2191 * in a mm share the same root anon vma.
2192 * So, we reuse mm->page_table_lock to guard
2193 * against concurrent vma expansions.
2194 */
2195 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002196 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002197 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002198 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002199 if (vma->vm_next)
2200 vma_gap_update(vma->vm_next);
2201 else
2202 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002203 spin_unlock(&vma->vm_mm->page_table_lock);
2204
Hugh Dickins42c36f62011-05-09 17:44:42 -07002205 perf_event_mmap(vma);
2206 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002209 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002210 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002211 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 return error;
2213}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002214#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2215
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216/*
2217 * vma is the first one with address < vma->vm_start. Have to extend vma.
2218 */
Michal Hockod05f3162011-05-24 17:11:44 -07002219int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002220 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221{
2222 int error;
2223
2224 /*
2225 * We must make sure the anon_vma is allocated
2226 * so that the anon_vma locking is not a noop.
2227 */
2228 if (unlikely(anon_vma_prepare(vma)))
2229 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002230
2231 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002232 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002233 if (error)
2234 return error;
2235
Rik van Rielbb4a3402010-08-09 17:18:37 -07002236 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
2238 /*
2239 * vma->vm_start/vm_end cannot change under us because the caller
2240 * is required to hold the mmap_sem in read mode. We need the
2241 * anon_vma lock to serialize against concurrent expand_stacks.
2242 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
2244 /* Somebody else might have raced and expanded it already */
2245 if (address < vma->vm_start) {
2246 unsigned long size, grow;
2247
2248 size = vma->vm_end - address;
2249 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2250
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002251 error = -ENOMEM;
2252 if (grow <= vma->vm_pgoff) {
2253 error = acct_stack_growth(vma, size, grow);
2254 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002255 /*
2256 * vma_gap_update() doesn't support concurrent
2257 * updates, but we only hold a shared mmap_sem
2258 * lock here, so we need to protect against
2259 * concurrent vma expansions.
2260 * vma_lock_anon_vma() doesn't help here, as
2261 * we don't guarantee that all growable vmas
2262 * in a mm share the same root anon vma.
2263 * So, we reuse mm->page_table_lock to guard
2264 * against concurrent vma expansions.
2265 */
2266 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002267 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002268 vma->vm_start = address;
2269 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002270 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002271 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002272 spin_unlock(&vma->vm_mm->page_table_lock);
2273
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002274 perf_event_mmap(vma);
2275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 }
2277 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002278 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002279 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002280 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 return error;
2282}
2283
Linus Torvalds09884962013-02-27 08:36:04 -08002284/*
2285 * Note how expand_stack() refuses to expand the stack all the way to
2286 * abut the next virtual mapping, *unless* that mapping itself is also
2287 * a stack mapping. We want to leave room for a guard page, after all
2288 * (the guard page itself is not added here, that is done by the
2289 * actual page faulting logic)
2290 *
2291 * This matches the behavior of the guard page logic (see mm/memory.c:
2292 * check_stack_guard_page()), which only allows the guard page to be
2293 * removed under these circumstances.
2294 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002295#ifdef CONFIG_STACK_GROWSUP
2296int expand_stack(struct vm_area_struct *vma, unsigned long address)
2297{
Linus Torvalds09884962013-02-27 08:36:04 -08002298 struct vm_area_struct *next;
2299
2300 address &= PAGE_MASK;
2301 next = vma->vm_next;
2302 if (next && next->vm_start == address + PAGE_SIZE) {
2303 if (!(next->vm_flags & VM_GROWSUP))
2304 return -ENOMEM;
2305 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002306 return expand_upwards(vma, address);
2307}
2308
2309struct vm_area_struct *
2310find_extend_vma(struct mm_struct *mm, unsigned long addr)
2311{
2312 struct vm_area_struct *vma, *prev;
2313
2314 addr &= PAGE_MASK;
2315 vma = find_vma_prev(mm, addr, &prev);
2316 if (vma && (vma->vm_start <= addr))
2317 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002318 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002319 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002320 if (prev->vm_flags & VM_LOCKED)
2321 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002322 return prev;
2323}
2324#else
2325int expand_stack(struct vm_area_struct *vma, unsigned long address)
2326{
Linus Torvalds09884962013-02-27 08:36:04 -08002327 struct vm_area_struct *prev;
2328
2329 address &= PAGE_MASK;
2330 prev = vma->vm_prev;
2331 if (prev && prev->vm_end == address) {
2332 if (!(prev->vm_flags & VM_GROWSDOWN))
2333 return -ENOMEM;
2334 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002335 return expand_downwards(vma, address);
2336}
2337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002339find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
vishnu.pscc71aba2014-10-09 15:26:29 -07002341 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 unsigned long start;
2343
2344 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002345 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 if (!vma)
2347 return NULL;
2348 if (vma->vm_start <= addr)
2349 return vma;
2350 if (!(vma->vm_flags & VM_GROWSDOWN))
2351 return NULL;
2352 start = vma->vm_start;
2353 if (expand_stack(vma, addr))
2354 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002355 if (vma->vm_flags & VM_LOCKED)
2356 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 return vma;
2358}
2359#endif
2360
Jesse Barnese1d6d012014-12-12 16:55:27 -08002361EXPORT_SYMBOL_GPL(find_extend_vma);
2362
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002364 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002366 *
2367 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002369static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002371 unsigned long nr_accounted = 0;
2372
Hugh Dickins365e9c872005-10-29 18:16:18 -07002373 /* Update high watermark before we lower total_vm */
2374 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002376 long nrpages = vma_pages(vma);
2377
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002378 if (vma->vm_flags & VM_ACCOUNT)
2379 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002380 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002381 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002382 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002383 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 validate_mm(mm);
2385}
2386
2387/*
2388 * Get rid of page table information in the indicated region.
2389 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002390 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 */
2392static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002393 struct vm_area_struct *vma, struct vm_area_struct *prev,
2394 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395{
vishnu.pscc71aba2014-10-09 15:26:29 -07002396 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002397 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
2399 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002400 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002401 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002402 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002403 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002404 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002405 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406}
2407
2408/*
2409 * Create a list of vma's touched by the unmap, removing them from the mm's
2410 * vma list as we go..
2411 */
2412static void
2413detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2414 struct vm_area_struct *prev, unsigned long end)
2415{
2416 struct vm_area_struct **insertion_point;
2417 struct vm_area_struct *tail_vma = NULL;
2418
2419 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002420 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002422 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 mm->map_count--;
2424 tail_vma = vma;
2425 vma = vma->vm_next;
2426 } while (vma && vma->vm_start < end);
2427 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002428 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002429 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002430 vma_gap_update(vma);
2431 } else
2432 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002434
2435 /* Kill the cache */
2436 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437}
2438
2439/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002440 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2441 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002443static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 unsigned long addr, int new_below)
2445{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08002447 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448
Andi Kleena5516432008-07-23 21:27:41 -07002449 if (is_vm_hugetlb_page(vma) && (addr &
2450 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 return -EINVAL;
2452
Christoph Lametere94b1762006-12-06 20:33:17 -08002453 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002455 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456
2457 /* most fields are the same, copy all, and then fixup */
2458 *new = *vma;
2459
Rik van Riel5beb4932010-03-05 13:42:07 -08002460 INIT_LIST_HEAD(&new->anon_vma_chain);
2461
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 if (new_below)
2463 new->vm_end = addr;
2464 else {
2465 new->vm_start = addr;
2466 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2467 }
2468
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002469 err = vma_dup_policy(vma, new);
2470 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002471 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002473 err = anon_vma_clone(new, vma);
2474 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002475 goto out_free_mpol;
2476
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002477 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 get_file(new->vm_file);
2479
2480 if (new->vm_ops && new->vm_ops->open)
2481 new->vm_ops->open(new);
2482
2483 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002484 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2486 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002487 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Rik van Riel5beb4932010-03-05 13:42:07 -08002489 /* Success. */
2490 if (!err)
2491 return 0;
2492
2493 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002494 if (new->vm_ops && new->vm_ops->close)
2495 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002496 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002497 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002498 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002499 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002500 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002501 out_free_vma:
2502 kmem_cache_free(vm_area_cachep, new);
2503 out_err:
2504 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505}
2506
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002507/*
2508 * Split a vma into two pieces at address 'addr', a new vma is allocated
2509 * either for the first part or the tail.
2510 */
2511int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2512 unsigned long addr, int new_below)
2513{
2514 if (mm->map_count >= sysctl_max_map_count)
2515 return -ENOMEM;
2516
2517 return __split_vma(mm, vma, addr, new_below);
2518}
2519
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520/* Munmap is split into 2 main parts -- this part which finds
2521 * what needs doing, and the areas themselves, which do the
2522 * work. This now handles partial unmappings.
2523 * Jeremy Fitzhardinge <jeremy@goop.org>
2524 */
2525int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2526{
2527 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002528 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529
2530 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2531 return -EINVAL;
2532
vishnu.pscc71aba2014-10-09 15:26:29 -07002533 len = PAGE_ALIGN(len);
2534 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 return -EINVAL;
2536
2537 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002538 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002539 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002541 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002542 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
2544 /* if it doesn't overlap, we have nothing.. */
2545 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002546 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 return 0;
2548
2549 /*
2550 * If we need to split any vma, do it now to save pain later.
2551 *
2552 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2553 * unmapped vm_area_struct will remain in use: so lower split_vma
2554 * places tmp vma above, and higher split_vma places tmp vma below.
2555 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002556 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002557 int error;
2558
2559 /*
2560 * Make sure that map_count on return from munmap() will
2561 * not exceed its limit; but let map_count go just above
2562 * its limit temporarily, to help free resources as expected.
2563 */
2564 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2565 return -ENOMEM;
2566
2567 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (error)
2569 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002570 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 }
2572
2573 /* Does it split the last one? */
2574 last = find_vma(mm, end);
2575 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002576 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 if (error)
2578 return error;
2579 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002580 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
2582 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002583 * unlock any mlock()ed ranges before detaching vmas
2584 */
2585 if (mm->locked_vm) {
2586 struct vm_area_struct *tmp = vma;
2587 while (tmp && tmp->vm_start < end) {
2588 if (tmp->vm_flags & VM_LOCKED) {
2589 mm->locked_vm -= vma_pages(tmp);
2590 munlock_vma_pages_all(tmp);
2591 }
2592 tmp = tmp->vm_next;
2593 }
2594 }
2595
2596 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 * Remove the vma's, and unmap the actual pages
2598 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002599 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2600 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601
Dave Hansen1de4fa12014-11-14 07:18:31 -08002602 arch_unmap(mm, vma, start, end);
2603
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002605 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606
2607 return 0;
2608}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Al Virobfce2812012-04-20 21:57:04 -04002610int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002611{
2612 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002613 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002614
2615 down_write(&mm->mmap_sem);
2616 ret = do_munmap(mm, start, len);
2617 up_write(&mm->mmap_sem);
2618 return ret;
2619}
2620EXPORT_SYMBOL(vm_munmap);
2621
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002622SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002625 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626}
2627
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002628
2629/*
2630 * Emulation of deprecated remap_file_pages() syscall.
2631 */
2632SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2633 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2634{
2635
2636 struct mm_struct *mm = current->mm;
2637 struct vm_area_struct *vma;
2638 unsigned long populate = 0;
2639 unsigned long ret = -EINVAL;
2640 struct file *file;
2641
2642 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
2643 "See Documentation/vm/remap_file_pages.txt.\n",
2644 current->comm, current->pid);
2645
2646 if (prot)
2647 return ret;
2648 start = start & PAGE_MASK;
2649 size = size & PAGE_MASK;
2650
2651 if (start + size <= start)
2652 return ret;
2653
2654 /* Does pgoff wrap? */
2655 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2656 return ret;
2657
2658 down_write(&mm->mmap_sem);
2659 vma = find_vma(mm, start);
2660
2661 if (!vma || !(vma->vm_flags & VM_SHARED))
2662 goto out;
2663
2664 if (start < vma->vm_start || start + size > vma->vm_end)
2665 goto out;
2666
2667 if (pgoff == linear_page_index(vma, start)) {
2668 ret = 0;
2669 goto out;
2670 }
2671
2672 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2673 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2674 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2675
2676 flags &= MAP_NONBLOCK;
2677 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2678 if (vma->vm_flags & VM_LOCKED) {
2679 flags |= MAP_LOCKED;
2680 /* drop PG_Mlocked flag for over-mapped range */
2681 munlock_vma_pages_range(vma, start, start + size);
2682 }
2683
2684 file = get_file(vma->vm_file);
2685 ret = do_mmap_pgoff(vma->vm_file, start, size,
2686 prot, flags, pgoff, &populate);
2687 fput(file);
2688out:
2689 up_write(&mm->mmap_sem);
2690 if (populate)
2691 mm_populate(ret, populate);
2692 if (!IS_ERR_VALUE(ret))
2693 ret = 0;
2694 return ret;
2695}
2696
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697static inline void verify_mm_writelocked(struct mm_struct *mm)
2698{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002699#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2701 WARN_ON(1);
2702 up_read(&mm->mmap_sem);
2703 }
2704#endif
2705}
2706
2707/*
2708 * this is really a simplified "do_mmap". it only handles
2709 * anonymous maps. eventually we may be able to do some
2710 * brk-specific accounting here.
2711 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002712static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713{
vishnu.pscc71aba2014-10-09 15:26:29 -07002714 struct mm_struct *mm = current->mm;
2715 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002717 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002719 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
2721 len = PAGE_ALIGN(len);
2722 if (!len)
2723 return addr;
2724
Kirill Korotaev3a459752006-09-07 14:17:04 +04002725 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2726
Al Viro2c6a1012009-12-03 19:40:46 -05002727 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2728 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002729 return error;
2730
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002731 error = mlock_future_check(mm, mm->def_flags, len);
2732 if (error)
2733 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
2735 /*
2736 * mm->mmap_sem is required to protect against another thread
2737 * changing the mappings in case we sleep.
2738 */
2739 verify_mm_writelocked(mm);
2740
2741 /*
2742 * Clear old maps. this also does some error checking for us
2743 */
2744 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002745 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 if (do_munmap(mm, addr, len))
2747 return -ENOMEM;
2748 goto munmap_back;
2749 }
2750
2751 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002752 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 return -ENOMEM;
2754
2755 if (mm->map_count > sysctl_max_map_count)
2756 return -ENOMEM;
2757
Al Viro191c5422012-02-13 03:58:52 +00002758 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 return -ENOMEM;
2760
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002762 vma = vma_merge(mm, prev, addr, addr + len, flags,
2763 NULL, NULL, pgoff, NULL);
2764 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 goto out;
2766
2767 /*
2768 * create a vma struct for an anonymous mapping
2769 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002770 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 if (!vma) {
2772 vm_unacct_memory(len >> PAGE_SHIFT);
2773 return -ENOMEM;
2774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775
Rik van Riel5beb4932010-03-05 13:42:07 -08002776 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 vma->vm_mm = mm;
2778 vma->vm_start = addr;
2779 vma->vm_end = addr + len;
2780 vma->vm_pgoff = pgoff;
2781 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002782 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 vma_link(mm, vma, prev, rb_link, rb_parent);
2784out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002785 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002787 if (flags & VM_LOCKED)
2788 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002789 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 return addr;
2791}
2792
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002793unsigned long vm_brk(unsigned long addr, unsigned long len)
2794{
2795 struct mm_struct *mm = current->mm;
2796 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002797 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002798
2799 down_write(&mm->mmap_sem);
2800 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002801 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002802 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002803 if (populate)
2804 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002805 return ret;
2806}
2807EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
2809/* Release all mmaps. */
2810void exit_mmap(struct mm_struct *mm)
2811{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002812 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002813 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 unsigned long nr_accounted = 0;
2815
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002816 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002817 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002818
Rik van Rielba470de2008-10-18 20:26:50 -07002819 if (mm->locked_vm) {
2820 vma = mm->mmap;
2821 while (vma) {
2822 if (vma->vm_flags & VM_LOCKED)
2823 munlock_vma_pages_all(vma);
2824 vma = vma->vm_next;
2825 }
2826 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002827
2828 arch_exit_mmap(mm);
2829
Rik van Rielba470de2008-10-18 20:26:50 -07002830 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002831 if (!vma) /* Can happen if dup_mmap() received an OOM */
2832 return;
2833
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002836 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002837 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002838 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002839 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002840
Hugh Dickins6ee86302013-04-29 15:07:44 -07002841 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002842 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002845 * Walk the list again, actually closing and freeing it,
2846 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002848 while (vma) {
2849 if (vma->vm_flags & VM_ACCOUNT)
2850 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002851 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002852 }
2853 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854}
2855
2856/* Insert vm structure into process list sorted by address
2857 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002858 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002860int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861{
Hugh Dickins6597d782012-10-08 16:29:07 -07002862 struct vm_area_struct *prev;
2863 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
2865 /*
2866 * The vm_pgoff of a purely anonymous vma should be irrelevant
2867 * until its first write fault, when page's anon_vma and index
2868 * are set. But now set the vm_pgoff it will almost certainly
2869 * end up with (unless mremap moves it elsewhere before that
2870 * first wfault), so /proc/pid/maps tells a consistent story.
2871 *
2872 * By setting it to reflect the virtual start address of the
2873 * vma, merges and splits can happen in a seamless way, just
2874 * using the existing file pgoff checks and manipulations.
2875 * Similarly in do_mmap_pgoff and in do_brk.
2876 */
2877 if (!vma->vm_file) {
2878 BUG_ON(vma->anon_vma);
2879 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2880 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002881 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2882 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002884 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002885 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002886 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302887
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 vma_link(mm, vma, prev, rb_link, rb_parent);
2889 return 0;
2890}
2891
2892/*
2893 * Copy the vma structure to a new location in the same mm,
2894 * prior to moving page table entries, to effect an mremap move.
2895 */
2896struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002897 unsigned long addr, unsigned long len, pgoff_t pgoff,
2898 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899{
2900 struct vm_area_struct *vma = *vmap;
2901 unsigned long vma_start = vma->vm_start;
2902 struct mm_struct *mm = vma->vm_mm;
2903 struct vm_area_struct *new_vma, *prev;
2904 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002905 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
2907 /*
2908 * If anonymous vma has not yet been faulted, update new pgoff
2909 * to match new location, to increase its chance of merging.
2910 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002911 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002913 faulted_in_anon_vma = false;
2914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915
Hugh Dickins6597d782012-10-08 16:29:07 -07002916 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2917 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2919 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2920 if (new_vma) {
2921 /*
2922 * Source vma may have been merged into new_vma
2923 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002924 if (unlikely(vma_start >= new_vma->vm_start &&
2925 vma_start < new_vma->vm_end)) {
2926 /*
2927 * The only way we can get a vma_merge with
2928 * self during an mremap is if the vma hasn't
2929 * been faulted in yet and we were allowed to
2930 * reset the dst vma->vm_pgoff to the
2931 * destination address of the mremap to allow
2932 * the merge to happen. mremap must change the
2933 * vm_pgoff linearity between src and dst vmas
2934 * (in turn preventing a vma_merge) to be
2935 * safe. It is only safe to keep the vm_pgoff
2936 * linear if there are no pages mapped yet.
2937 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002938 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002939 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002940 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002941 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002943 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 if (new_vma) {
2945 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 new_vma->vm_start = addr;
2947 new_vma->vm_end = addr + len;
2948 new_vma->vm_pgoff = pgoff;
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002949 if (vma_dup_policy(vma, new_vma))
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002950 goto out_free_vma;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002951 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2952 if (anon_vma_clone(new_vma, vma))
2953 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002954 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 get_file(new_vma->vm_file);
2956 if (new_vma->vm_ops && new_vma->vm_ops->open)
2957 new_vma->vm_ops->open(new_vma);
2958 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002959 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 }
2961 }
2962 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002963
2964 out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002965 mpol_put(vma_policy(new_vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002966 out_free_vma:
2967 kmem_cache_free(vm_area_cachep, new_vma);
2968 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002970
2971/*
2972 * Return true if the calling process may expand its vm space by the passed
2973 * number of pages
2974 */
2975int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2976{
2977 unsigned long cur = mm->total_vm; /* pages */
2978 unsigned long lim;
2979
Jiri Slaby59e99e52010-03-05 13:41:44 -08002980 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002981
2982 if (cur + npages > lim)
2983 return 0;
2984 return 1;
2985}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002986
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002987static int special_mapping_fault(struct vm_area_struct *vma,
2988 struct vm_fault *vmf);
2989
2990/*
2991 * Having a close hook prevents vma merging regardless of flags.
2992 */
2993static void special_mapping_close(struct vm_area_struct *vma)
2994{
2995}
2996
2997static const char *special_mapping_name(struct vm_area_struct *vma)
2998{
2999 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3000}
3001
3002static const struct vm_operations_struct special_mapping_vmops = {
3003 .close = special_mapping_close,
3004 .fault = special_mapping_fault,
3005 .name = special_mapping_name,
3006};
3007
3008static const struct vm_operations_struct legacy_special_mapping_vmops = {
3009 .close = special_mapping_close,
3010 .fault = special_mapping_fault,
3011};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003012
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003013static int special_mapping_fault(struct vm_area_struct *vma,
3014 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003015{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003016 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003017 struct page **pages;
3018
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003019 /*
3020 * special mappings have no vm_file, and in that case, the mm
3021 * uses vm_pgoff internally. So we have to subtract it from here.
3022 * We are allowed to do this because we are the mm; do not copy
3023 * this code into drivers!
3024 */
3025 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003026
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003027 if (vma->vm_ops == &legacy_special_mapping_vmops)
3028 pages = vma->vm_private_data;
3029 else
3030 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
3031 pages;
3032
3033 for (; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003034 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003035
3036 if (*pages) {
3037 struct page *page = *pages;
3038 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003039 vmf->page = page;
3040 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003041 }
3042
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003043 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003044}
3045
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003046static struct vm_area_struct *__install_special_mapping(
3047 struct mm_struct *mm,
3048 unsigned long addr, unsigned long len,
3049 unsigned long vm_flags, const struct vm_operations_struct *ops,
3050 void *priv)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003051{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003052 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003053 struct vm_area_struct *vma;
3054
3055 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3056 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003057 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003058
Rik van Riel5beb4932010-03-05 13:42:07 -08003059 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003060 vma->vm_mm = mm;
3061 vma->vm_start = addr;
3062 vma->vm_end = addr + len;
3063
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003064 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003065 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003066
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003067 vma->vm_ops = ops;
3068 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003069
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003070 ret = insert_vm_struct(mm, vma);
3071 if (ret)
3072 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003073
3074 mm->total_vm += len >> PAGE_SHIFT;
3075
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003076 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003077
Stefani Seibold3935ed62014-03-17 23:22:02 +01003078 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003079
3080out:
3081 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003082 return ERR_PTR(ret);
3083}
3084
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003085/*
3086 * Called with mm->mmap_sem held for writing.
3087 * Insert a new vma covering the given region, with the given flags.
3088 * Its pages are supplied by the given array of struct page *.
3089 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3090 * The region past the last page supplied will always produce SIGBUS.
3091 * The array pointer and the pages it points to are assumed to stay alive
3092 * for as long as this mapping might exist.
3093 */
3094struct vm_area_struct *_install_special_mapping(
3095 struct mm_struct *mm,
3096 unsigned long addr, unsigned long len,
3097 unsigned long vm_flags, const struct vm_special_mapping *spec)
3098{
3099 return __install_special_mapping(mm, addr, len, vm_flags,
3100 &special_mapping_vmops, (void *)spec);
3101}
3102
Stefani Seibold3935ed62014-03-17 23:22:02 +01003103int install_special_mapping(struct mm_struct *mm,
3104 unsigned long addr, unsigned long len,
3105 unsigned long vm_flags, struct page **pages)
3106{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003107 struct vm_area_struct *vma = __install_special_mapping(
3108 mm, addr, len, vm_flags, &legacy_special_mapping_vmops,
3109 (void *)pages);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003110
Duan Jiong14bd5b42014-06-04 16:07:05 -07003111 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003112}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003113
3114static DEFINE_MUTEX(mm_all_locks_mutex);
3115
Peter Zijlstra454ed842008-08-11 09:30:25 +02003116static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003117{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003118 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003119 /*
3120 * The LSB of head.next can't change from under us
3121 * because we hold the mm_all_locks_mutex.
3122 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003123 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003124 /*
3125 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003126 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003127 * the same anon_vma we won't take it again.
3128 *
3129 * No need of atomic instructions here, head.next
3130 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003131 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003132 */
3133 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003134 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003135 BUG();
3136 }
3137}
3138
Peter Zijlstra454ed842008-08-11 09:30:25 +02003139static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003140{
3141 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3142 /*
3143 * AS_MM_ALL_LOCKS can't change from under us because
3144 * we hold the mm_all_locks_mutex.
3145 *
3146 * Operations on ->flags have to be atomic because
3147 * even if AS_MM_ALL_LOCKS is stable thanks to the
3148 * mm_all_locks_mutex, there may be other cpus
3149 * changing other bitflags in parallel to us.
3150 */
3151 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3152 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003153 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003154 }
3155}
3156
3157/*
3158 * This operation locks against the VM for all pte/vma/mm related
3159 * operations that could ever happen on a certain mm. This includes
3160 * vmtruncate, try_to_unmap, and all page faults.
3161 *
3162 * The caller must take the mmap_sem in write mode before calling
3163 * mm_take_all_locks(). The caller isn't allowed to release the
3164 * mmap_sem until mm_drop_all_locks() returns.
3165 *
3166 * mmap_sem in write mode is required in order to block all operations
3167 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003168 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003169 * anon_vmas to be associated with existing vmas.
3170 *
3171 * A single task can't take more than one mm_take_all_locks() in a row
3172 * or it would deadlock.
3173 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003174 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003175 * mapping->flags avoid to take the same lock twice, if more than one
3176 * vma in this mm is backed by the same anon_vma or address_space.
3177 *
3178 * We can take all the locks in random order because the VM code
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003179 * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003180 * takes more than one of them in a row. Secondly we're protected
3181 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3182 *
3183 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3184 * that may have to take thousand of locks.
3185 *
3186 * mm_take_all_locks() can fail if it's interrupted by signals.
3187 */
3188int mm_take_all_locks(struct mm_struct *mm)
3189{
3190 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003191 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003192
3193 BUG_ON(down_read_trylock(&mm->mmap_sem));
3194
3195 mutex_lock(&mm_all_locks_mutex);
3196
3197 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3198 if (signal_pending(current))
3199 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003200 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003201 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003202 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003203
3204 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3205 if (signal_pending(current))
3206 goto out_unlock;
3207 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003208 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3209 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003210 }
3211
Kautuk Consul584cff52011-10-31 17:08:59 -07003212 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003213
3214out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003215 mm_drop_all_locks(mm);
3216 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003217}
3218
3219static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3220{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003221 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003222 /*
3223 * The LSB of head.next can't change to 0 from under
3224 * us because we hold the mm_all_locks_mutex.
3225 *
3226 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003227 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003228 * never see our bitflag.
3229 *
3230 * No need of atomic instructions here, head.next
3231 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003232 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003233 */
3234 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003235 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003236 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003237 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003238 }
3239}
3240
3241static void vm_unlock_mapping(struct address_space *mapping)
3242{
3243 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3244 /*
3245 * AS_MM_ALL_LOCKS can't change to 0 from under us
3246 * because we hold the mm_all_locks_mutex.
3247 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003248 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003249 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3250 &mapping->flags))
3251 BUG();
3252 }
3253}
3254
3255/*
3256 * The mmap_sem cannot be released by the caller until
3257 * mm_drop_all_locks() returns.
3258 */
3259void mm_drop_all_locks(struct mm_struct *mm)
3260{
3261 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003262 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003263
3264 BUG_ON(down_read_trylock(&mm->mmap_sem));
3265 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3266
3267 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3268 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003269 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3270 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003271 if (vma->vm_file && vma->vm_file->f_mapping)
3272 vm_unlock_mapping(vma->vm_file->f_mapping);
3273 }
3274
3275 mutex_unlock(&mm_all_locks_mutex);
3276}
David Howells8feae132009-01-08 12:04:47 +00003277
3278/*
3279 * initialise the VMA slab
3280 */
3281void __init mmap_init(void)
3282{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003283 int ret;
3284
Tejun Heo908c7f12014-09-08 09:51:29 +09003285 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003286 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003287}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003288
3289/*
3290 * Initialise sysctl_user_reserve_kbytes.
3291 *
3292 * This is intended to prevent a user from starting a single memory hogging
3293 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3294 * mode.
3295 *
3296 * The default value is min(3% of free memory, 128MB)
3297 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3298 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003299static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003300{
3301 unsigned long free_kbytes;
3302
3303 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3304
3305 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3306 return 0;
3307}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003308subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003309
3310/*
3311 * Initialise sysctl_admin_reserve_kbytes.
3312 *
3313 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3314 * to log in and kill a memory hogging process.
3315 *
3316 * Systems with more than 256MB will reserve 8MB, enough to recover
3317 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3318 * only reserve 3% of free pages by default.
3319 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003320static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003321{
3322 unsigned long free_kbytes;
3323
3324 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3325
3326 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3327 return 0;
3328}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003329subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003330
3331/*
3332 * Reinititalise user and admin reserves if memory is added or removed.
3333 *
3334 * The default user reserve max is 128MB, and the default max for the
3335 * admin reserve is 8MB. These are usually, but not always, enough to
3336 * enable recovery from a memory hogging process using login/sshd, a shell,
3337 * and tools like top. It may make sense to increase or even disable the
3338 * reserve depending on the existence of swap or variations in the recovery
3339 * tools. So, the admin may have changed them.
3340 *
3341 * If memory is added and the reserves have been eliminated or increased above
3342 * the default max, then we'll trust the admin.
3343 *
3344 * If memory is removed and there isn't enough free memory, then we
3345 * need to reset the reserves.
3346 *
3347 * Otherwise keep the reserve set by the admin.
3348 */
3349static int reserve_mem_notifier(struct notifier_block *nb,
3350 unsigned long action, void *data)
3351{
3352 unsigned long tmp, free_kbytes;
3353
3354 switch (action) {
3355 case MEM_ONLINE:
3356 /* Default max is 128MB. Leave alone if modified by operator. */
3357 tmp = sysctl_user_reserve_kbytes;
3358 if (0 < tmp && tmp < (1UL << 17))
3359 init_user_reserve();
3360
3361 /* Default max is 8MB. Leave alone if modified by operator. */
3362 tmp = sysctl_admin_reserve_kbytes;
3363 if (0 < tmp && tmp < (1UL << 13))
3364 init_admin_reserve();
3365
3366 break;
3367 case MEM_OFFLINE:
3368 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3369
3370 if (sysctl_user_reserve_kbytes > free_kbytes) {
3371 init_user_reserve();
3372 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3373 sysctl_user_reserve_kbytes);
3374 }
3375
3376 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3377 init_admin_reserve();
3378 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3379 sysctl_admin_reserve_kbytes);
3380 }
3381 break;
3382 default:
3383 break;
3384 }
3385 return NOTIFY_OK;
3386}
3387
3388static struct notifier_block reserve_mem_nb = {
3389 .notifier_call = reserve_mem_notifier,
3390};
3391
3392static int __meminit init_reserve_notifier(void)
3393{
3394 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003395 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003396
3397 return 0;
3398}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003399subsys_initcall(init_reserve_notifier);