blob: 76d1ec29149bf25f5e6ffe3e56e86467efccc641 [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>
Andrea Arcangeli19a809a2015-09-04 15:46:24 -070044#include <linux/userfaultfd_k.h>
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080045#include <linux/moduleparam.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47#include <asm/uaccess.h>
48#include <asm/cacheflush.h>
49#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020050#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Jan Beulich42b77722008-07-23 21:27:10 -070052#include "internal.h"
53
Kirill Korotaev3a459752006-09-07 14:17:04 +040054#ifndef arch_mmap_check
55#define arch_mmap_check(addr, len, flags) (0)
56#endif
57
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080058#ifndef arch_rebalance_pgtables
59#define arch_rebalance_pgtables(addr, len) (addr)
60#endif
61
Daniel Cashmand07e2252016-01-14 15:19:53 -080062#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
63const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
64const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
65int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
66#endif
67#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
68const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
69const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
70int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
71#endif
72
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080073static bool ignore_rlimit_data = true;
74core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
Daniel Cashmand07e2252016-01-14 15:19:53 -080075
Hugh Dickinse0da3822005-04-19 13:29:15 -070076static void unmap_region(struct mm_struct *mm,
77 struct vm_area_struct *vma, struct vm_area_struct *prev,
78 unsigned long start, unsigned long end);
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080/* description of effects of mapping type and prot in current implementation.
81 * this is due to the limited x86 page protection hardware. The expected
82 * behavior is in parens:
83 *
84 * map_type prot
85 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
86 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 * w: (no) no w: (no) no w: (yes) yes w: (no) no
88 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070089 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
91 * w: (no) no w: (no) no w: (copy) copy w: (no) no
92 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
93 *
94 */
95pgprot_t protection_map[16] = {
96 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
97 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
98};
99
Hugh Dickins804af2c2006-07-26 21:39:49 +0100100pgprot_t vm_get_page_prot(unsigned long vm_flags)
101{
Dave Kleikampb845f312008-07-08 00:28:51 +1000102 return __pgprot(pgprot_val(protection_map[vm_flags &
103 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
104 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100105}
106EXPORT_SYMBOL(vm_get_page_prot);
107
Peter Feiner64e45502014-10-13 15:55:46 -0700108static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
109{
110 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
111}
112
113/* Update vma->vm_page_prot to reflect vma->vm_flags. */
114void vma_set_page_prot(struct vm_area_struct *vma)
115{
116 unsigned long vm_flags = vma->vm_flags;
117
118 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
119 if (vma_wants_writenotify(vma)) {
120 vm_flags &= ~VM_SHARED;
121 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
122 vm_flags);
123 }
124}
125
126
Shaohua Li34679d72011-05-24 17:11:18 -0700127int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
128int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800129unsigned long sysctl_overcommit_kbytes __read_mostly;
Christoph Lameterc3d8c142005-09-06 15:16:33 -0700130int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700131unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700132unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -0700133/*
134 * Make sure vm_committed_as in one cacheline and not cacheline shared with
135 * other variables. It can be updated by several CPUs frequently.
136 */
137struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
139/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -0800140 * The global memory commitment made in the system can be a metric
141 * that can be used to drive ballooning decisions when Linux is hosted
142 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
143 * balancing memory across competing virtual machines that are hosted.
144 * Several metrics drive this policy engine including the guest reported
145 * memory commitment.
146 */
147unsigned long vm_memory_committed(void)
148{
149 return percpu_counter_read_positive(&vm_committed_as);
150}
151EXPORT_SYMBOL_GPL(vm_memory_committed);
152
153/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 * Check that a process has enough memory to allocate a new virtual
155 * mapping. 0 means there is enough memory for the allocation to
156 * succeed and -ENOMEM implies there is not.
157 *
158 * We currently support three overcommit policies, which are set via the
159 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
160 *
161 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
162 * Additional code 2002 Jul 20 by Robert Love.
163 *
164 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
165 *
166 * Note this is a helper function intended to be used by LSMs which
167 * wish to use this logic.
168 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700169int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Roman Gushchin5703b082015-02-11 15:28:39 -0800171 long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -0700173 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
174 -(s64)vm_committed_as_batch * num_online_cpus(),
175 "memory commitment underflow");
176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 vm_acct_memory(pages);
178
179 /*
180 * Sometimes we want to use more memory than we have
181 */
182 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
183 return 0;
184
185 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700186 free = global_page_state(NR_FREE_PAGES);
187 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Dmitry Finkc15bef32011-07-25 17:12:19 -0700189 /*
190 * shmem pages shouldn't be counted as free in this
191 * case, they can't be purged, only swapped out, and
192 * that won't affect the overall amount of available
193 * memory in the system.
194 */
195 free -= global_page_state(NR_SHMEM);
196
Shaohua Liec8acf22013-02-22 16:34:38 -0800197 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
199 /*
200 * Any slabs which are created with the
201 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
202 * which are reclaimable, under pressure. The dentry
203 * cache and most inode caches should fall into this
204 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700205 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700208 * Leave reserved pages. The pages are not for anonymous pages.
209 */
210 if (free <= totalreserve_pages)
211 goto error;
212 else
213 free -= totalreserve_pages;
214
215 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700216 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 */
218 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700219 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
221 if (free > pages)
222 return 0;
223
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700224 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 }
226
Jerome Marchand00619bc2013-11-12 15:08:31 -0800227 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700229 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 */
231 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700232 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700234 /*
235 * Don't let a single process grow so big a user can't recover
236 */
237 if (mm) {
238 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
Roman Gushchin5703b082015-02-11 15:28:39 -0800239 allowed -= min_t(long, mm->total_vm / 32, reserve);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700240 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700242 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700244error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 vm_unacct_memory(pages);
246
247 return -ENOMEM;
248}
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800251 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 */
253static void __remove_shared_vm_struct(struct vm_area_struct *vma,
254 struct file *file, struct address_space *mapping)
255{
256 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500257 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700259 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800262 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 flush_dcache_mmap_unlock(mapping);
264}
265
266/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700267 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700268 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700270void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
272 struct file *file = vma->vm_file;
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 if (file) {
275 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800276 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800278 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700280}
281
282/*
283 * Close a vm structure and free it, returning the next.
284 */
285static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
286{
287 struct vm_area_struct *next = vma->vm_next;
288
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700289 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 if (vma->vm_ops && vma->vm_ops->close)
291 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700292 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700293 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700294 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700296 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700299static unsigned long do_brk(unsigned long addr, unsigned long len);
300
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100301SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700303 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 unsigned long newbrk, oldbrk;
305 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700306 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800307 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
309 down_write(&mm->mmap_sem);
310
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700311#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800312 /*
313 * CONFIG_COMPAT_BRK can still be overridden by setting
314 * randomize_va_space to 2, which will still cause mm->start_brk
315 * to be arbitrarily shifted
316 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700317 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800318 min_brk = mm->start_brk;
319 else
320 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700321#else
322 min_brk = mm->start_brk;
323#endif
324 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700326
327 /*
328 * Check against rlimit here. If this check is done later after the test
329 * of oldbrk with newbrk then it can escape the test and let the data
330 * segment grow beyond its set limit the in case where the limit is
331 * not page aligned -Ram Gupta
332 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700333 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
334 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700335 goto out;
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 newbrk = PAGE_ALIGN(brk);
338 oldbrk = PAGE_ALIGN(mm->brk);
339 if (oldbrk == newbrk)
340 goto set_brk;
341
342 /* Always allow shrinking brk. */
343 if (brk <= mm->brk) {
344 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
345 goto set_brk;
346 goto out;
347 }
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 /* Check against existing mmap mappings. */
350 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
351 goto out;
352
353 /* Ok, looks good - let it rip. */
354 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
355 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357set_brk:
358 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800359 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
360 up_write(&mm->mmap_sem);
361 if (populate)
362 mm_populate(oldbrk, newbrk - oldbrk);
363 return brk;
364
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365out:
366 retval = mm->brk;
367 up_write(&mm->mmap_sem);
368 return retval;
369}
370
Michel Lespinassed3737182012-12-11 16:01:38 -0800371static long vma_compute_subtree_gap(struct vm_area_struct *vma)
372{
373 unsigned long max, subtree_gap;
374 max = vma->vm_start;
375 if (vma->vm_prev)
376 max -= vma->vm_prev->vm_end;
377 if (vma->vm_rb.rb_left) {
378 subtree_gap = rb_entry(vma->vm_rb.rb_left,
379 struct vm_area_struct, vm_rb)->rb_subtree_gap;
380 if (subtree_gap > max)
381 max = subtree_gap;
382 }
383 if (vma->vm_rb.rb_right) {
384 subtree_gap = rb_entry(vma->vm_rb.rb_right,
385 struct vm_area_struct, vm_rb)->rb_subtree_gap;
386 if (subtree_gap > max)
387 max = subtree_gap;
388 }
389 return max;
390}
391
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700392#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800393static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800395 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800396 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 struct rb_node *nd, *pn = NULL;
398 unsigned long prev = 0, pend = 0;
399
400 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
401 struct vm_area_struct *vma;
402 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800403 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700404 pr_emerg("vm_start %lx < prev %lx\n",
405 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800406 bug = 1;
407 }
408 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700409 pr_emerg("vm_start %lx < pend %lx\n",
410 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800411 bug = 1;
412 }
413 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700414 pr_emerg("vm_start %lx > vm_end %lx\n",
415 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800416 bug = 1;
417 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800418 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800419 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700420 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800421 vma->rb_subtree_gap,
422 vma_compute_subtree_gap(vma));
423 bug = 1;
424 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800425 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 i++;
427 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800428 prev = vma->vm_start;
429 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 }
431 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800432 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800434 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700435 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800436 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800438 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
440
Michel Lespinassed3737182012-12-11 16:01:38 -0800441static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
442{
443 struct rb_node *nd;
444
445 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
446 struct vm_area_struct *vma;
447 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700448 VM_BUG_ON_VMA(vma != ignore &&
449 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
450 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700454static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
456 int bug = 0;
457 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800458 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700459 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700460
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700461 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800462 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700463 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700464
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800465 if (anon_vma) {
466 anon_vma_lock_read(anon_vma);
467 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
468 anon_vma_interval_tree_verify(avc);
469 anon_vma_unlock_read(anon_vma);
470 }
471
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800472 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700473 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 i++;
475 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800476 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700477 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800478 bug = 1;
479 }
480 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700481 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700482 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800483 bug = 1;
484 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800485 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800486 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700487 if (i != -1)
488 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800489 bug = 1;
490 }
Sasha Levin96dad672014-10-09 15:28:39 -0700491 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492}
493#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800494#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495#define validate_mm(mm) do { } while (0)
496#endif
497
Michel Lespinassed3737182012-12-11 16:01:38 -0800498RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
499 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
500
501/*
502 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
503 * vma->vm_prev->vm_end values changed, without modifying the vma's position
504 * in the rbtree.
505 */
506static void vma_gap_update(struct vm_area_struct *vma)
507{
508 /*
509 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
510 * function that does exacltly what we want.
511 */
512 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
513}
514
515static inline void vma_rb_insert(struct vm_area_struct *vma,
516 struct rb_root *root)
517{
518 /* All rb_subtree_gap values must be consistent prior to insertion */
519 validate_mm_rb(root, NULL);
520
521 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
522}
523
524static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
525{
526 /*
527 * All rb_subtree_gap values must be consistent prior to erase,
528 * with the possible exception of the vma being erased.
529 */
530 validate_mm_rb(root, vma);
531
532 /*
533 * Note rb_erase_augmented is a fairly large inline function,
534 * so make sure we instantiate it only once with our desired
535 * augmented rbtree callbacks.
536 */
537 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
538}
539
Michel Lespinassebf181b92012-10-08 16:31:39 -0700540/*
541 * vma has some anon_vma assigned, and is already inserted on that
542 * anon_vma's interval trees.
543 *
544 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
545 * vma must be removed from the anon_vma's interval trees using
546 * anon_vma_interval_tree_pre_update_vma().
547 *
548 * After the update, the vma will be reinserted using
549 * anon_vma_interval_tree_post_update_vma().
550 *
551 * The entire update must be protected by exclusive mmap_sem and by
552 * the root anon_vma's mutex.
553 */
554static inline void
555anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
556{
557 struct anon_vma_chain *avc;
558
559 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
560 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
561}
562
563static inline void
564anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
565{
566 struct anon_vma_chain *avc;
567
568 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
569 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
570}
571
Hugh Dickins6597d782012-10-08 16:29:07 -0700572static int find_vma_links(struct mm_struct *mm, unsigned long addr,
573 unsigned long end, struct vm_area_struct **pprev,
574 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575{
Hugh Dickins6597d782012-10-08 16:29:07 -0700576 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 __rb_link = &mm->mm_rb.rb_node;
579 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 while (*__rb_link) {
582 struct vm_area_struct *vma_tmp;
583
584 __rb_parent = *__rb_link;
585 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
586
587 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700588 /* Fail if an existing vma overlaps the area */
589 if (vma_tmp->vm_start < end)
590 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 __rb_link = &__rb_parent->rb_left;
592 } else {
593 rb_prev = __rb_parent;
594 __rb_link = &__rb_parent->rb_right;
595 }
596 }
597
598 *pprev = NULL;
599 if (rb_prev)
600 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
601 *rb_link = __rb_link;
602 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700603 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
Cyril Hrubise8420a82013-04-29 15:08:33 -0700606static unsigned long count_vma_pages_range(struct mm_struct *mm,
607 unsigned long addr, unsigned long end)
608{
609 unsigned long nr_pages = 0;
610 struct vm_area_struct *vma;
611
612 /* Find first overlaping mapping */
613 vma = find_vma_intersection(mm, addr, end);
614 if (!vma)
615 return 0;
616
617 nr_pages = (min(end, vma->vm_end) -
618 max(addr, vma->vm_start)) >> PAGE_SHIFT;
619
620 /* Iterate over the rest of the overlaps */
621 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
622 unsigned long overlap_len;
623
624 if (vma->vm_start > end)
625 break;
626
627 overlap_len = min(end, vma->vm_end) - vma->vm_start;
628 nr_pages += overlap_len >> PAGE_SHIFT;
629 }
630
631 return nr_pages;
632}
633
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
635 struct rb_node **rb_link, struct rb_node *rb_parent)
636{
Michel Lespinassed3737182012-12-11 16:01:38 -0800637 /* Update tracking information for the gap following the new vma. */
638 if (vma->vm_next)
639 vma_gap_update(vma->vm_next);
640 else
641 mm->highest_vm_end = vma->vm_end;
642
643 /*
644 * vma->vm_prev wasn't known when we followed the rbtree to find the
645 * correct insertion point for that vma. As a result, we could not
646 * update the vma vm_rb parents rb_subtree_gap values on the way down.
647 * So, we first insert the vma with a zero rb_subtree_gap value
648 * (to be consistent with what we did on the way down), and then
649 * immediately update the gap to the correct value. Finally we
650 * rebalance the rbtree after all augmented values have been set.
651 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800653 vma->rb_subtree_gap = 0;
654 vma_gap_update(vma);
655 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656}
657
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700658static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
ZhenwenXu48aae422009-01-06 14:40:21 -0800660 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
662 file = vma->vm_file;
663 if (file) {
664 struct address_space *mapping = file->f_mapping;
665
666 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500667 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700669 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
671 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800672 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 flush_dcache_mmap_unlock(mapping);
674 }
675}
676
677static void
678__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
679 struct vm_area_struct *prev, struct rb_node **rb_link,
680 struct rb_node *rb_parent)
681{
682 __vma_link_list(mm, vma, prev, rb_parent);
683 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
686static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
687 struct vm_area_struct *prev, struct rb_node **rb_link,
688 struct rb_node *rb_parent)
689{
690 struct address_space *mapping = NULL;
691
Huang Shijie64ac4942014-06-04 16:07:33 -0700692 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800694 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
697 __vma_link(mm, vma, prev, rb_link, rb_parent);
698 __vma_link_file(vma);
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800701 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
703 mm->map_count++;
704 validate_mm(mm);
705}
706
707/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700708 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700709 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800711static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Hugh Dickins6597d782012-10-08 16:29:07 -0700713 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800714 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Hugh Dickins6597d782012-10-08 16:29:07 -0700716 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
717 &prev, &rb_link, &rb_parent))
718 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 __vma_link(mm, vma, prev, rb_link, rb_parent);
720 mm->map_count++;
721}
722
723static inline void
724__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
725 struct vm_area_struct *prev)
726{
Michel Lespinassed3737182012-12-11 16:01:38 -0800727 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700728
Michel Lespinassed3737182012-12-11 16:01:38 -0800729 vma_rb_erase(vma, &mm->mm_rb);
730 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700731 if (next)
732 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700733
734 /* Kill the cache */
735 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736}
737
738/*
739 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
740 * is already present in an i_mmap tree without adjusting the tree.
741 * The following helper function should be used when such adjustments
742 * are necessary. The "insert" vma (if any) is to be inserted
743 * before we drop the necessary locks.
744 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800745int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
747{
748 struct mm_struct *mm = vma->vm_mm;
749 struct vm_area_struct *next = vma->vm_next;
750 struct vm_area_struct *importer = NULL;
751 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700752 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700753 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800755 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 long adjust_next = 0;
757 int remove_next = 0;
758
759 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700760 struct vm_area_struct *exporter = NULL;
761
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 if (end >= next->vm_end) {
763 /*
764 * vma expands, overlapping all the next, and
765 * perhaps the one after too (mprotect case 6).
766 */
767again: remove_next = 1 + (end > next->vm_end);
768 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700769 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 importer = vma;
771 } else if (end > next->vm_start) {
772 /*
773 * vma expands, overlapping part of the next:
774 * mprotect case 5 shifting the boundary up.
775 */
776 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700777 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 importer = vma;
779 } else if (end < vma->vm_end) {
780 /*
781 * vma shrinks, and !insert tells it's not
782 * split_vma inserting another: so it must be
783 * mprotect case 4 shifting the boundary down.
784 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700785 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700786 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 importer = next;
788 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Rik van Riel5beb4932010-03-05 13:42:07 -0800790 /*
791 * Easily overlooked: when mprotect shifts the boundary,
792 * make sure the expanding vma has anon_vma set if the
793 * shrinking vma had, to cover any anon pages imported.
794 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700795 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800796 int error;
797
Linus Torvalds287d97a2010-04-10 15:22:30 -0700798 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300799 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700800 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300801 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800802 }
803 }
804
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 if (file) {
806 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800807 root = &mapping->i_mmap;
808 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530809
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800810 if (adjust_next)
811 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530812
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800813 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700816 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 * are visible to arm/parisc __flush_dcache_page
818 * throughout; but we cannot insert into address
819 * space until vma start or end is updated.
820 */
821 __vma_link_file(insert);
822 }
823 }
824
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800825 vma_adjust_trans_huge(vma, start, end, adjust_next);
826
Michel Lespinassebf181b92012-10-08 16:31:39 -0700827 anon_vma = vma->anon_vma;
828 if (!anon_vma && adjust_next)
829 anon_vma = next->anon_vma;
830 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700831 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
832 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000833 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700834 anon_vma_interval_tree_pre_update_vma(vma);
835 if (adjust_next)
836 anon_vma_interval_tree_pre_update_vma(next);
837 }
Rik van Riel012f18002010-08-09 17:18:40 -0700838
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (root) {
840 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700841 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700843 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 }
845
Michel Lespinassed3737182012-12-11 16:01:38 -0800846 if (start != vma->vm_start) {
847 vma->vm_start = start;
848 start_changed = true;
849 }
850 if (end != vma->vm_end) {
851 vma->vm_end = end;
852 end_changed = true;
853 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 vma->vm_pgoff = pgoff;
855 if (adjust_next) {
856 next->vm_start += adjust_next << PAGE_SHIFT;
857 next->vm_pgoff += adjust_next;
858 }
859
860 if (root) {
861 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700862 vma_interval_tree_insert(next, root);
863 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 flush_dcache_mmap_unlock(mapping);
865 }
866
867 if (remove_next) {
868 /*
869 * vma_merge has merged next into vma, and needs
870 * us to remove next before dropping the locks.
871 */
872 __vma_unlink(mm, next, vma);
873 if (file)
874 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 } else if (insert) {
876 /*
877 * split_vma has split insert from vma, and needs
878 * us to insert it before dropping the locks
879 * (it may either follow vma or precede it).
880 */
881 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800882 } else {
883 if (start_changed)
884 vma_gap_update(vma);
885 if (end_changed) {
886 if (!next)
887 mm->highest_vm_end = end;
888 else if (!adjust_next)
889 vma_gap_update(next);
890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 }
892
Michel Lespinassebf181b92012-10-08 16:31:39 -0700893 if (anon_vma) {
894 anon_vma_interval_tree_post_update_vma(vma);
895 if (adjust_next)
896 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800897 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700898 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800900 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530902 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100903 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530904
905 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100906 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530907 }
908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700910 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530911 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700913 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800914 if (next->anon_vma)
915 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700917 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 kmem_cache_free(vm_area_cachep, next);
919 /*
920 * In mprotect's case 6 (see comments on vma_merge),
921 * we must remove another next too. It would clutter
922 * up the code too much to do both in one go.
923 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800924 next = vma->vm_next;
925 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800927 else if (next)
928 vma_gap_update(next);
929 else
930 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530932 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100933 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800936
937 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938}
939
940/*
941 * If the vma has a ->close operation then the driver probably needs to release
942 * per-vma resources, so we don't attempt to merge those.
943 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700945 struct file *file, unsigned long vm_flags,
946 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800948 /*
949 * VM_SOFTDIRTY should not prevent from VMA merging, if we
950 * match the flags but dirty bit -- the caller should mark
951 * merged VMA as dirty. If dirty bit won't be excluded from
952 * comparison, we increase pressue on the memory system forcing
953 * the kernel to generate new VMAs when old one could be
954 * extended instead.
955 */
956 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 return 0;
958 if (vma->vm_file != file)
959 return 0;
960 if (vma->vm_ops && vma->vm_ops->close)
961 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700962 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
963 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 return 1;
965}
966
967static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700968 struct anon_vma *anon_vma2,
969 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970{
Shaohua Li965f55d2011-05-24 17:11:20 -0700971 /*
972 * The list_is_singular() test is to avoid merging VMA cloned from
973 * parents. This can improve scalability caused by anon_vma lock.
974 */
975 if ((!anon_vma1 || !anon_vma2) && (!vma ||
976 list_is_singular(&vma->anon_vma_chain)))
977 return 1;
978 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979}
980
981/*
982 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
983 * in front of (at a lower virtual address and file offset than) the vma.
984 *
985 * We cannot merge two vmas if they have differently assigned (non-NULL)
986 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
987 *
988 * We don't check here for the merged mmap wrapping around the end of pagecache
989 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
990 * wrap, nor mmaps which cover the final page at index -1UL.
991 */
992static int
993can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700994 struct anon_vma *anon_vma, struct file *file,
995 pgoff_t vm_pgoff,
996 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700998 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700999 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (vma->vm_pgoff == vm_pgoff)
1001 return 1;
1002 }
1003 return 0;
1004}
1005
1006/*
1007 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1008 * beyond (at a higher virtual address and file offset than) the vma.
1009 *
1010 * We cannot merge two vmas if they have differently assigned (non-NULL)
1011 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1012 */
1013static int
1014can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001015 struct anon_vma *anon_vma, struct file *file,
1016 pgoff_t vm_pgoff,
1017 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001019 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001020 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001022 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1024 return 1;
1025 }
1026 return 0;
1027}
1028
1029/*
1030 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1031 * whether that can be merged with its predecessor or its successor.
1032 * Or both (it neatly fills a hole).
1033 *
1034 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1035 * certain not to be mapped by the time vma_merge is called; but when
1036 * called for mprotect, it is certain to be already mapped (either at
1037 * an offset within prev, or at the start of next), and the flags of
1038 * this area are about to be changed to vm_flags - and the no-change
1039 * case has already been eliminated.
1040 *
1041 * The following mprotect cases have to be considered, where AAAA is
1042 * the area passed down from mprotect_fixup, never extending beyond one
1043 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1044 *
1045 * AAAA AAAA AAAA AAAA
1046 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1047 * cannot merge might become might become might become
1048 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1049 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1050 * mremap move: PPPPNNNNNNNN 8
1051 * AAAA
1052 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1053 * might become case 1 below case 2 below case 3 below
1054 *
1055 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1056 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1057 */
1058struct vm_area_struct *vma_merge(struct mm_struct *mm,
1059 struct vm_area_struct *prev, unsigned long addr,
1060 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001061 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001062 pgoff_t pgoff, struct mempolicy *policy,
1063 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
1065 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1066 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001067 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069 /*
1070 * We later require that vma->vm_flags == vm_flags,
1071 * so this tests vma->vm_flags & VM_SPECIAL, too.
1072 */
1073 if (vm_flags & VM_SPECIAL)
1074 return NULL;
1075
1076 if (prev)
1077 next = prev->vm_next;
1078 else
1079 next = mm->mmap;
1080 area = next;
1081 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1082 next = next->vm_next;
1083
1084 /*
1085 * Can it merge with the predecessor?
1086 */
1087 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001088 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001090 anon_vma, file, pgoff,
1091 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 /*
1093 * OK, it can. Can we now merge in the successor as well?
1094 */
1095 if (next && end == next->vm_start &&
1096 mpol_equal(policy, vma_policy(next)) &&
1097 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001098 anon_vma, file,
1099 pgoff+pglen,
1100 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001102 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001104 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 next->vm_end, prev->vm_pgoff, NULL);
1106 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001107 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001109 if (err)
1110 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001111 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 return prev;
1113 }
1114
1115 /*
1116 * Can this new request be merged in front of next?
1117 */
1118 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001119 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001121 anon_vma, file, pgoff+pglen,
1122 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001124 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 addr, prev->vm_pgoff, NULL);
1126 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001127 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001129 if (err)
1130 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001131 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 return area;
1133 }
1134
1135 return NULL;
1136}
1137
1138/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001139 * Rough compatbility check to quickly see if it's even worth looking
1140 * at sharing an anon_vma.
1141 *
1142 * They need to have the same vm_file, and the flags can only differ
1143 * in things that mprotect may change.
1144 *
1145 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1146 * we can merge the two vma's. For example, we refuse to merge a vma if
1147 * there is a vm_ops->close() function, because that indicates that the
1148 * driver is doing some kind of reference counting. But that doesn't
1149 * really matter for the anon_vma sharing case.
1150 */
1151static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1152{
1153 return a->vm_end == b->vm_start &&
1154 mpol_equal(vma_policy(a), vma_policy(b)) &&
1155 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001156 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001157 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1158}
1159
1160/*
1161 * Do some basic sanity checking to see if we can re-use the anon_vma
1162 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1163 * the same as 'old', the other will be the new one that is trying
1164 * to share the anon_vma.
1165 *
1166 * NOTE! This runs with mm_sem held for reading, so it is possible that
1167 * the anon_vma of 'old' is concurrently in the process of being set up
1168 * by another page fault trying to merge _that_. But that's ok: if it
1169 * is being set up, that automatically means that it will be a singleton
1170 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001171 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001172 *
1173 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1174 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1175 * is to return an anon_vma that is "complex" due to having gone through
1176 * a fork).
1177 *
1178 * We also make sure that the two vma's are compatible (adjacent,
1179 * and with the same memory policies). That's all stable, even with just
1180 * a read lock on the mm_sem.
1181 */
1182static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1183{
1184 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001185 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001186
1187 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1188 return anon_vma;
1189 }
1190 return NULL;
1191}
1192
1193/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1195 * neighbouring vmas for a suitable anon_vma, before it goes off
1196 * to allocate a new anon_vma. It checks because a repetitive
1197 * sequence of mprotects and faults may otherwise lead to distinct
1198 * anon_vmas being allocated, preventing vma merge in subsequent
1199 * mprotect.
1200 */
1201struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1202{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001203 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
1206 near = vma->vm_next;
1207 if (!near)
1208 goto try_prev;
1209
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001210 anon_vma = reusable_anon_vma(near, vma, near);
1211 if (anon_vma)
1212 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001214 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (!near)
1216 goto none;
1217
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001218 anon_vma = reusable_anon_vma(near, near, vma);
1219 if (anon_vma)
1220 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221none:
1222 /*
1223 * There's no absolute need to look only at touching neighbours:
1224 * we could search further afield for "compatible" anon_vmas.
1225 * But it would probably just be a waste of time searching,
1226 * or lead to too many vmas hanging off the same anon_vma.
1227 * We're trying to allow mprotect remerging later on,
1228 * not trying to minimize memory used for anon_vmas.
1229 */
1230 return NULL;
1231}
1232
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233/*
Al Viro40401532012-02-13 03:58:52 +00001234 * If a hint addr is less than mmap_min_addr change hint to be as
1235 * low as possible but still greater than mmap_min_addr
1236 */
1237static inline unsigned long round_hint_to_min(unsigned long hint)
1238{
1239 hint &= PAGE_MASK;
1240 if (((void *)hint != NULL) &&
1241 (hint < mmap_min_addr))
1242 return PAGE_ALIGN(mmap_min_addr);
1243 return hint;
1244}
1245
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001246static inline int mlock_future_check(struct mm_struct *mm,
1247 unsigned long flags,
1248 unsigned long len)
1249{
1250 unsigned long locked, lock_limit;
1251
1252 /* mlock MCL_FUTURE? */
1253 if (flags & VM_LOCKED) {
1254 locked = len >> PAGE_SHIFT;
1255 locked += mm->locked_vm;
1256 lock_limit = rlimit(RLIMIT_MEMLOCK);
1257 lock_limit >>= PAGE_SHIFT;
1258 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1259 return -EAGAIN;
1260 }
1261 return 0;
1262}
1263
Al Viro40401532012-02-13 03:58:52 +00001264/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001265 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001267unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001269 unsigned long flags, vm_flags_t vm_flags,
1270 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271{
vishnu.pscc71aba2014-10-09 15:26:29 -07001272 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Michel Lespinasse41badc12013-02-22 16:32:47 -08001274 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001275
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001276 if (!len)
1277 return -EINVAL;
1278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 /*
1280 * Does the application expect PROT_READ to imply PROT_EXEC?
1281 *
1282 * (the exception is when the underlying filesystem is noexec
1283 * mounted, in which case we dont add PROT_EXEC.)
1284 */
1285 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001286 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 prot |= PROT_EXEC;
1288
Eric Paris7cd94142007-11-26 18:47:40 -05001289 if (!(flags & MAP_FIXED))
1290 addr = round_hint_to_min(addr);
1291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 /* Careful about overflows.. */
1293 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001294 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 return -ENOMEM;
1296
1297 /* offset overflow? */
1298 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001299 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
1301 /* Too many mappings? */
1302 if (mm->map_count > sysctl_max_map_count)
1303 return -ENOMEM;
1304
1305 /* Obtain the address to map to. we verify (or select) it and ensure
1306 * that it represents a valid section of the address space.
1307 */
1308 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001309 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 return addr;
1311
1312 /* Do simple checking here so the lower-level routines won't have
1313 * to. we assume access permissions have been handled by the open
1314 * of the memory object, so we don't do any here.
1315 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001316 vm_flags |= calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1318
Huang Shijiecdf7b342009-09-21 17:03:36 -07001319 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 if (!can_do_mlock())
1321 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001322
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001323 if (mlock_future_check(mm, vm_flags, len))
1324 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001327 struct inode *inode = file_inode(file);
1328
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 switch (flags & MAP_TYPE) {
1330 case MAP_SHARED:
1331 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1332 return -EACCES;
1333
1334 /*
1335 * Make sure we don't allow writing to an append-only
1336 * file..
1337 */
1338 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1339 return -EACCES;
1340
1341 /*
1342 * Make sure there are no mandatory locks on the file.
1343 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001344 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 return -EAGAIN;
1346
1347 vm_flags |= VM_SHARED | VM_MAYSHARE;
1348 if (!(file->f_mode & FMODE_WRITE))
1349 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1350
1351 /* fall through */
1352 case MAP_PRIVATE:
1353 if (!(file->f_mode & FMODE_READ))
1354 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001355 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001356 if (vm_flags & VM_EXEC)
1357 return -EPERM;
1358 vm_flags &= ~VM_MAYEXEC;
1359 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001360
Al Viro72c2d532013-09-22 16:27:52 -04001361 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001362 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001363 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1364 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 break;
1366
1367 default:
1368 return -EINVAL;
1369 }
1370 } else {
1371 switch (flags & MAP_TYPE) {
1372 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001373 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1374 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001375 /*
1376 * Ignore pgoff.
1377 */
1378 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 vm_flags |= VM_SHARED | VM_MAYSHARE;
1380 break;
1381 case MAP_PRIVATE:
1382 /*
1383 * Set pgoff according to addr for anon_vma.
1384 */
1385 pgoff = addr >> PAGE_SHIFT;
1386 break;
1387 default:
1388 return -EINVAL;
1389 }
1390 }
1391
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001392 /*
1393 * Set 'VM_NORESERVE' if we should not account for the
1394 * memory use of this mapping.
1395 */
1396 if (flags & MAP_NORESERVE) {
1397 /* We honor MAP_NORESERVE if allowed to overcommit */
1398 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1399 vm_flags |= VM_NORESERVE;
1400
1401 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1402 if (file && is_file_hugepages(file))
1403 vm_flags |= VM_NORESERVE;
1404 }
1405
1406 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001407 if (!IS_ERR_VALUE(addr) &&
1408 ((vm_flags & VM_LOCKED) ||
1409 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001410 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001411 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001412}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001413
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001414SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1415 unsigned long, prot, unsigned long, flags,
1416 unsigned long, fd, unsigned long, pgoff)
1417{
1418 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001419 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001420
1421 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001422 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001423 file = fget(fd);
1424 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001425 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001426 if (is_file_hugepages(file))
1427 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001428 retval = -EINVAL;
1429 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1430 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001431 } else if (flags & MAP_HUGETLB) {
1432 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001433 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001434
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001435 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001436 if (!hs)
1437 return -EINVAL;
1438
1439 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001440 /*
1441 * VM_NORESERVE is used because the reservations will be
1442 * taken when vm_ops->mmap() is called
1443 * A dummy user value is used because we are not locking
1444 * memory so no accounting is necessary
1445 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001446 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001447 VM_NORESERVE,
1448 &user, HUGETLB_ANONHUGE_INODE,
1449 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001450 if (IS_ERR(file))
1451 return PTR_ERR(file);
1452 }
1453
1454 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1455
Al Viroeb36c582012-05-30 20:17:35 -04001456 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001457out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001458 if (file)
1459 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001460 return retval;
1461}
1462
Christoph Hellwiga4679372010-03-10 15:21:15 -08001463#ifdef __ARCH_WANT_SYS_OLD_MMAP
1464struct mmap_arg_struct {
1465 unsigned long addr;
1466 unsigned long len;
1467 unsigned long prot;
1468 unsigned long flags;
1469 unsigned long fd;
1470 unsigned long offset;
1471};
1472
1473SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1474{
1475 struct mmap_arg_struct a;
1476
1477 if (copy_from_user(&a, arg, sizeof(a)))
1478 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001479 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001480 return -EINVAL;
1481
1482 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1483 a.offset >> PAGE_SHIFT);
1484}
1485#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1486
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001487/*
1488 * Some shared mappigns will want the pages marked read-only
1489 * to track write events. If so, we'll downgrade vm_page_prot
1490 * to the private version (using protection_map[] without the
1491 * VM_SHARED bit).
1492 */
1493int vma_wants_writenotify(struct vm_area_struct *vma)
1494{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001495 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001496 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001497
1498 /* If it was private or non-writable, the write bit is already clear */
1499 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1500 return 0;
1501
1502 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001503 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001504 return 1;
1505
Peter Feiner64e45502014-10-13 15:55:46 -07001506 /* The open routine did something to the protections that pgprot_modify
1507 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001508 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001509 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001510 return 0;
1511
Peter Feiner64e45502014-10-13 15:55:46 -07001512 /* Do we need to track softdirty? */
1513 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1514 return 1;
1515
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001516 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001517 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001518 return 0;
1519
1520 /* Can the mapping track the dirty pages? */
1521 return vma->vm_file && vma->vm_file->f_mapping &&
1522 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1523}
1524
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001525/*
1526 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001527 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001528 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001529static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001530{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001531 /*
1532 * hugetlb has its own accounting separate from the core VM
1533 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1534 */
1535 if (file && is_file_hugepages(file))
1536 return 0;
1537
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001538 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1539}
1540
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001541unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001542 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001543{
1544 struct mm_struct *mm = current->mm;
1545 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001546 int error;
1547 struct rb_node **rb_link, *rb_parent;
1548 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001549
Cyril Hrubise8420a82013-04-29 15:08:33 -07001550 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001551 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001552 unsigned long nr_pages;
1553
1554 /*
1555 * MAP_FIXED may remove pages of mappings that intersects with
1556 * requested mapping. Account for the pages it would unmap.
1557 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001558 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1559
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001560 if (!may_expand_vm(mm, vm_flags,
1561 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001562 return -ENOMEM;
1563 }
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001566 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1567 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 if (do_munmap(mm, addr, len))
1569 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
1571
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001572 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001573 * Private writable mapping: check memory availability
1574 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001575 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001576 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001577 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001578 return -ENOMEM;
1579 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581
1582 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001583 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001585 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1586 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001587 if (vma)
1588 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
1590 /*
1591 * Determine the object being mapped and call the appropriate
1592 * specific mapper. the address has already been validated, but
1593 * not unmapped, but the maps are removed from the list.
1594 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001595 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 if (!vma) {
1597 error = -ENOMEM;
1598 goto unacct_error;
1599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 vma->vm_mm = mm;
1602 vma->vm_start = addr;
1603 vma->vm_end = addr + len;
1604 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001605 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001607 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
1609 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 if (vm_flags & VM_DENYWRITE) {
1611 error = deny_write_access(file);
1612 if (error)
1613 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001615 if (vm_flags & VM_SHARED) {
1616 error = mapping_map_writable(file->f_mapping);
1617 if (error)
1618 goto allow_write_and_free_vma;
1619 }
1620
1621 /* ->mmap() can change vma->vm_file, but must guarantee that
1622 * vma_link() below can deny write-access if VM_DENYWRITE is set
1623 * and map writably if VM_SHARED is set. This usually means the
1624 * new file must not have been exposed to user-space, yet.
1625 */
Al Virocb0942b2012-08-27 14:48:26 -04001626 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 error = file->f_op->mmap(file, vma);
1628 if (error)
1629 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001630
1631 /* Can addr have changed??
1632 *
1633 * Answer: Yes, several device drivers can do it in their
1634 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001635 * Bug: If addr is changed, prev, rb_link, rb_parent should
1636 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001637 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001638 WARN_ON_ONCE(addr != vma->vm_start);
1639
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001640 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001641 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 } else if (vm_flags & VM_SHARED) {
1643 error = shmem_zero_setup(vma);
1644 if (error)
1645 goto free_vma;
1646 }
1647
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001648 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001649 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001650 if (file) {
1651 if (vm_flags & VM_SHARED)
1652 mapping_unmap_writable(file->f_mapping);
1653 if (vm_flags & VM_DENYWRITE)
1654 allow_write_access(file);
1655 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001656 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001657out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001658 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001659
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001660 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001662 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1663 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001664 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001665 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001666 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001667 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301668
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001669 if (file)
1670 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301671
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001672 /*
1673 * New (or expanded) vma always get soft dirty status.
1674 * Otherwise user-space soft-dirty page tracker won't
1675 * be able to distinguish situation when vma area unmapped,
1676 * then new mapped in-place (which must be aimed as
1677 * a completely new data area).
1678 */
1679 vma->vm_flags |= VM_SOFTDIRTY;
1680
Peter Feiner64e45502014-10-13 15:55:46 -07001681 vma_set_page_prot(vma);
1682
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 return addr;
1684
1685unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 vma->vm_file = NULL;
1687 fput(file);
1688
1689 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001690 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1691 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001692 if (vm_flags & VM_SHARED)
1693 mapping_unmap_writable(file->f_mapping);
1694allow_write_and_free_vma:
1695 if (vm_flags & VM_DENYWRITE)
1696 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697free_vma:
1698 kmem_cache_free(vm_area_cachep, vma);
1699unacct_error:
1700 if (charged)
1701 vm_unacct_memory(charged);
1702 return error;
1703}
1704
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001705unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1706{
1707 /*
1708 * We implement the search by looking for an rbtree node that
1709 * immediately follows a suitable gap. That is,
1710 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1711 * - gap_end = vma->vm_start >= info->low_limit + length;
1712 * - gap_end - gap_start >= length
1713 */
1714
1715 struct mm_struct *mm = current->mm;
1716 struct vm_area_struct *vma;
1717 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1718
1719 /* Adjust search length to account for worst case alignment overhead */
1720 length = info->length + info->align_mask;
1721 if (length < info->length)
1722 return -ENOMEM;
1723
1724 /* Adjust search limits by the desired length */
1725 if (info->high_limit < length)
1726 return -ENOMEM;
1727 high_limit = info->high_limit - length;
1728
1729 if (info->low_limit > high_limit)
1730 return -ENOMEM;
1731 low_limit = info->low_limit + length;
1732
1733 /* Check if rbtree root looks promising */
1734 if (RB_EMPTY_ROOT(&mm->mm_rb))
1735 goto check_highest;
1736 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1737 if (vma->rb_subtree_gap < length)
1738 goto check_highest;
1739
1740 while (true) {
1741 /* Visit left subtree if it looks promising */
1742 gap_end = vma->vm_start;
1743 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1744 struct vm_area_struct *left =
1745 rb_entry(vma->vm_rb.rb_left,
1746 struct vm_area_struct, vm_rb);
1747 if (left->rb_subtree_gap >= length) {
1748 vma = left;
1749 continue;
1750 }
1751 }
1752
1753 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1754check_current:
1755 /* Check if current node has a suitable gap */
1756 if (gap_start > high_limit)
1757 return -ENOMEM;
1758 if (gap_end >= low_limit && gap_end - gap_start >= length)
1759 goto found;
1760
1761 /* Visit right subtree if it looks promising */
1762 if (vma->vm_rb.rb_right) {
1763 struct vm_area_struct *right =
1764 rb_entry(vma->vm_rb.rb_right,
1765 struct vm_area_struct, vm_rb);
1766 if (right->rb_subtree_gap >= length) {
1767 vma = right;
1768 continue;
1769 }
1770 }
1771
1772 /* Go back up the rbtree to find next candidate node */
1773 while (true) {
1774 struct rb_node *prev = &vma->vm_rb;
1775 if (!rb_parent(prev))
1776 goto check_highest;
1777 vma = rb_entry(rb_parent(prev),
1778 struct vm_area_struct, vm_rb);
1779 if (prev == vma->vm_rb.rb_left) {
1780 gap_start = vma->vm_prev->vm_end;
1781 gap_end = vma->vm_start;
1782 goto check_current;
1783 }
1784 }
1785 }
1786
1787check_highest:
1788 /* Check highest gap, which does not precede any rbtree node */
1789 gap_start = mm->highest_vm_end;
1790 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1791 if (gap_start > high_limit)
1792 return -ENOMEM;
1793
1794found:
1795 /* We found a suitable gap. Clip it with the original low_limit. */
1796 if (gap_start < info->low_limit)
1797 gap_start = info->low_limit;
1798
1799 /* Adjust gap address to the desired alignment */
1800 gap_start += (info->align_offset - gap_start) & info->align_mask;
1801
1802 VM_BUG_ON(gap_start + info->length > info->high_limit);
1803 VM_BUG_ON(gap_start + info->length > gap_end);
1804 return gap_start;
1805}
1806
1807unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1808{
1809 struct mm_struct *mm = current->mm;
1810 struct vm_area_struct *vma;
1811 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1812
1813 /* Adjust search length to account for worst case alignment overhead */
1814 length = info->length + info->align_mask;
1815 if (length < info->length)
1816 return -ENOMEM;
1817
1818 /*
1819 * Adjust search limits by the desired length.
1820 * See implementation comment at top of unmapped_area().
1821 */
1822 gap_end = info->high_limit;
1823 if (gap_end < length)
1824 return -ENOMEM;
1825 high_limit = gap_end - length;
1826
1827 if (info->low_limit > high_limit)
1828 return -ENOMEM;
1829 low_limit = info->low_limit + length;
1830
1831 /* Check highest gap, which does not precede any rbtree node */
1832 gap_start = mm->highest_vm_end;
1833 if (gap_start <= high_limit)
1834 goto found_highest;
1835
1836 /* Check if rbtree root looks promising */
1837 if (RB_EMPTY_ROOT(&mm->mm_rb))
1838 return -ENOMEM;
1839 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1840 if (vma->rb_subtree_gap < length)
1841 return -ENOMEM;
1842
1843 while (true) {
1844 /* Visit right subtree if it looks promising */
1845 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1846 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1847 struct vm_area_struct *right =
1848 rb_entry(vma->vm_rb.rb_right,
1849 struct vm_area_struct, vm_rb);
1850 if (right->rb_subtree_gap >= length) {
1851 vma = right;
1852 continue;
1853 }
1854 }
1855
1856check_current:
1857 /* Check if current node has a suitable gap */
1858 gap_end = vma->vm_start;
1859 if (gap_end < low_limit)
1860 return -ENOMEM;
1861 if (gap_start <= high_limit && gap_end - gap_start >= length)
1862 goto found;
1863
1864 /* Visit left subtree if it looks promising */
1865 if (vma->vm_rb.rb_left) {
1866 struct vm_area_struct *left =
1867 rb_entry(vma->vm_rb.rb_left,
1868 struct vm_area_struct, vm_rb);
1869 if (left->rb_subtree_gap >= length) {
1870 vma = left;
1871 continue;
1872 }
1873 }
1874
1875 /* Go back up the rbtree to find next candidate node */
1876 while (true) {
1877 struct rb_node *prev = &vma->vm_rb;
1878 if (!rb_parent(prev))
1879 return -ENOMEM;
1880 vma = rb_entry(rb_parent(prev),
1881 struct vm_area_struct, vm_rb);
1882 if (prev == vma->vm_rb.rb_right) {
1883 gap_start = vma->vm_prev ?
1884 vma->vm_prev->vm_end : 0;
1885 goto check_current;
1886 }
1887 }
1888 }
1889
1890found:
1891 /* We found a suitable gap. Clip it with the original high_limit. */
1892 if (gap_end > info->high_limit)
1893 gap_end = info->high_limit;
1894
1895found_highest:
1896 /* Compute highest gap address at the desired alignment */
1897 gap_end -= info->length;
1898 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1899
1900 VM_BUG_ON(gap_end < info->low_limit);
1901 VM_BUG_ON(gap_end < gap_start);
1902 return gap_end;
1903}
1904
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905/* Get an address range which is currently unmapped.
1906 * For shmat() with addr=0.
1907 *
1908 * Ugly calling convention alert:
1909 * Return value with the low bits set means error value,
1910 * ie
1911 * if (ret & ~PAGE_MASK)
1912 * error = ret;
1913 *
1914 * This function "knows" that -ENOMEM has the bits set.
1915 */
1916#ifndef HAVE_ARCH_UNMAPPED_AREA
1917unsigned long
1918arch_get_unmapped_area(struct file *filp, unsigned long addr,
1919 unsigned long len, unsigned long pgoff, unsigned long flags)
1920{
1921 struct mm_struct *mm = current->mm;
1922 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001923 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001925 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 return -ENOMEM;
1927
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001928 if (flags & MAP_FIXED)
1929 return addr;
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 if (addr) {
1932 addr = PAGE_ALIGN(addr);
1933 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001934 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 (!vma || addr + len <= vma->vm_start))
1936 return addr;
1937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001939 info.flags = 0;
1940 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001941 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001942 info.high_limit = TASK_SIZE;
1943 info.align_mask = 0;
1944 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945}
vishnu.pscc71aba2014-10-09 15:26:29 -07001946#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948/*
1949 * This mmap-allocator allocates new areas top-down from below the
1950 * stack's low limit (the base):
1951 */
1952#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1953unsigned long
1954arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1955 const unsigned long len, const unsigned long pgoff,
1956 const unsigned long flags)
1957{
1958 struct vm_area_struct *vma;
1959 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001960 unsigned long addr = addr0;
1961 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001964 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return -ENOMEM;
1966
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001967 if (flags & MAP_FIXED)
1968 return addr;
1969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 /* requesting a specific address */
1971 if (addr) {
1972 addr = PAGE_ALIGN(addr);
1973 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001974 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 (!vma || addr + len <= vma->vm_start))
1976 return addr;
1977 }
1978
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001979 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1980 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001981 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001982 info.high_limit = mm->mmap_base;
1983 info.align_mask = 0;
1984 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 /*
1987 * A failed mmap() very likely causes application failure,
1988 * so fall back to the bottom-up function here. This scenario
1989 * can happen with large stack limits and large mmap()
1990 * allocations.
1991 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001992 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001993 VM_BUG_ON(addr != -ENOMEM);
1994 info.flags = 0;
1995 info.low_limit = TASK_UNMAPPED_BASE;
1996 info.high_limit = TASK_SIZE;
1997 addr = vm_unmapped_area(&info);
1998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 return addr;
2001}
2002#endif
2003
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004unsigned long
2005get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2006 unsigned long pgoff, unsigned long flags)
2007{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002008 unsigned long (*get_area)(struct file *, unsigned long,
2009 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
Al Viro9206de92009-12-03 15:23:11 -05002011 unsigned long error = arch_mmap_check(addr, len, flags);
2012 if (error)
2013 return error;
2014
2015 /* Careful about overflows.. */
2016 if (len > TASK_SIZE)
2017 return -ENOMEM;
2018
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002019 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04002020 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002021 get_area = file->f_op->get_unmapped_area;
2022 addr = get_area(file, addr, len, pgoff, flags);
2023 if (IS_ERR_VALUE(addr))
2024 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002026 if (addr > TASK_SIZE - len)
2027 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002028 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002029 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002030
Al Viro9ac4ed42012-05-30 17:13:15 -04002031 addr = arch_rebalance_pgtables(addr, len);
2032 error = security_mmap_addr(addr);
2033 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034}
2035
2036EXPORT_SYMBOL(get_unmapped_area);
2037
2038/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002039struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002041 struct rb_node *rb_node;
2042 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002044 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002045 vma = vmacache_find(mm, addr);
2046 if (likely(vma))
2047 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002049 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002051 while (rb_node) {
2052 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002054 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002055
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002056 if (tmp->vm_end > addr) {
2057 vma = tmp;
2058 if (tmp->vm_start <= addr)
2059 break;
2060 rb_node = rb_node->rb_left;
2061 } else
2062 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002064
2065 if (vma)
2066 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 return vma;
2068}
2069
2070EXPORT_SYMBOL(find_vma);
2071
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002072/*
2073 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002074 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075struct vm_area_struct *
2076find_vma_prev(struct mm_struct *mm, unsigned long addr,
2077 struct vm_area_struct **pprev)
2078{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002079 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002081 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002082 if (vma) {
2083 *pprev = vma->vm_prev;
2084 } else {
2085 struct rb_node *rb_node = mm->mm_rb.rb_node;
2086 *pprev = NULL;
2087 while (rb_node) {
2088 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2089 rb_node = rb_node->rb_right;
2090 }
2091 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002092 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093}
2094
2095/*
2096 * Verify that the stack growth is acceptable and
2097 * update accounting. This is shared with both the
2098 * grow-up and grow-down cases.
2099 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002100static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101{
2102 struct mm_struct *mm = vma->vm_mm;
2103 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08002104 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002107 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 return -ENOMEM;
2109
2110 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08002111 actual_size = size;
2112 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2113 actual_size -= PAGE_SIZE;
Jason Low4db0c3c2015-04-15 16:14:08 -07002114 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 return -ENOMEM;
2116
2117 /* mlock limit tests */
2118 if (vma->vm_flags & VM_LOCKED) {
2119 unsigned long locked;
2120 unsigned long limit;
2121 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002122 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002123 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 if (locked > limit && !capable(CAP_IPC_LOCK))
2125 return -ENOMEM;
2126 }
2127
Adam Litke0d59a012007-01-30 14:35:39 -08002128 /* Check to ensure the stack will not grow into a hugetlb-only region */
2129 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2130 vma->vm_end - size;
2131 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2132 return -EFAULT;
2133
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 /*
2135 * Overcommit.. This must be the final test, as it will
2136 * update security statistics.
2137 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002138 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 return -ENOMEM;
2140
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 return 0;
2142}
2143
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002144#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002146 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2147 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002149int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150{
Oleg Nesterov09357812015-11-05 18:48:17 -08002151 struct mm_struct *mm = vma->vm_mm;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002152 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
2154 if (!(vma->vm_flags & VM_GROWSUP))
2155 return -EFAULT;
2156
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002157 /* Guard against wrapping around to address 0. */
2158 if (address < PAGE_ALIGN(address+4))
2159 address = PAGE_ALIGN(address+4);
2160 else
2161 return -ENOMEM;
2162
2163 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (unlikely(anon_vma_prepare(vma)))
2165 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
2167 /*
2168 * vma->vm_start/vm_end cannot change under us because the caller
2169 * is required to hold the mmap_sem in read mode. We need the
2170 * anon_vma lock to serialize against concurrent expand_stacks.
2171 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002172 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
2174 /* Somebody else might have raced and expanded it already */
2175 if (address > vma->vm_end) {
2176 unsigned long size, grow;
2177
2178 size = address - vma->vm_start;
2179 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2180
Hugh Dickins42c36f62011-05-09 17:44:42 -07002181 error = -ENOMEM;
2182 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2183 error = acct_stack_growth(vma, size, grow);
2184 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002185 /*
2186 * vma_gap_update() doesn't support concurrent
2187 * updates, but we only hold a shared mmap_sem
2188 * lock here, so we need to protect against
2189 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002190 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002191 * we don't guarantee that all growable vmas
2192 * in a mm share the same root anon vma.
2193 * So, we reuse mm->page_table_lock to guard
2194 * against concurrent vma expansions.
2195 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002196 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002197 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002198 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002199 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002200 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002201 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002202 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002203 if (vma->vm_next)
2204 vma_gap_update(vma->vm_next);
2205 else
Oleg Nesterov09357812015-11-05 18:48:17 -08002206 mm->highest_vm_end = address;
2207 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002208
Hugh Dickins42c36f62011-05-09 17:44:42 -07002209 perf_event_mmap(vma);
2210 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002213 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002214 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002215 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 return error;
2217}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002218#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220/*
2221 * vma is the first one with address < vma->vm_start. Have to extend vma.
2222 */
Michal Hockod05f3162011-05-24 17:11:44 -07002223int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002224 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225{
Oleg Nesterov09357812015-11-05 18:48:17 -08002226 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 int error;
2228
Eric Paris88694772007-11-26 18:47:26 -05002229 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002230 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002231 if (error)
2232 return error;
2233
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002234 /* We must make sure the anon_vma is allocated. */
2235 if (unlikely(anon_vma_prepare(vma)))
2236 return -ENOMEM;
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 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002243 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245 /* Somebody else might have raced and expanded it already */
2246 if (address < vma->vm_start) {
2247 unsigned long size, grow;
2248
2249 size = vma->vm_end - address;
2250 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2251
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002252 error = -ENOMEM;
2253 if (grow <= vma->vm_pgoff) {
2254 error = acct_stack_growth(vma, size, grow);
2255 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002256 /*
2257 * vma_gap_update() doesn't support concurrent
2258 * updates, but we only hold a shared mmap_sem
2259 * lock here, so we need to protect against
2260 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002261 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002262 * we don't guarantee that all growable vmas
2263 * in a mm share the same root anon vma.
2264 * So, we reuse mm->page_table_lock to guard
2265 * against concurrent vma expansions.
2266 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002267 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002268 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002269 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002270 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002271 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002272 vma->vm_start = address;
2273 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002274 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002275 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002276 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002277
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002278 perf_event_mmap(vma);
2279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 }
2281 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002282 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002283 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002284 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 return error;
2286}
2287
Linus Torvalds09884962013-02-27 08:36:04 -08002288/*
2289 * Note how expand_stack() refuses to expand the stack all the way to
2290 * abut the next virtual mapping, *unless* that mapping itself is also
2291 * a stack mapping. We want to leave room for a guard page, after all
2292 * (the guard page itself is not added here, that is done by the
2293 * actual page faulting logic)
2294 *
2295 * This matches the behavior of the guard page logic (see mm/memory.c:
2296 * check_stack_guard_page()), which only allows the guard page to be
2297 * removed under these circumstances.
2298 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002299#ifdef CONFIG_STACK_GROWSUP
2300int expand_stack(struct vm_area_struct *vma, unsigned long address)
2301{
Linus Torvalds09884962013-02-27 08:36:04 -08002302 struct vm_area_struct *next;
2303
2304 address &= PAGE_MASK;
2305 next = vma->vm_next;
2306 if (next && next->vm_start == address + PAGE_SIZE) {
2307 if (!(next->vm_flags & VM_GROWSUP))
2308 return -ENOMEM;
2309 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002310 return expand_upwards(vma, address);
2311}
2312
2313struct vm_area_struct *
2314find_extend_vma(struct mm_struct *mm, unsigned long addr)
2315{
2316 struct vm_area_struct *vma, *prev;
2317
2318 addr &= PAGE_MASK;
2319 vma = find_vma_prev(mm, addr, &prev);
2320 if (vma && (vma->vm_start <= addr))
2321 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002322 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002323 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002324 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002325 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002326 return prev;
2327}
2328#else
2329int expand_stack(struct vm_area_struct *vma, unsigned long address)
2330{
Linus Torvalds09884962013-02-27 08:36:04 -08002331 struct vm_area_struct *prev;
2332
2333 address &= PAGE_MASK;
2334 prev = vma->vm_prev;
2335 if (prev && prev->vm_end == address) {
2336 if (!(prev->vm_flags & VM_GROWSDOWN))
2337 return -ENOMEM;
2338 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002339 return expand_downwards(vma, address);
2340}
2341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002343find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344{
vishnu.pscc71aba2014-10-09 15:26:29 -07002345 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 unsigned long start;
2347
2348 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002349 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 if (!vma)
2351 return NULL;
2352 if (vma->vm_start <= addr)
2353 return vma;
2354 if (!(vma->vm_flags & VM_GROWSDOWN))
2355 return NULL;
2356 start = vma->vm_start;
2357 if (expand_stack(vma, addr))
2358 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002359 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002360 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 return vma;
2362}
2363#endif
2364
Jesse Barnese1d6d012014-12-12 16:55:27 -08002365EXPORT_SYMBOL_GPL(find_extend_vma);
2366
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002368 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002370 *
2371 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002373static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002375 unsigned long nr_accounted = 0;
2376
Hugh Dickins365e9c872005-10-29 18:16:18 -07002377 /* Update high watermark before we lower total_vm */
2378 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002380 long nrpages = vma_pages(vma);
2381
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002382 if (vma->vm_flags & VM_ACCOUNT)
2383 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002384 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002385 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002386 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002387 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 validate_mm(mm);
2389}
2390
2391/*
2392 * Get rid of page table information in the indicated region.
2393 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002394 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 */
2396static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002397 struct vm_area_struct *vma, struct vm_area_struct *prev,
2398 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399{
vishnu.pscc71aba2014-10-09 15:26:29 -07002400 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002401 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
2403 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002404 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002405 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002406 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002407 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002408 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002409 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410}
2411
2412/*
2413 * Create a list of vma's touched by the unmap, removing them from the mm's
2414 * vma list as we go..
2415 */
2416static void
2417detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2418 struct vm_area_struct *prev, unsigned long end)
2419{
2420 struct vm_area_struct **insertion_point;
2421 struct vm_area_struct *tail_vma = NULL;
2422
2423 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002424 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002426 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 mm->map_count--;
2428 tail_vma = vma;
2429 vma = vma->vm_next;
2430 } while (vma && vma->vm_start < end);
2431 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002432 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002433 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002434 vma_gap_update(vma);
2435 } else
2436 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002438
2439 /* Kill the cache */
2440 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441}
2442
2443/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002444 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2445 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002447static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 unsigned long addr, int new_below)
2449{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002451 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
Andi Kleena5516432008-07-23 21:27:41 -07002453 if (is_vm_hugetlb_page(vma) && (addr &
2454 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 return -EINVAL;
2456
Christoph Lametere94b1762006-12-06 20:33:17 -08002457 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002459 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
2461 /* most fields are the same, copy all, and then fixup */
2462 *new = *vma;
2463
Rik van Riel5beb4932010-03-05 13:42:07 -08002464 INIT_LIST_HEAD(&new->anon_vma_chain);
2465
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 if (new_below)
2467 new->vm_end = addr;
2468 else {
2469 new->vm_start = addr;
2470 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2471 }
2472
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002473 err = vma_dup_policy(vma, new);
2474 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002475 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002477 err = anon_vma_clone(new, vma);
2478 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002479 goto out_free_mpol;
2480
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002481 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 get_file(new->vm_file);
2483
2484 if (new->vm_ops && new->vm_ops->open)
2485 new->vm_ops->open(new);
2486
2487 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002488 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2490 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002491 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Rik van Riel5beb4932010-03-05 13:42:07 -08002493 /* Success. */
2494 if (!err)
2495 return 0;
2496
2497 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002498 if (new->vm_ops && new->vm_ops->close)
2499 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002500 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002501 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002502 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002503 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002504 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002505 out_free_vma:
2506 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002507 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508}
2509
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002510/*
2511 * Split a vma into two pieces at address 'addr', a new vma is allocated
2512 * either for the first part or the tail.
2513 */
2514int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2515 unsigned long addr, int new_below)
2516{
2517 if (mm->map_count >= sysctl_max_map_count)
2518 return -ENOMEM;
2519
2520 return __split_vma(mm, vma, addr, new_below);
2521}
2522
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523/* Munmap is split into 2 main parts -- this part which finds
2524 * what needs doing, and the areas themselves, which do the
2525 * work. This now handles partial unmappings.
2526 * Jeremy Fitzhardinge <jeremy@goop.org>
2527 */
2528int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2529{
2530 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002531 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002533 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 return -EINVAL;
2535
vishnu.pscc71aba2014-10-09 15:26:29 -07002536 len = PAGE_ALIGN(len);
2537 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 return -EINVAL;
2539
2540 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002541 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002542 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002544 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002545 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
2547 /* if it doesn't overlap, we have nothing.. */
2548 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002549 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 return 0;
2551
2552 /*
2553 * If we need to split any vma, do it now to save pain later.
2554 *
2555 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2556 * unmapped vm_area_struct will remain in use: so lower split_vma
2557 * places tmp vma above, and higher split_vma places tmp vma below.
2558 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002559 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002560 int error;
2561
2562 /*
2563 * Make sure that map_count on return from munmap() will
2564 * not exceed its limit; but let map_count go just above
2565 * its limit temporarily, to help free resources as expected.
2566 */
2567 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2568 return -ENOMEM;
2569
2570 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 if (error)
2572 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002573 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 }
2575
2576 /* Does it split the last one? */
2577 last = find_vma(mm, end);
2578 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002579 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 if (error)
2581 return error;
2582 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002583 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
2585 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002586 * unlock any mlock()ed ranges before detaching vmas
2587 */
2588 if (mm->locked_vm) {
2589 struct vm_area_struct *tmp = vma;
2590 while (tmp && tmp->vm_start < end) {
2591 if (tmp->vm_flags & VM_LOCKED) {
2592 mm->locked_vm -= vma_pages(tmp);
2593 munlock_vma_pages_all(tmp);
2594 }
2595 tmp = tmp->vm_next;
2596 }
2597 }
2598
2599 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 * Remove the vma's, and unmap the actual pages
2601 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002602 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2603 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604
Dave Hansen1de4fa12014-11-14 07:18:31 -08002605 arch_unmap(mm, vma, start, end);
2606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002608 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
2610 return 0;
2611}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612
Al Virobfce2812012-04-20 21:57:04 -04002613int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002614{
2615 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002616 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002617
2618 down_write(&mm->mmap_sem);
2619 ret = do_munmap(mm, start, len);
2620 up_write(&mm->mmap_sem);
2621 return ret;
2622}
2623EXPORT_SYMBOL(vm_munmap);
2624
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002625SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002628 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629}
2630
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002631
2632/*
2633 * Emulation of deprecated remap_file_pages() syscall.
2634 */
2635SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2636 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2637{
2638
2639 struct mm_struct *mm = current->mm;
2640 struct vm_area_struct *vma;
2641 unsigned long populate = 0;
2642 unsigned long ret = -EINVAL;
2643 struct file *file;
2644
2645 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
2646 "See Documentation/vm/remap_file_pages.txt.\n",
2647 current->comm, current->pid);
2648
2649 if (prot)
2650 return ret;
2651 start = start & PAGE_MASK;
2652 size = size & PAGE_MASK;
2653
2654 if (start + size <= start)
2655 return ret;
2656
2657 /* Does pgoff wrap? */
2658 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2659 return ret;
2660
2661 down_write(&mm->mmap_sem);
2662 vma = find_vma(mm, start);
2663
2664 if (!vma || !(vma->vm_flags & VM_SHARED))
2665 goto out;
2666
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002667 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002668 goto out;
2669
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002670 if (start + size > vma->vm_end) {
2671 struct vm_area_struct *next;
2672
2673 for (next = vma->vm_next; next; next = next->vm_next) {
2674 /* hole between vmas ? */
2675 if (next->vm_start != next->vm_prev->vm_end)
2676 goto out;
2677
2678 if (next->vm_file != vma->vm_file)
2679 goto out;
2680
2681 if (next->vm_flags != vma->vm_flags)
2682 goto out;
2683
2684 if (start + size <= next->vm_end)
2685 break;
2686 }
2687
2688 if (!next)
2689 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002690 }
2691
2692 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2693 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2694 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2695
2696 flags &= MAP_NONBLOCK;
2697 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2698 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002699 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002700 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002701
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002702 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002703 for (tmp = vma; tmp->vm_start >= start + size;
2704 tmp = tmp->vm_next) {
2705 munlock_vma_pages_range(tmp,
2706 max(tmp->vm_start, start),
2707 min(tmp->vm_end, start + size));
2708 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002709 }
2710
2711 file = get_file(vma->vm_file);
2712 ret = do_mmap_pgoff(vma->vm_file, start, size,
2713 prot, flags, pgoff, &populate);
2714 fput(file);
2715out:
2716 up_write(&mm->mmap_sem);
2717 if (populate)
2718 mm_populate(ret, populate);
2719 if (!IS_ERR_VALUE(ret))
2720 ret = 0;
2721 return ret;
2722}
2723
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724static inline void verify_mm_writelocked(struct mm_struct *mm)
2725{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002726#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2728 WARN_ON(1);
2729 up_read(&mm->mmap_sem);
2730 }
2731#endif
2732}
2733
2734/*
2735 * this is really a simplified "do_mmap". it only handles
2736 * anonymous maps. eventually we may be able to do some
2737 * brk-specific accounting here.
2738 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002739static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740{
vishnu.pscc71aba2014-10-09 15:26:29 -07002741 struct mm_struct *mm = current->mm;
2742 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002744 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002746 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
2748 len = PAGE_ALIGN(len);
2749 if (!len)
2750 return addr;
2751
Kirill Korotaev3a459752006-09-07 14:17:04 +04002752 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2753
Al Viro2c6a1012009-12-03 19:40:46 -05002754 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002755 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002756 return error;
2757
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002758 error = mlock_future_check(mm, mm->def_flags, len);
2759 if (error)
2760 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
2762 /*
2763 * mm->mmap_sem is required to protect against another thread
2764 * changing the mappings in case we sleep.
2765 */
2766 verify_mm_writelocked(mm);
2767
2768 /*
2769 * Clear old maps. this also does some error checking for us
2770 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002771 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2772 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 if (do_munmap(mm, addr, len))
2774 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 }
2776
2777 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002778 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 return -ENOMEM;
2780
2781 if (mm->map_count > sysctl_max_map_count)
2782 return -ENOMEM;
2783
Al Viro191c5422012-02-13 03:58:52 +00002784 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 return -ENOMEM;
2786
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002788 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002789 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002790 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 goto out;
2792
2793 /*
2794 * create a vma struct for an anonymous mapping
2795 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002796 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 if (!vma) {
2798 vm_unacct_memory(len >> PAGE_SHIFT);
2799 return -ENOMEM;
2800 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801
Rik van Riel5beb4932010-03-05 13:42:07 -08002802 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 vma->vm_mm = mm;
2804 vma->vm_start = addr;
2805 vma->vm_end = addr + len;
2806 vma->vm_pgoff = pgoff;
2807 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002808 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 vma_link(mm, vma, prev, rb_link, rb_parent);
2810out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002811 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002813 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002814 if (flags & VM_LOCKED)
2815 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002816 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 return addr;
2818}
2819
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002820unsigned long vm_brk(unsigned long addr, unsigned long len)
2821{
2822 struct mm_struct *mm = current->mm;
2823 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002824 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002825
2826 down_write(&mm->mmap_sem);
2827 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002828 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002829 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002830 if (populate)
2831 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002832 return ret;
2833}
2834EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835
2836/* Release all mmaps. */
2837void exit_mmap(struct mm_struct *mm)
2838{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002839 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002840 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 unsigned long nr_accounted = 0;
2842
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002843 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002844 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002845
Rik van Rielba470de2008-10-18 20:26:50 -07002846 if (mm->locked_vm) {
2847 vma = mm->mmap;
2848 while (vma) {
2849 if (vma->vm_flags & VM_LOCKED)
2850 munlock_vma_pages_all(vma);
2851 vma = vma->vm_next;
2852 }
2853 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002854
2855 arch_exit_mmap(mm);
2856
Rik van Rielba470de2008-10-18 20:26:50 -07002857 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002858 if (!vma) /* Can happen if dup_mmap() received an OOM */
2859 return;
2860
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002863 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002864 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002865 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002866 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002867
Hugh Dickins6ee86302013-04-29 15:07:44 -07002868 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002869 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002872 * Walk the list again, actually closing and freeing it,
2873 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002875 while (vma) {
2876 if (vma->vm_flags & VM_ACCOUNT)
2877 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002878 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002879 }
2880 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881}
2882
2883/* Insert vm structure into process list sorted by address
2884 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002885 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002887int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888{
Hugh Dickins6597d782012-10-08 16:29:07 -07002889 struct vm_area_struct *prev;
2890 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891
Chen Gangc9d13f52015-09-08 15:04:08 -07002892 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2893 &prev, &rb_link, &rb_parent))
2894 return -ENOMEM;
2895 if ((vma->vm_flags & VM_ACCOUNT) &&
2896 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2897 return -ENOMEM;
2898
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 /*
2900 * The vm_pgoff of a purely anonymous vma should be irrelevant
2901 * until its first write fault, when page's anon_vma and index
2902 * are set. But now set the vm_pgoff it will almost certainly
2903 * end up with (unless mremap moves it elsewhere before that
2904 * first wfault), so /proc/pid/maps tells a consistent story.
2905 *
2906 * By setting it to reflect the virtual start address of the
2907 * vma, merges and splits can happen in a seamless way, just
2908 * using the existing file pgoff checks and manipulations.
2909 * Similarly in do_mmap_pgoff and in do_brk.
2910 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002911 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 BUG_ON(vma->anon_vma);
2913 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2914 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302915
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 vma_link(mm, vma, prev, rb_link, rb_parent);
2917 return 0;
2918}
2919
2920/*
2921 * Copy the vma structure to a new location in the same mm,
2922 * prior to moving page table entries, to effect an mremap move.
2923 */
2924struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002925 unsigned long addr, unsigned long len, pgoff_t pgoff,
2926 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927{
2928 struct vm_area_struct *vma = *vmap;
2929 unsigned long vma_start = vma->vm_start;
2930 struct mm_struct *mm = vma->vm_mm;
2931 struct vm_area_struct *new_vma, *prev;
2932 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002933 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934
2935 /*
2936 * If anonymous vma has not yet been faulted, update new pgoff
2937 * to match new location, to increase its chance of merging.
2938 */
Oleg Nesterovce757992015-09-08 14:58:34 -07002939 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002941 faulted_in_anon_vma = false;
2942 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943
Hugh Dickins6597d782012-10-08 16:29:07 -07002944 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2945 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002947 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2948 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 if (new_vma) {
2950 /*
2951 * Source vma may have been merged into new_vma
2952 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002953 if (unlikely(vma_start >= new_vma->vm_start &&
2954 vma_start < new_vma->vm_end)) {
2955 /*
2956 * The only way we can get a vma_merge with
2957 * self during an mremap is if the vma hasn't
2958 * been faulted in yet and we were allowed to
2959 * reset the dst vma->vm_pgoff to the
2960 * destination address of the mremap to allow
2961 * the merge to happen. mremap must change the
2962 * vm_pgoff linearity between src and dst vmas
2963 * (in turn preventing a vma_merge) to be
2964 * safe. It is only safe to keep the vm_pgoff
2965 * linear if there are no pages mapped yet.
2966 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002967 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002968 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002969 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002970 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002972 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07002973 if (!new_vma)
2974 goto out;
2975 *new_vma = *vma;
2976 new_vma->vm_start = addr;
2977 new_vma->vm_end = addr + len;
2978 new_vma->vm_pgoff = pgoff;
2979 if (vma_dup_policy(vma, new_vma))
2980 goto out_free_vma;
2981 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2982 if (anon_vma_clone(new_vma, vma))
2983 goto out_free_mempol;
2984 if (new_vma->vm_file)
2985 get_file(new_vma->vm_file);
2986 if (new_vma->vm_ops && new_vma->vm_ops->open)
2987 new_vma->vm_ops->open(new_vma);
2988 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2989 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 }
2991 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002992
Chen Gange3975892015-09-08 15:03:38 -07002993out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002994 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07002995out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08002996 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07002997out:
Rik van Riel5beb4932010-03-05 13:42:07 -08002998 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003000
3001/*
3002 * Return true if the calling process may expand its vm space by the passed
3003 * number of pages
3004 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003005bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003006{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003007 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3008 return false;
3009
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003010 if (is_data_mapping(flags) &&
3011 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
3012 if (ignore_rlimit_data)
3013 pr_warn_once("%s (%d): VmData %lu exceed data ulimit "
3014 "%lu. Will be forbidden soon.\n",
3015 current->comm, current->pid,
3016 (mm->data_vm + npages) << PAGE_SHIFT,
3017 rlimit(RLIMIT_DATA));
3018 else
3019 return false;
3020 }
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003021
3022 return true;
3023}
3024
3025void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3026{
3027 mm->total_vm += npages;
3028
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003029 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003030 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003031 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003032 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003033 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003034 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003035}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003036
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003037static int special_mapping_fault(struct vm_area_struct *vma,
3038 struct vm_fault *vmf);
3039
3040/*
3041 * Having a close hook prevents vma merging regardless of flags.
3042 */
3043static void special_mapping_close(struct vm_area_struct *vma)
3044{
3045}
3046
3047static const char *special_mapping_name(struct vm_area_struct *vma)
3048{
3049 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3050}
3051
3052static const struct vm_operations_struct special_mapping_vmops = {
3053 .close = special_mapping_close,
3054 .fault = special_mapping_fault,
3055 .name = special_mapping_name,
3056};
3057
3058static const struct vm_operations_struct legacy_special_mapping_vmops = {
3059 .close = special_mapping_close,
3060 .fault = special_mapping_fault,
3061};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003062
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003063static int special_mapping_fault(struct vm_area_struct *vma,
3064 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003065{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003066 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003067 struct page **pages;
3068
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003069 if (vma->vm_ops == &legacy_special_mapping_vmops)
3070 pages = vma->vm_private_data;
3071 else
3072 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
3073 pages;
3074
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003075 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003076 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003077
3078 if (*pages) {
3079 struct page *page = *pages;
3080 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003081 vmf->page = page;
3082 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003083 }
3084
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003085 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003086}
3087
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003088static struct vm_area_struct *__install_special_mapping(
3089 struct mm_struct *mm,
3090 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003091 unsigned long vm_flags, void *priv,
3092 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003093{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003094 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003095 struct vm_area_struct *vma;
3096
3097 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3098 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003099 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003100
Rik van Riel5beb4932010-03-05 13:42:07 -08003101 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003102 vma->vm_mm = mm;
3103 vma->vm_start = addr;
3104 vma->vm_end = addr + len;
3105
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003106 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003107 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003108
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003109 vma->vm_ops = ops;
3110 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003111
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003112 ret = insert_vm_struct(mm, vma);
3113 if (ret)
3114 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003115
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003116 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003117
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003118 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003119
Stefani Seibold3935ed62014-03-17 23:22:02 +01003120 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003121
3122out:
3123 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003124 return ERR_PTR(ret);
3125}
3126
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003127/*
3128 * Called with mm->mmap_sem held for writing.
3129 * Insert a new vma covering the given region, with the given flags.
3130 * Its pages are supplied by the given array of struct page *.
3131 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3132 * The region past the last page supplied will always produce SIGBUS.
3133 * The array pointer and the pages it points to are assumed to stay alive
3134 * for as long as this mapping might exist.
3135 */
3136struct vm_area_struct *_install_special_mapping(
3137 struct mm_struct *mm,
3138 unsigned long addr, unsigned long len,
3139 unsigned long vm_flags, const struct vm_special_mapping *spec)
3140{
Chen Gang27f28b92015-11-05 18:48:41 -08003141 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3142 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003143}
3144
Stefani Seibold3935ed62014-03-17 23:22:02 +01003145int install_special_mapping(struct mm_struct *mm,
3146 unsigned long addr, unsigned long len,
3147 unsigned long vm_flags, struct page **pages)
3148{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003149 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003150 mm, addr, len, vm_flags, (void *)pages,
3151 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003152
Duan Jiong14bd5b42014-06-04 16:07:05 -07003153 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003154}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003155
3156static DEFINE_MUTEX(mm_all_locks_mutex);
3157
Peter Zijlstra454ed842008-08-11 09:30:25 +02003158static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003159{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003160 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003161 /*
3162 * The LSB of head.next can't change from under us
3163 * because we hold the mm_all_locks_mutex.
3164 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003165 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003166 /*
3167 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003168 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003169 * the same anon_vma we won't take it again.
3170 *
3171 * No need of atomic instructions here, head.next
3172 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003173 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003174 */
3175 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003176 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003177 BUG();
3178 }
3179}
3180
Peter Zijlstra454ed842008-08-11 09:30:25 +02003181static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003182{
3183 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3184 /*
3185 * AS_MM_ALL_LOCKS can't change from under us because
3186 * we hold the mm_all_locks_mutex.
3187 *
3188 * Operations on ->flags have to be atomic because
3189 * even if AS_MM_ALL_LOCKS is stable thanks to the
3190 * mm_all_locks_mutex, there may be other cpus
3191 * changing other bitflags in parallel to us.
3192 */
3193 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3194 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003195 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003196 }
3197}
3198
3199/*
3200 * This operation locks against the VM for all pte/vma/mm related
3201 * operations that could ever happen on a certain mm. This includes
3202 * vmtruncate, try_to_unmap, and all page faults.
3203 *
3204 * The caller must take the mmap_sem in write mode before calling
3205 * mm_take_all_locks(). The caller isn't allowed to release the
3206 * mmap_sem until mm_drop_all_locks() returns.
3207 *
3208 * mmap_sem in write mode is required in order to block all operations
3209 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003210 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003211 * anon_vmas to be associated with existing vmas.
3212 *
3213 * A single task can't take more than one mm_take_all_locks() in a row
3214 * or it would deadlock.
3215 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003216 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003217 * mapping->flags avoid to take the same lock twice, if more than one
3218 * vma in this mm is backed by the same anon_vma or address_space.
3219 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003220 * We take locks in following order, accordingly to comment at beginning
3221 * of mm/rmap.c:
3222 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3223 * hugetlb mapping);
3224 * - all i_mmap_rwsem locks;
3225 * - all anon_vma->rwseml
3226 *
3227 * We can take all locks within these types randomly because the VM code
3228 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3229 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003230 *
3231 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3232 * that may have to take thousand of locks.
3233 *
3234 * mm_take_all_locks() can fail if it's interrupted by signals.
3235 */
3236int mm_take_all_locks(struct mm_struct *mm)
3237{
3238 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003239 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003240
3241 BUG_ON(down_read_trylock(&mm->mmap_sem));
3242
3243 mutex_lock(&mm_all_locks_mutex);
3244
3245 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3246 if (signal_pending(current))
3247 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003248 if (vma->vm_file && vma->vm_file->f_mapping &&
3249 is_vm_hugetlb_page(vma))
3250 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3251 }
3252
3253 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3254 if (signal_pending(current))
3255 goto out_unlock;
3256 if (vma->vm_file && vma->vm_file->f_mapping &&
3257 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003258 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003259 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003260
3261 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3262 if (signal_pending(current))
3263 goto out_unlock;
3264 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003265 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3266 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003267 }
3268
Kautuk Consul584cff52011-10-31 17:08:59 -07003269 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003270
3271out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003272 mm_drop_all_locks(mm);
3273 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003274}
3275
3276static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3277{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003278 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003279 /*
3280 * The LSB of head.next can't change to 0 from under
3281 * us because we hold the mm_all_locks_mutex.
3282 *
3283 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003284 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003285 * never see our bitflag.
3286 *
3287 * No need of atomic instructions here, head.next
3288 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003289 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003290 */
3291 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003292 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003293 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003294 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003295 }
3296}
3297
3298static void vm_unlock_mapping(struct address_space *mapping)
3299{
3300 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3301 /*
3302 * AS_MM_ALL_LOCKS can't change to 0 from under us
3303 * because we hold the mm_all_locks_mutex.
3304 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003305 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003306 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3307 &mapping->flags))
3308 BUG();
3309 }
3310}
3311
3312/*
3313 * The mmap_sem cannot be released by the caller until
3314 * mm_drop_all_locks() returns.
3315 */
3316void mm_drop_all_locks(struct mm_struct *mm)
3317{
3318 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003319 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003320
3321 BUG_ON(down_read_trylock(&mm->mmap_sem));
3322 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3323
3324 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3325 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003326 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3327 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003328 if (vma->vm_file && vma->vm_file->f_mapping)
3329 vm_unlock_mapping(vma->vm_file->f_mapping);
3330 }
3331
3332 mutex_unlock(&mm_all_locks_mutex);
3333}
David Howells8feae132009-01-08 12:04:47 +00003334
3335/*
3336 * initialise the VMA slab
3337 */
3338void __init mmap_init(void)
3339{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003340 int ret;
3341
Tejun Heo908c7f12014-09-08 09:51:29 +09003342 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003343 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003344}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003345
3346/*
3347 * Initialise sysctl_user_reserve_kbytes.
3348 *
3349 * This is intended to prevent a user from starting a single memory hogging
3350 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3351 * mode.
3352 *
3353 * The default value is min(3% of free memory, 128MB)
3354 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3355 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003356static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003357{
3358 unsigned long free_kbytes;
3359
3360 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3361
3362 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3363 return 0;
3364}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003365subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003366
3367/*
3368 * Initialise sysctl_admin_reserve_kbytes.
3369 *
3370 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3371 * to log in and kill a memory hogging process.
3372 *
3373 * Systems with more than 256MB will reserve 8MB, enough to recover
3374 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3375 * only reserve 3% of free pages by default.
3376 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003377static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003378{
3379 unsigned long free_kbytes;
3380
3381 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3382
3383 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3384 return 0;
3385}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003386subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003387
3388/*
3389 * Reinititalise user and admin reserves if memory is added or removed.
3390 *
3391 * The default user reserve max is 128MB, and the default max for the
3392 * admin reserve is 8MB. These are usually, but not always, enough to
3393 * enable recovery from a memory hogging process using login/sshd, a shell,
3394 * and tools like top. It may make sense to increase or even disable the
3395 * reserve depending on the existence of swap or variations in the recovery
3396 * tools. So, the admin may have changed them.
3397 *
3398 * If memory is added and the reserves have been eliminated or increased above
3399 * the default max, then we'll trust the admin.
3400 *
3401 * If memory is removed and there isn't enough free memory, then we
3402 * need to reset the reserves.
3403 *
3404 * Otherwise keep the reserve set by the admin.
3405 */
3406static int reserve_mem_notifier(struct notifier_block *nb,
3407 unsigned long action, void *data)
3408{
3409 unsigned long tmp, free_kbytes;
3410
3411 switch (action) {
3412 case MEM_ONLINE:
3413 /* Default max is 128MB. Leave alone if modified by operator. */
3414 tmp = sysctl_user_reserve_kbytes;
3415 if (0 < tmp && tmp < (1UL << 17))
3416 init_user_reserve();
3417
3418 /* Default max is 8MB. Leave alone if modified by operator. */
3419 tmp = sysctl_admin_reserve_kbytes;
3420 if (0 < tmp && tmp < (1UL << 13))
3421 init_admin_reserve();
3422
3423 break;
3424 case MEM_OFFLINE:
3425 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3426
3427 if (sysctl_user_reserve_kbytes > free_kbytes) {
3428 init_user_reserve();
3429 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3430 sysctl_user_reserve_kbytes);
3431 }
3432
3433 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3434 init_admin_reserve();
3435 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3436 sysctl_admin_reserve_kbytes);
3437 }
3438 break;
3439 default:
3440 break;
3441 }
3442 return NOTIFY_OK;
3443}
3444
3445static struct notifier_block reserve_mem_nb = {
3446 .notifier_call = reserve_mem_notifier,
3447};
3448
3449static int __meminit init_reserve_notifier(void)
3450{
3451 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003452 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003453
3454 return 0;
3455}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003456subsys_initcall(init_reserve_notifier);