blob: b3f00b616b810e4362effddd437b320acdcedb7e [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46#include <asm/uaccess.h>
47#include <asm/cacheflush.h>
48#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020049#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Jan Beulich42b77722008-07-23 21:27:10 -070051#include "internal.h"
52
Kirill Korotaev3a459752006-09-07 14:17:04 +040053#ifndef arch_mmap_check
54#define arch_mmap_check(addr, len, flags) (0)
55#endif
56
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080057#ifndef arch_rebalance_pgtables
58#define arch_rebalance_pgtables(addr, len) (addr)
59#endif
60
Daniel Cashmand07e2252016-01-14 15:19:53 -080061#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
62const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
63const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
64int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
65#endif
66#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
67const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
68const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
69int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
70#endif
71
72
Hugh Dickinse0da3822005-04-19 13:29:15 -070073static void unmap_region(struct mm_struct *mm,
74 struct vm_area_struct *vma, struct vm_area_struct *prev,
75 unsigned long start, unsigned long end);
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/* description of effects of mapping type and prot in current implementation.
78 * this is due to the limited x86 page protection hardware. The expected
79 * behavior is in parens:
80 *
81 * map_type prot
82 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
83 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
84 * w: (no) no w: (no) no w: (yes) yes w: (no) no
85 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070086 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
88 * w: (no) no w: (no) no w: (copy) copy w: (no) no
89 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
90 *
91 */
92pgprot_t protection_map[16] = {
93 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
94 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
95};
96
Hugh Dickins804af2c2006-07-26 21:39:49 +010097pgprot_t vm_get_page_prot(unsigned long vm_flags)
98{
Dave Kleikampb845f312008-07-08 00:28:51 +100099 return __pgprot(pgprot_val(protection_map[vm_flags &
100 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
101 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100102}
103EXPORT_SYMBOL(vm_get_page_prot);
104
Peter Feiner64e45502014-10-13 15:55:46 -0700105static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
106{
107 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
108}
109
110/* Update vma->vm_page_prot to reflect vma->vm_flags. */
111void vma_set_page_prot(struct vm_area_struct *vma)
112{
113 unsigned long vm_flags = vma->vm_flags;
114
115 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
116 if (vma_wants_writenotify(vma)) {
117 vm_flags &= ~VM_SHARED;
118 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
119 vm_flags);
120 }
121}
122
123
Shaohua Li34679d72011-05-24 17:11:18 -0700124int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
125int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800126unsigned long sysctl_overcommit_kbytes __read_mostly;
Christoph Lameterc3d8c142005-09-06 15:16:33 -0700127int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700128unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700129unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -0700130/*
131 * Make sure vm_committed_as in one cacheline and not cacheline shared with
132 * other variables. It can be updated by several CPUs frequently.
133 */
134struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
136/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -0800137 * The global memory commitment made in the system can be a metric
138 * that can be used to drive ballooning decisions when Linux is hosted
139 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
140 * balancing memory across competing virtual machines that are hosted.
141 * Several metrics drive this policy engine including the guest reported
142 * memory commitment.
143 */
144unsigned long vm_memory_committed(void)
145{
146 return percpu_counter_read_positive(&vm_committed_as);
147}
148EXPORT_SYMBOL_GPL(vm_memory_committed);
149
150/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 * Check that a process has enough memory to allocate a new virtual
152 * mapping. 0 means there is enough memory for the allocation to
153 * succeed and -ENOMEM implies there is not.
154 *
155 * We currently support three overcommit policies, which are set via the
156 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
157 *
158 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
159 * Additional code 2002 Jul 20 by Robert Love.
160 *
161 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
162 *
163 * Note this is a helper function intended to be used by LSMs which
164 * wish to use this logic.
165 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700166int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167{
Roman Gushchin5703b082015-02-11 15:28:39 -0800168 long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -0700170 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
171 -(s64)vm_committed_as_batch * num_online_cpus(),
172 "memory commitment underflow");
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 vm_acct_memory(pages);
175
176 /*
177 * Sometimes we want to use more memory than we have
178 */
179 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
180 return 0;
181
182 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700183 free = global_page_state(NR_FREE_PAGES);
184 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Dmitry Finkc15bef32011-07-25 17:12:19 -0700186 /*
187 * shmem pages shouldn't be counted as free in this
188 * case, they can't be purged, only swapped out, and
189 * that won't affect the overall amount of available
190 * memory in the system.
191 */
192 free -= global_page_state(NR_SHMEM);
193
Shaohua Liec8acf22013-02-22 16:34:38 -0800194 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196 /*
197 * Any slabs which are created with the
198 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
199 * which are reclaimable, under pressure. The dentry
200 * cache and most inode caches should fall into this
201 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700202 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700205 * Leave reserved pages. The pages are not for anonymous pages.
206 */
207 if (free <= totalreserve_pages)
208 goto error;
209 else
210 free -= totalreserve_pages;
211
212 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700213 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 */
215 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700216 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218 if (free > pages)
219 return 0;
220
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700221 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 }
223
Jerome Marchand00619bc2013-11-12 15:08:31 -0800224 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700226 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 */
228 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700229 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700231 /*
232 * Don't let a single process grow so big a user can't recover
233 */
234 if (mm) {
235 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
Roman Gushchin5703b082015-02-11 15:28:39 -0800236 allowed -= min_t(long, mm->total_vm / 32, reserve);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700239 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700241error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 vm_unacct_memory(pages);
243
244 return -ENOMEM;
245}
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800248 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 */
250static void __remove_shared_vm_struct(struct vm_area_struct *vma,
251 struct file *file, struct address_space *mapping)
252{
253 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500254 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700256 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800259 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 flush_dcache_mmap_unlock(mapping);
261}
262
263/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700264 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700265 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700267void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
269 struct file *file = vma->vm_file;
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 if (file) {
272 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800273 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800275 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700277}
278
279/*
280 * Close a vm structure and free it, returning the next.
281 */
282static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
283{
284 struct vm_area_struct *next = vma->vm_next;
285
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700286 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 if (vma->vm_ops && vma->vm_ops->close)
288 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700289 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700290 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700291 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700293 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700296static unsigned long do_brk(unsigned long addr, unsigned long len);
297
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100298SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700300 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 unsigned long newbrk, oldbrk;
302 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700303 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800304 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
306 down_write(&mm->mmap_sem);
307
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700308#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800309 /*
310 * CONFIG_COMPAT_BRK can still be overridden by setting
311 * randomize_va_space to 2, which will still cause mm->start_brk
312 * to be arbitrarily shifted
313 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700314 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800315 min_brk = mm->start_brk;
316 else
317 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700318#else
319 min_brk = mm->start_brk;
320#endif
321 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700323
324 /*
325 * Check against rlimit here. If this check is done later after the test
326 * of oldbrk with newbrk then it can escape the test and let the data
327 * segment grow beyond its set limit the in case where the limit is
328 * not page aligned -Ram Gupta
329 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700330 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
331 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700332 goto out;
333
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 newbrk = PAGE_ALIGN(brk);
335 oldbrk = PAGE_ALIGN(mm->brk);
336 if (oldbrk == newbrk)
337 goto set_brk;
338
339 /* Always allow shrinking brk. */
340 if (brk <= mm->brk) {
341 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
342 goto set_brk;
343 goto out;
344 }
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 /* Check against existing mmap mappings. */
347 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
348 goto out;
349
350 /* Ok, looks good - let it rip. */
351 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
352 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354set_brk:
355 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800356 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
357 up_write(&mm->mmap_sem);
358 if (populate)
359 mm_populate(oldbrk, newbrk - oldbrk);
360 return brk;
361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362out:
363 retval = mm->brk;
364 up_write(&mm->mmap_sem);
365 return retval;
366}
367
Michel Lespinassed3737182012-12-11 16:01:38 -0800368static long vma_compute_subtree_gap(struct vm_area_struct *vma)
369{
370 unsigned long max, subtree_gap;
371 max = vma->vm_start;
372 if (vma->vm_prev)
373 max -= vma->vm_prev->vm_end;
374 if (vma->vm_rb.rb_left) {
375 subtree_gap = rb_entry(vma->vm_rb.rb_left,
376 struct vm_area_struct, vm_rb)->rb_subtree_gap;
377 if (subtree_gap > max)
378 max = subtree_gap;
379 }
380 if (vma->vm_rb.rb_right) {
381 subtree_gap = rb_entry(vma->vm_rb.rb_right,
382 struct vm_area_struct, vm_rb)->rb_subtree_gap;
383 if (subtree_gap > max)
384 max = subtree_gap;
385 }
386 return max;
387}
388
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700389#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390static int browse_rb(struct rb_root *root)
391{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800392 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 struct rb_node *nd, *pn = NULL;
394 unsigned long prev = 0, pend = 0;
395
396 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
397 struct vm_area_struct *vma;
398 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800399 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700400 pr_emerg("vm_start %lx < prev %lx\n",
401 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800402 bug = 1;
403 }
404 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700405 pr_emerg("vm_start %lx < pend %lx\n",
406 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800407 bug = 1;
408 }
409 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700410 pr_emerg("vm_start %lx > vm_end %lx\n",
411 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800412 bug = 1;
413 }
414 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700415 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800416 vma->rb_subtree_gap,
417 vma_compute_subtree_gap(vma));
418 bug = 1;
419 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 i++;
421 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800422 prev = vma->vm_start;
423 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 }
425 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800426 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800428 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700429 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800430 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800432 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
Michel Lespinassed3737182012-12-11 16:01:38 -0800435static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
436{
437 struct rb_node *nd;
438
439 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
440 struct vm_area_struct *vma;
441 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700442 VM_BUG_ON_VMA(vma != ignore &&
443 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
444 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700448static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
450 int bug = 0;
451 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800452 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700453 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700454
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700455 while (vma) {
456 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700457
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800458 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700459 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
460 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800461 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800462 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700463 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 i++;
465 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800466 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700467 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800468 bug = 1;
469 }
470 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700471 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700472 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800473 bug = 1;
474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800476 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700477 if (i != -1)
478 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800479 bug = 1;
480 }
Sasha Levin96dad672014-10-09 15:28:39 -0700481 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800484#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485#define validate_mm(mm) do { } while (0)
486#endif
487
Michel Lespinassed3737182012-12-11 16:01:38 -0800488RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
489 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
490
491/*
492 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
493 * vma->vm_prev->vm_end values changed, without modifying the vma's position
494 * in the rbtree.
495 */
496static void vma_gap_update(struct vm_area_struct *vma)
497{
498 /*
499 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
500 * function that does exacltly what we want.
501 */
502 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
503}
504
505static inline void vma_rb_insert(struct vm_area_struct *vma,
506 struct rb_root *root)
507{
508 /* All rb_subtree_gap values must be consistent prior to insertion */
509 validate_mm_rb(root, NULL);
510
511 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
512}
513
514static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
515{
516 /*
517 * All rb_subtree_gap values must be consistent prior to erase,
518 * with the possible exception of the vma being erased.
519 */
520 validate_mm_rb(root, vma);
521
522 /*
523 * Note rb_erase_augmented is a fairly large inline function,
524 * so make sure we instantiate it only once with our desired
525 * augmented rbtree callbacks.
526 */
527 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
528}
529
Michel Lespinassebf181b92012-10-08 16:31:39 -0700530/*
531 * vma has some anon_vma assigned, and is already inserted on that
532 * anon_vma's interval trees.
533 *
534 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
535 * vma must be removed from the anon_vma's interval trees using
536 * anon_vma_interval_tree_pre_update_vma().
537 *
538 * After the update, the vma will be reinserted using
539 * anon_vma_interval_tree_post_update_vma().
540 *
541 * The entire update must be protected by exclusive mmap_sem and by
542 * the root anon_vma's mutex.
543 */
544static inline void
545anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
546{
547 struct anon_vma_chain *avc;
548
549 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
550 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
551}
552
553static inline void
554anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
555{
556 struct anon_vma_chain *avc;
557
558 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
559 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
560}
561
Hugh Dickins6597d782012-10-08 16:29:07 -0700562static int find_vma_links(struct mm_struct *mm, unsigned long addr,
563 unsigned long end, struct vm_area_struct **pprev,
564 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Hugh Dickins6597d782012-10-08 16:29:07 -0700566 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568 __rb_link = &mm->mm_rb.rb_node;
569 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
571 while (*__rb_link) {
572 struct vm_area_struct *vma_tmp;
573
574 __rb_parent = *__rb_link;
575 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
576
577 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700578 /* Fail if an existing vma overlaps the area */
579 if (vma_tmp->vm_start < end)
580 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 __rb_link = &__rb_parent->rb_left;
582 } else {
583 rb_prev = __rb_parent;
584 __rb_link = &__rb_parent->rb_right;
585 }
586 }
587
588 *pprev = NULL;
589 if (rb_prev)
590 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
591 *rb_link = __rb_link;
592 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700593 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594}
595
Cyril Hrubise8420a82013-04-29 15:08:33 -0700596static unsigned long count_vma_pages_range(struct mm_struct *mm,
597 unsigned long addr, unsigned long end)
598{
599 unsigned long nr_pages = 0;
600 struct vm_area_struct *vma;
601
602 /* Find first overlaping mapping */
603 vma = find_vma_intersection(mm, addr, end);
604 if (!vma)
605 return 0;
606
607 nr_pages = (min(end, vma->vm_end) -
608 max(addr, vma->vm_start)) >> PAGE_SHIFT;
609
610 /* Iterate over the rest of the overlaps */
611 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
612 unsigned long overlap_len;
613
614 if (vma->vm_start > end)
615 break;
616
617 overlap_len = min(end, vma->vm_end) - vma->vm_start;
618 nr_pages += overlap_len >> PAGE_SHIFT;
619 }
620
621 return nr_pages;
622}
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
625 struct rb_node **rb_link, struct rb_node *rb_parent)
626{
Michel Lespinassed3737182012-12-11 16:01:38 -0800627 /* Update tracking information for the gap following the new vma. */
628 if (vma->vm_next)
629 vma_gap_update(vma->vm_next);
630 else
631 mm->highest_vm_end = vma->vm_end;
632
633 /*
634 * vma->vm_prev wasn't known when we followed the rbtree to find the
635 * correct insertion point for that vma. As a result, we could not
636 * update the vma vm_rb parents rb_subtree_gap values on the way down.
637 * So, we first insert the vma with a zero rb_subtree_gap value
638 * (to be consistent with what we did on the way down), and then
639 * immediately update the gap to the correct value. Finally we
640 * rebalance the rbtree after all augmented values have been set.
641 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800643 vma->rb_subtree_gap = 0;
644 vma_gap_update(vma);
645 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700648static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
ZhenwenXu48aae422009-01-06 14:40:21 -0800650 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
652 file = vma->vm_file;
653 if (file) {
654 struct address_space *mapping = file->f_mapping;
655
656 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500657 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700659 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800662 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 flush_dcache_mmap_unlock(mapping);
664 }
665}
666
667static void
668__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
669 struct vm_area_struct *prev, struct rb_node **rb_link,
670 struct rb_node *rb_parent)
671{
672 __vma_link_list(mm, vma, prev, rb_parent);
673 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
677 struct vm_area_struct *prev, struct rb_node **rb_link,
678 struct rb_node *rb_parent)
679{
680 struct address_space *mapping = NULL;
681
Huang Shijie64ac4942014-06-04 16:07:33 -0700682 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800684 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
687 __vma_link(mm, vma, prev, rb_link, rb_parent);
688 __vma_link_file(vma);
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800691 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 mm->map_count++;
694 validate_mm(mm);
695}
696
697/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700698 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700699 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800701static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
Hugh Dickins6597d782012-10-08 16:29:07 -0700703 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800704 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Hugh Dickins6597d782012-10-08 16:29:07 -0700706 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
707 &prev, &rb_link, &rb_parent))
708 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 __vma_link(mm, vma, prev, rb_link, rb_parent);
710 mm->map_count++;
711}
712
713static inline void
714__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
715 struct vm_area_struct *prev)
716{
Michel Lespinassed3737182012-12-11 16:01:38 -0800717 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700718
Michel Lespinassed3737182012-12-11 16:01:38 -0800719 vma_rb_erase(vma, &mm->mm_rb);
720 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700721 if (next)
722 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700723
724 /* Kill the cache */
725 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
728/*
729 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
730 * is already present in an i_mmap tree without adjusting the tree.
731 * The following helper function should be used when such adjustments
732 * are necessary. The "insert" vma (if any) is to be inserted
733 * before we drop the necessary locks.
734 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800735int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
737{
738 struct mm_struct *mm = vma->vm_mm;
739 struct vm_area_struct *next = vma->vm_next;
740 struct vm_area_struct *importer = NULL;
741 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700742 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700743 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800745 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 long adjust_next = 0;
747 int remove_next = 0;
748
749 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700750 struct vm_area_struct *exporter = NULL;
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 if (end >= next->vm_end) {
753 /*
754 * vma expands, overlapping all the next, and
755 * perhaps the one after too (mprotect case 6).
756 */
757again: remove_next = 1 + (end > next->vm_end);
758 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700759 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 importer = vma;
761 } else if (end > next->vm_start) {
762 /*
763 * vma expands, overlapping part of the next:
764 * mprotect case 5 shifting the boundary up.
765 */
766 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700767 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 importer = vma;
769 } else if (end < vma->vm_end) {
770 /*
771 * vma shrinks, and !insert tells it's not
772 * split_vma inserting another: so it must be
773 * mprotect case 4 shifting the boundary down.
774 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700775 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700776 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 importer = next;
778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
Rik van Riel5beb4932010-03-05 13:42:07 -0800780 /*
781 * Easily overlooked: when mprotect shifts the boundary,
782 * make sure the expanding vma has anon_vma set if the
783 * shrinking vma had, to cover any anon pages imported.
784 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700785 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800786 int error;
787
Linus Torvalds287d97a2010-04-10 15:22:30 -0700788 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300789 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700790 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300791 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800792 }
793 }
794
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 if (file) {
796 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800797 root = &mapping->i_mmap;
798 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530799
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800800 if (adjust_next)
801 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530802
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800803 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700806 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 * are visible to arm/parisc __flush_dcache_page
808 * throughout; but we cannot insert into address
809 * space until vma start or end is updated.
810 */
811 __vma_link_file(insert);
812 }
813 }
814
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800815 vma_adjust_trans_huge(vma, start, end, adjust_next);
816
Michel Lespinassebf181b92012-10-08 16:31:39 -0700817 anon_vma = vma->anon_vma;
818 if (!anon_vma && adjust_next)
819 anon_vma = next->anon_vma;
820 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700821 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
822 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000823 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700824 anon_vma_interval_tree_pre_update_vma(vma);
825 if (adjust_next)
826 anon_vma_interval_tree_pre_update_vma(next);
827 }
Rik van Riel012f18002010-08-09 17:18:40 -0700828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 if (root) {
830 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700831 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700833 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
835
Michel Lespinassed3737182012-12-11 16:01:38 -0800836 if (start != vma->vm_start) {
837 vma->vm_start = start;
838 start_changed = true;
839 }
840 if (end != vma->vm_end) {
841 vma->vm_end = end;
842 end_changed = true;
843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 vma->vm_pgoff = pgoff;
845 if (adjust_next) {
846 next->vm_start += adjust_next << PAGE_SHIFT;
847 next->vm_pgoff += adjust_next;
848 }
849
850 if (root) {
851 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700852 vma_interval_tree_insert(next, root);
853 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 flush_dcache_mmap_unlock(mapping);
855 }
856
857 if (remove_next) {
858 /*
859 * vma_merge has merged next into vma, and needs
860 * us to remove next before dropping the locks.
861 */
862 __vma_unlink(mm, next, vma);
863 if (file)
864 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 } else if (insert) {
866 /*
867 * split_vma has split insert from vma, and needs
868 * us to insert it before dropping the locks
869 * (it may either follow vma or precede it).
870 */
871 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800872 } else {
873 if (start_changed)
874 vma_gap_update(vma);
875 if (end_changed) {
876 if (!next)
877 mm->highest_vm_end = end;
878 else if (!adjust_next)
879 vma_gap_update(next);
880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 }
882
Michel Lespinassebf181b92012-10-08 16:31:39 -0700883 if (anon_vma) {
884 anon_vma_interval_tree_post_update_vma(vma);
885 if (adjust_next)
886 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800887 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700888 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800890 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530892 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100893 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530894
895 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100896 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530897 }
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700900 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530901 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700903 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800904 if (next->anon_vma)
905 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700907 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 kmem_cache_free(vm_area_cachep, next);
909 /*
910 * In mprotect's case 6 (see comments on vma_merge),
911 * we must remove another next too. It would clutter
912 * up the code too much to do both in one go.
913 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800914 next = vma->vm_next;
915 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800917 else if (next)
918 vma_gap_update(next);
919 else
920 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530922 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100923 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800926
927 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
930/*
931 * If the vma has a ->close operation then the driver probably needs to release
932 * per-vma resources, so we don't attempt to merge those.
933 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700935 struct file *file, unsigned long vm_flags,
936 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800938 /*
939 * VM_SOFTDIRTY should not prevent from VMA merging, if we
940 * match the flags but dirty bit -- the caller should mark
941 * merged VMA as dirty. If dirty bit won't be excluded from
942 * comparison, we increase pressue on the memory system forcing
943 * the kernel to generate new VMAs when old one could be
944 * extended instead.
945 */
946 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 return 0;
948 if (vma->vm_file != file)
949 return 0;
950 if (vma->vm_ops && vma->vm_ops->close)
951 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700952 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
953 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 return 1;
955}
956
957static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700958 struct anon_vma *anon_vma2,
959 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960{
Shaohua Li965f55d2011-05-24 17:11:20 -0700961 /*
962 * The list_is_singular() test is to avoid merging VMA cloned from
963 * parents. This can improve scalability caused by anon_vma lock.
964 */
965 if ((!anon_vma1 || !anon_vma2) && (!vma ||
966 list_is_singular(&vma->anon_vma_chain)))
967 return 1;
968 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969}
970
971/*
972 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
973 * in front of (at a lower virtual address and file offset than) the vma.
974 *
975 * We cannot merge two vmas if they have differently assigned (non-NULL)
976 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
977 *
978 * We don't check here for the merged mmap wrapping around the end of pagecache
979 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
980 * wrap, nor mmaps which cover the final page at index -1UL.
981 */
982static int
983can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700984 struct anon_vma *anon_vma, struct file *file,
985 pgoff_t vm_pgoff,
986 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700988 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700989 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 if (vma->vm_pgoff == vm_pgoff)
991 return 1;
992 }
993 return 0;
994}
995
996/*
997 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
998 * beyond (at a higher virtual address and file offset than) the vma.
999 *
1000 * We cannot merge two vmas if they have differently assigned (non-NULL)
1001 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1002 */
1003static int
1004can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001005 struct anon_vma *anon_vma, struct file *file,
1006 pgoff_t vm_pgoff,
1007 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001009 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001010 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001012 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1014 return 1;
1015 }
1016 return 0;
1017}
1018
1019/*
1020 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1021 * whether that can be merged with its predecessor or its successor.
1022 * Or both (it neatly fills a hole).
1023 *
1024 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1025 * certain not to be mapped by the time vma_merge is called; but when
1026 * called for mprotect, it is certain to be already mapped (either at
1027 * an offset within prev, or at the start of next), and the flags of
1028 * this area are about to be changed to vm_flags - and the no-change
1029 * case has already been eliminated.
1030 *
1031 * The following mprotect cases have to be considered, where AAAA is
1032 * the area passed down from mprotect_fixup, never extending beyond one
1033 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1034 *
1035 * AAAA AAAA AAAA AAAA
1036 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1037 * cannot merge might become might become might become
1038 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1039 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1040 * mremap move: PPPPNNNNNNNN 8
1041 * AAAA
1042 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1043 * might become case 1 below case 2 below case 3 below
1044 *
1045 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1046 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1047 */
1048struct vm_area_struct *vma_merge(struct mm_struct *mm,
1049 struct vm_area_struct *prev, unsigned long addr,
1050 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001051 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001052 pgoff_t pgoff, struct mempolicy *policy,
1053 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
1055 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1056 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001057 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
1059 /*
1060 * We later require that vma->vm_flags == vm_flags,
1061 * so this tests vma->vm_flags & VM_SPECIAL, too.
1062 */
1063 if (vm_flags & VM_SPECIAL)
1064 return NULL;
1065
1066 if (prev)
1067 next = prev->vm_next;
1068 else
1069 next = mm->mmap;
1070 area = next;
1071 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1072 next = next->vm_next;
1073
1074 /*
1075 * Can it merge with the predecessor?
1076 */
1077 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001078 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001080 anon_vma, file, pgoff,
1081 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 /*
1083 * OK, it can. Can we now merge in the successor as well?
1084 */
1085 if (next && end == next->vm_start &&
1086 mpol_equal(policy, vma_policy(next)) &&
1087 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001088 anon_vma, file,
1089 pgoff+pglen,
1090 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001092 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001094 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 next->vm_end, prev->vm_pgoff, NULL);
1096 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001097 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001099 if (err)
1100 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001101 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return prev;
1103 }
1104
1105 /*
1106 * Can this new request be merged in front of next?
1107 */
1108 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001109 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001111 anon_vma, file, pgoff+pglen,
1112 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001114 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 addr, prev->vm_pgoff, NULL);
1116 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001117 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001119 if (err)
1120 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001121 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 return area;
1123 }
1124
1125 return NULL;
1126}
1127
1128/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001129 * Rough compatbility check to quickly see if it's even worth looking
1130 * at sharing an anon_vma.
1131 *
1132 * They need to have the same vm_file, and the flags can only differ
1133 * in things that mprotect may change.
1134 *
1135 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1136 * we can merge the two vma's. For example, we refuse to merge a vma if
1137 * there is a vm_ops->close() function, because that indicates that the
1138 * driver is doing some kind of reference counting. But that doesn't
1139 * really matter for the anon_vma sharing case.
1140 */
1141static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1142{
1143 return a->vm_end == b->vm_start &&
1144 mpol_equal(vma_policy(a), vma_policy(b)) &&
1145 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001146 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001147 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1148}
1149
1150/*
1151 * Do some basic sanity checking to see if we can re-use the anon_vma
1152 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1153 * the same as 'old', the other will be the new one that is trying
1154 * to share the anon_vma.
1155 *
1156 * NOTE! This runs with mm_sem held for reading, so it is possible that
1157 * the anon_vma of 'old' is concurrently in the process of being set up
1158 * by another page fault trying to merge _that_. But that's ok: if it
1159 * is being set up, that automatically means that it will be a singleton
1160 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001161 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001162 *
1163 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1164 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1165 * is to return an anon_vma that is "complex" due to having gone through
1166 * a fork).
1167 *
1168 * We also make sure that the two vma's are compatible (adjacent,
1169 * and with the same memory policies). That's all stable, even with just
1170 * a read lock on the mm_sem.
1171 */
1172static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1173{
1174 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001175 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001176
1177 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1178 return anon_vma;
1179 }
1180 return NULL;
1181}
1182
1183/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1185 * neighbouring vmas for a suitable anon_vma, before it goes off
1186 * to allocate a new anon_vma. It checks because a repetitive
1187 * sequence of mprotects and faults may otherwise lead to distinct
1188 * anon_vmas being allocated, preventing vma merge in subsequent
1189 * mprotect.
1190 */
1191struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1192{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001193 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196 near = vma->vm_next;
1197 if (!near)
1198 goto try_prev;
1199
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001200 anon_vma = reusable_anon_vma(near, vma, near);
1201 if (anon_vma)
1202 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001204 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 if (!near)
1206 goto none;
1207
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001208 anon_vma = reusable_anon_vma(near, near, vma);
1209 if (anon_vma)
1210 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211none:
1212 /*
1213 * There's no absolute need to look only at touching neighbours:
1214 * we could search further afield for "compatible" anon_vmas.
1215 * But it would probably just be a waste of time searching,
1216 * or lead to too many vmas hanging off the same anon_vma.
1217 * We're trying to allow mprotect remerging later on,
1218 * not trying to minimize memory used for anon_vmas.
1219 */
1220 return NULL;
1221}
1222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223/*
Al Viro40401532012-02-13 03:58:52 +00001224 * If a hint addr is less than mmap_min_addr change hint to be as
1225 * low as possible but still greater than mmap_min_addr
1226 */
1227static inline unsigned long round_hint_to_min(unsigned long hint)
1228{
1229 hint &= PAGE_MASK;
1230 if (((void *)hint != NULL) &&
1231 (hint < mmap_min_addr))
1232 return PAGE_ALIGN(mmap_min_addr);
1233 return hint;
1234}
1235
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001236static inline int mlock_future_check(struct mm_struct *mm,
1237 unsigned long flags,
1238 unsigned long len)
1239{
1240 unsigned long locked, lock_limit;
1241
1242 /* mlock MCL_FUTURE? */
1243 if (flags & VM_LOCKED) {
1244 locked = len >> PAGE_SHIFT;
1245 locked += mm->locked_vm;
1246 lock_limit = rlimit(RLIMIT_MEMLOCK);
1247 lock_limit >>= PAGE_SHIFT;
1248 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1249 return -EAGAIN;
1250 }
1251 return 0;
1252}
1253
Al Viro40401532012-02-13 03:58:52 +00001254/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001255 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001257unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001259 unsigned long flags, vm_flags_t vm_flags,
1260 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
vishnu.pscc71aba2014-10-09 15:26:29 -07001262 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
Michel Lespinasse41badc12013-02-22 16:32:47 -08001264 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001265
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001266 if (!len)
1267 return -EINVAL;
1268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 /*
1270 * Does the application expect PROT_READ to imply PROT_EXEC?
1271 *
1272 * (the exception is when the underlying filesystem is noexec
1273 * mounted, in which case we dont add PROT_EXEC.)
1274 */
1275 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001276 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 prot |= PROT_EXEC;
1278
Eric Paris7cd94142007-11-26 18:47:40 -05001279 if (!(flags & MAP_FIXED))
1280 addr = round_hint_to_min(addr);
1281
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 /* Careful about overflows.. */
1283 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001284 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 return -ENOMEM;
1286
1287 /* offset overflow? */
1288 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001289 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 /* Too many mappings? */
1292 if (mm->map_count > sysctl_max_map_count)
1293 return -ENOMEM;
1294
1295 /* Obtain the address to map to. we verify (or select) it and ensure
1296 * that it represents a valid section of the address space.
1297 */
1298 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001299 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 return addr;
1301
1302 /* Do simple checking here so the lower-level routines won't have
1303 * to. we assume access permissions have been handled by the open
1304 * of the memory object, so we don't do any here.
1305 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001306 vm_flags |= calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1308
Huang Shijiecdf7b342009-09-21 17:03:36 -07001309 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 if (!can_do_mlock())
1311 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001312
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001313 if (mlock_future_check(mm, vm_flags, len))
1314 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001317 struct inode *inode = file_inode(file);
1318
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 switch (flags & MAP_TYPE) {
1320 case MAP_SHARED:
1321 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1322 return -EACCES;
1323
1324 /*
1325 * Make sure we don't allow writing to an append-only
1326 * file..
1327 */
1328 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1329 return -EACCES;
1330
1331 /*
1332 * Make sure there are no mandatory locks on the file.
1333 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001334 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 return -EAGAIN;
1336
1337 vm_flags |= VM_SHARED | VM_MAYSHARE;
1338 if (!(file->f_mode & FMODE_WRITE))
1339 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1340
1341 /* fall through */
1342 case MAP_PRIVATE:
1343 if (!(file->f_mode & FMODE_READ))
1344 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001345 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001346 if (vm_flags & VM_EXEC)
1347 return -EPERM;
1348 vm_flags &= ~VM_MAYEXEC;
1349 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001350
Al Viro72c2d532013-09-22 16:27:52 -04001351 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001352 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001353 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1354 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 break;
1356
1357 default:
1358 return -EINVAL;
1359 }
1360 } else {
1361 switch (flags & MAP_TYPE) {
1362 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001363 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1364 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001365 /*
1366 * Ignore pgoff.
1367 */
1368 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 vm_flags |= VM_SHARED | VM_MAYSHARE;
1370 break;
1371 case MAP_PRIVATE:
1372 /*
1373 * Set pgoff according to addr for anon_vma.
1374 */
1375 pgoff = addr >> PAGE_SHIFT;
1376 break;
1377 default:
1378 return -EINVAL;
1379 }
1380 }
1381
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001382 /*
1383 * Set 'VM_NORESERVE' if we should not account for the
1384 * memory use of this mapping.
1385 */
1386 if (flags & MAP_NORESERVE) {
1387 /* We honor MAP_NORESERVE if allowed to overcommit */
1388 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1389 vm_flags |= VM_NORESERVE;
1390
1391 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1392 if (file && is_file_hugepages(file))
1393 vm_flags |= VM_NORESERVE;
1394 }
1395
1396 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001397 if (!IS_ERR_VALUE(addr) &&
1398 ((vm_flags & VM_LOCKED) ||
1399 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001400 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001401 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001402}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001403
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001404SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1405 unsigned long, prot, unsigned long, flags,
1406 unsigned long, fd, unsigned long, pgoff)
1407{
1408 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001409 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001410
1411 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001412 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001413 file = fget(fd);
1414 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001415 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001416 if (is_file_hugepages(file))
1417 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001418 retval = -EINVAL;
1419 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1420 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001421 } else if (flags & MAP_HUGETLB) {
1422 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001423 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001424
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001425 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001426 if (!hs)
1427 return -EINVAL;
1428
1429 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001430 /*
1431 * VM_NORESERVE is used because the reservations will be
1432 * taken when vm_ops->mmap() is called
1433 * A dummy user value is used because we are not locking
1434 * memory so no accounting is necessary
1435 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001436 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001437 VM_NORESERVE,
1438 &user, HUGETLB_ANONHUGE_INODE,
1439 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001440 if (IS_ERR(file))
1441 return PTR_ERR(file);
1442 }
1443
1444 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1445
Al Viroeb36c582012-05-30 20:17:35 -04001446 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001447out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001448 if (file)
1449 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001450 return retval;
1451}
1452
Christoph Hellwiga4679372010-03-10 15:21:15 -08001453#ifdef __ARCH_WANT_SYS_OLD_MMAP
1454struct mmap_arg_struct {
1455 unsigned long addr;
1456 unsigned long len;
1457 unsigned long prot;
1458 unsigned long flags;
1459 unsigned long fd;
1460 unsigned long offset;
1461};
1462
1463SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1464{
1465 struct mmap_arg_struct a;
1466
1467 if (copy_from_user(&a, arg, sizeof(a)))
1468 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001469 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001470 return -EINVAL;
1471
1472 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1473 a.offset >> PAGE_SHIFT);
1474}
1475#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1476
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001477/*
1478 * Some shared mappigns will want the pages marked read-only
1479 * to track write events. If so, we'll downgrade vm_page_prot
1480 * to the private version (using protection_map[] without the
1481 * VM_SHARED bit).
1482 */
1483int vma_wants_writenotify(struct vm_area_struct *vma)
1484{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001485 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001486 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001487
1488 /* If it was private or non-writable, the write bit is already clear */
1489 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1490 return 0;
1491
1492 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001493 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001494 return 1;
1495
Peter Feiner64e45502014-10-13 15:55:46 -07001496 /* The open routine did something to the protections that pgprot_modify
1497 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001498 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001499 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001500 return 0;
1501
Peter Feiner64e45502014-10-13 15:55:46 -07001502 /* Do we need to track softdirty? */
1503 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1504 return 1;
1505
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001506 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001507 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001508 return 0;
1509
1510 /* Can the mapping track the dirty pages? */
1511 return vma->vm_file && vma->vm_file->f_mapping &&
1512 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1513}
1514
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001515/*
1516 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001517 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001518 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001519static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001520{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001521 /*
1522 * hugetlb has its own accounting separate from the core VM
1523 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1524 */
1525 if (file && is_file_hugepages(file))
1526 return 0;
1527
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001528 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1529}
1530
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001531unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001532 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001533{
1534 struct mm_struct *mm = current->mm;
1535 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001536 int error;
1537 struct rb_node **rb_link, *rb_parent;
1538 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001539
Cyril Hrubise8420a82013-04-29 15:08:33 -07001540 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001541 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001542 unsigned long nr_pages;
1543
1544 /*
1545 * MAP_FIXED may remove pages of mappings that intersects with
1546 * requested mapping. Account for the pages it would unmap.
1547 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001548 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1549
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001550 if (!may_expand_vm(mm, vm_flags,
1551 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001552 return -ENOMEM;
1553 }
1554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001556 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1557 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (do_munmap(mm, addr, len))
1559 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 }
1561
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001562 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001563 * Private writable mapping: check memory availability
1564 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001565 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001566 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001567 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001568 return -ENOMEM;
1569 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
1571
1572 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001573 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001575 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1576 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001577 if (vma)
1578 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 /*
1581 * Determine the object being mapped and call the appropriate
1582 * specific mapper. the address has already been validated, but
1583 * not unmapped, but the maps are removed from the list.
1584 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001585 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 if (!vma) {
1587 error = -ENOMEM;
1588 goto unacct_error;
1589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 vma->vm_mm = mm;
1592 vma->vm_start = addr;
1593 vma->vm_end = addr + len;
1594 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001595 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001597 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
1599 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if (vm_flags & VM_DENYWRITE) {
1601 error = deny_write_access(file);
1602 if (error)
1603 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001605 if (vm_flags & VM_SHARED) {
1606 error = mapping_map_writable(file->f_mapping);
1607 if (error)
1608 goto allow_write_and_free_vma;
1609 }
1610
1611 /* ->mmap() can change vma->vm_file, but must guarantee that
1612 * vma_link() below can deny write-access if VM_DENYWRITE is set
1613 * and map writably if VM_SHARED is set. This usually means the
1614 * new file must not have been exposed to user-space, yet.
1615 */
Al Virocb0942b2012-08-27 14:48:26 -04001616 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 error = file->f_op->mmap(file, vma);
1618 if (error)
1619 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001620
1621 /* Can addr have changed??
1622 *
1623 * Answer: Yes, several device drivers can do it in their
1624 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001625 * Bug: If addr is changed, prev, rb_link, rb_parent should
1626 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001627 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001628 WARN_ON_ONCE(addr != vma->vm_start);
1629
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001630 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001631 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 } else if (vm_flags & VM_SHARED) {
1633 error = shmem_zero_setup(vma);
1634 if (error)
1635 goto free_vma;
1636 }
1637
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001638 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001639 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001640 if (file) {
1641 if (vm_flags & VM_SHARED)
1642 mapping_unmap_writable(file->f_mapping);
1643 if (vm_flags & VM_DENYWRITE)
1644 allow_write_access(file);
1645 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001646 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001647out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001648 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001649
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001650 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001652 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1653 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001654 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001655 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001656 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001657 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301658
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001659 if (file)
1660 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301661
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001662 /*
1663 * New (or expanded) vma always get soft dirty status.
1664 * Otherwise user-space soft-dirty page tracker won't
1665 * be able to distinguish situation when vma area unmapped,
1666 * then new mapped in-place (which must be aimed as
1667 * a completely new data area).
1668 */
1669 vma->vm_flags |= VM_SOFTDIRTY;
1670
Peter Feiner64e45502014-10-13 15:55:46 -07001671 vma_set_page_prot(vma);
1672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return addr;
1674
1675unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 vma->vm_file = NULL;
1677 fput(file);
1678
1679 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001680 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1681 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001682 if (vm_flags & VM_SHARED)
1683 mapping_unmap_writable(file->f_mapping);
1684allow_write_and_free_vma:
1685 if (vm_flags & VM_DENYWRITE)
1686 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687free_vma:
1688 kmem_cache_free(vm_area_cachep, vma);
1689unacct_error:
1690 if (charged)
1691 vm_unacct_memory(charged);
1692 return error;
1693}
1694
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001695unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1696{
1697 /*
1698 * We implement the search by looking for an rbtree node that
1699 * immediately follows a suitable gap. That is,
1700 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1701 * - gap_end = vma->vm_start >= info->low_limit + length;
1702 * - gap_end - gap_start >= length
1703 */
1704
1705 struct mm_struct *mm = current->mm;
1706 struct vm_area_struct *vma;
1707 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1708
1709 /* Adjust search length to account for worst case alignment overhead */
1710 length = info->length + info->align_mask;
1711 if (length < info->length)
1712 return -ENOMEM;
1713
1714 /* Adjust search limits by the desired length */
1715 if (info->high_limit < length)
1716 return -ENOMEM;
1717 high_limit = info->high_limit - length;
1718
1719 if (info->low_limit > high_limit)
1720 return -ENOMEM;
1721 low_limit = info->low_limit + length;
1722
1723 /* Check if rbtree root looks promising */
1724 if (RB_EMPTY_ROOT(&mm->mm_rb))
1725 goto check_highest;
1726 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1727 if (vma->rb_subtree_gap < length)
1728 goto check_highest;
1729
1730 while (true) {
1731 /* Visit left subtree if it looks promising */
1732 gap_end = vma->vm_start;
1733 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1734 struct vm_area_struct *left =
1735 rb_entry(vma->vm_rb.rb_left,
1736 struct vm_area_struct, vm_rb);
1737 if (left->rb_subtree_gap >= length) {
1738 vma = left;
1739 continue;
1740 }
1741 }
1742
1743 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1744check_current:
1745 /* Check if current node has a suitable gap */
1746 if (gap_start > high_limit)
1747 return -ENOMEM;
1748 if (gap_end >= low_limit && gap_end - gap_start >= length)
1749 goto found;
1750
1751 /* Visit right subtree if it looks promising */
1752 if (vma->vm_rb.rb_right) {
1753 struct vm_area_struct *right =
1754 rb_entry(vma->vm_rb.rb_right,
1755 struct vm_area_struct, vm_rb);
1756 if (right->rb_subtree_gap >= length) {
1757 vma = right;
1758 continue;
1759 }
1760 }
1761
1762 /* Go back up the rbtree to find next candidate node */
1763 while (true) {
1764 struct rb_node *prev = &vma->vm_rb;
1765 if (!rb_parent(prev))
1766 goto check_highest;
1767 vma = rb_entry(rb_parent(prev),
1768 struct vm_area_struct, vm_rb);
1769 if (prev == vma->vm_rb.rb_left) {
1770 gap_start = vma->vm_prev->vm_end;
1771 gap_end = vma->vm_start;
1772 goto check_current;
1773 }
1774 }
1775 }
1776
1777check_highest:
1778 /* Check highest gap, which does not precede any rbtree node */
1779 gap_start = mm->highest_vm_end;
1780 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1781 if (gap_start > high_limit)
1782 return -ENOMEM;
1783
1784found:
1785 /* We found a suitable gap. Clip it with the original low_limit. */
1786 if (gap_start < info->low_limit)
1787 gap_start = info->low_limit;
1788
1789 /* Adjust gap address to the desired alignment */
1790 gap_start += (info->align_offset - gap_start) & info->align_mask;
1791
1792 VM_BUG_ON(gap_start + info->length > info->high_limit);
1793 VM_BUG_ON(gap_start + info->length > gap_end);
1794 return gap_start;
1795}
1796
1797unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1798{
1799 struct mm_struct *mm = current->mm;
1800 struct vm_area_struct *vma;
1801 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1802
1803 /* Adjust search length to account for worst case alignment overhead */
1804 length = info->length + info->align_mask;
1805 if (length < info->length)
1806 return -ENOMEM;
1807
1808 /*
1809 * Adjust search limits by the desired length.
1810 * See implementation comment at top of unmapped_area().
1811 */
1812 gap_end = info->high_limit;
1813 if (gap_end < length)
1814 return -ENOMEM;
1815 high_limit = gap_end - length;
1816
1817 if (info->low_limit > high_limit)
1818 return -ENOMEM;
1819 low_limit = info->low_limit + length;
1820
1821 /* Check highest gap, which does not precede any rbtree node */
1822 gap_start = mm->highest_vm_end;
1823 if (gap_start <= high_limit)
1824 goto found_highest;
1825
1826 /* Check if rbtree root looks promising */
1827 if (RB_EMPTY_ROOT(&mm->mm_rb))
1828 return -ENOMEM;
1829 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1830 if (vma->rb_subtree_gap < length)
1831 return -ENOMEM;
1832
1833 while (true) {
1834 /* Visit right subtree if it looks promising */
1835 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1836 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1837 struct vm_area_struct *right =
1838 rb_entry(vma->vm_rb.rb_right,
1839 struct vm_area_struct, vm_rb);
1840 if (right->rb_subtree_gap >= length) {
1841 vma = right;
1842 continue;
1843 }
1844 }
1845
1846check_current:
1847 /* Check if current node has a suitable gap */
1848 gap_end = vma->vm_start;
1849 if (gap_end < low_limit)
1850 return -ENOMEM;
1851 if (gap_start <= high_limit && gap_end - gap_start >= length)
1852 goto found;
1853
1854 /* Visit left subtree if it looks promising */
1855 if (vma->vm_rb.rb_left) {
1856 struct vm_area_struct *left =
1857 rb_entry(vma->vm_rb.rb_left,
1858 struct vm_area_struct, vm_rb);
1859 if (left->rb_subtree_gap >= length) {
1860 vma = left;
1861 continue;
1862 }
1863 }
1864
1865 /* Go back up the rbtree to find next candidate node */
1866 while (true) {
1867 struct rb_node *prev = &vma->vm_rb;
1868 if (!rb_parent(prev))
1869 return -ENOMEM;
1870 vma = rb_entry(rb_parent(prev),
1871 struct vm_area_struct, vm_rb);
1872 if (prev == vma->vm_rb.rb_right) {
1873 gap_start = vma->vm_prev ?
1874 vma->vm_prev->vm_end : 0;
1875 goto check_current;
1876 }
1877 }
1878 }
1879
1880found:
1881 /* We found a suitable gap. Clip it with the original high_limit. */
1882 if (gap_end > info->high_limit)
1883 gap_end = info->high_limit;
1884
1885found_highest:
1886 /* Compute highest gap address at the desired alignment */
1887 gap_end -= info->length;
1888 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1889
1890 VM_BUG_ON(gap_end < info->low_limit);
1891 VM_BUG_ON(gap_end < gap_start);
1892 return gap_end;
1893}
1894
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895/* Get an address range which is currently unmapped.
1896 * For shmat() with addr=0.
1897 *
1898 * Ugly calling convention alert:
1899 * Return value with the low bits set means error value,
1900 * ie
1901 * if (ret & ~PAGE_MASK)
1902 * error = ret;
1903 *
1904 * This function "knows" that -ENOMEM has the bits set.
1905 */
1906#ifndef HAVE_ARCH_UNMAPPED_AREA
1907unsigned long
1908arch_get_unmapped_area(struct file *filp, unsigned long addr,
1909 unsigned long len, unsigned long pgoff, unsigned long flags)
1910{
1911 struct mm_struct *mm = current->mm;
1912 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001913 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001915 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 return -ENOMEM;
1917
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001918 if (flags & MAP_FIXED)
1919 return addr;
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 if (addr) {
1922 addr = PAGE_ALIGN(addr);
1923 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001924 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 (!vma || addr + len <= vma->vm_start))
1926 return addr;
1927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001929 info.flags = 0;
1930 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001931 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001932 info.high_limit = TASK_SIZE;
1933 info.align_mask = 0;
1934 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935}
vishnu.pscc71aba2014-10-09 15:26:29 -07001936#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938/*
1939 * This mmap-allocator allocates new areas top-down from below the
1940 * stack's low limit (the base):
1941 */
1942#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1943unsigned long
1944arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1945 const unsigned long len, const unsigned long pgoff,
1946 const unsigned long flags)
1947{
1948 struct vm_area_struct *vma;
1949 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001950 unsigned long addr = addr0;
1951 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001954 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return -ENOMEM;
1956
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001957 if (flags & MAP_FIXED)
1958 return addr;
1959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 /* requesting a specific address */
1961 if (addr) {
1962 addr = PAGE_ALIGN(addr);
1963 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001964 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 (!vma || addr + len <= vma->vm_start))
1966 return addr;
1967 }
1968
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001969 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1970 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001971 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001972 info.high_limit = mm->mmap_base;
1973 info.align_mask = 0;
1974 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001975
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 /*
1977 * A failed mmap() very likely causes application failure,
1978 * so fall back to the bottom-up function here. This scenario
1979 * can happen with large stack limits and large mmap()
1980 * allocations.
1981 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001982 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001983 VM_BUG_ON(addr != -ENOMEM);
1984 info.flags = 0;
1985 info.low_limit = TASK_UNMAPPED_BASE;
1986 info.high_limit = TASK_SIZE;
1987 addr = vm_unmapped_area(&info);
1988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
1990 return addr;
1991}
1992#endif
1993
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994unsigned long
1995get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1996 unsigned long pgoff, unsigned long flags)
1997{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001998 unsigned long (*get_area)(struct file *, unsigned long,
1999 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
Al Viro9206de92009-12-03 15:23:11 -05002001 unsigned long error = arch_mmap_check(addr, len, flags);
2002 if (error)
2003 return error;
2004
2005 /* Careful about overflows.. */
2006 if (len > TASK_SIZE)
2007 return -ENOMEM;
2008
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002009 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04002010 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002011 get_area = file->f_op->get_unmapped_area;
2012 addr = get_area(file, addr, len, pgoff, flags);
2013 if (IS_ERR_VALUE(addr))
2014 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002016 if (addr > TASK_SIZE - len)
2017 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002018 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002019 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002020
Al Viro9ac4ed42012-05-30 17:13:15 -04002021 addr = arch_rebalance_pgtables(addr, len);
2022 error = security_mmap_addr(addr);
2023 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024}
2025
2026EXPORT_SYMBOL(get_unmapped_area);
2027
2028/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002029struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002031 struct rb_node *rb_node;
2032 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002034 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002035 vma = vmacache_find(mm, addr);
2036 if (likely(vma))
2037 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002039 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002041 while (rb_node) {
2042 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002044 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002045
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002046 if (tmp->vm_end > addr) {
2047 vma = tmp;
2048 if (tmp->vm_start <= addr)
2049 break;
2050 rb_node = rb_node->rb_left;
2051 } else
2052 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002054
2055 if (vma)
2056 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 return vma;
2058}
2059
2060EXPORT_SYMBOL(find_vma);
2061
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002062/*
2063 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002064 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065struct vm_area_struct *
2066find_vma_prev(struct mm_struct *mm, unsigned long addr,
2067 struct vm_area_struct **pprev)
2068{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002069 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002071 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002072 if (vma) {
2073 *pprev = vma->vm_prev;
2074 } else {
2075 struct rb_node *rb_node = mm->mm_rb.rb_node;
2076 *pprev = NULL;
2077 while (rb_node) {
2078 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2079 rb_node = rb_node->rb_right;
2080 }
2081 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002082 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083}
2084
2085/*
2086 * Verify that the stack growth is acceptable and
2087 * update accounting. This is shared with both the
2088 * grow-up and grow-down cases.
2089 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002090static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091{
2092 struct mm_struct *mm = vma->vm_mm;
2093 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08002094 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
2096 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002097 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return -ENOMEM;
2099
2100 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08002101 actual_size = size;
2102 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2103 actual_size -= PAGE_SIZE;
Jason Low4db0c3c2015-04-15 16:14:08 -07002104 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 return -ENOMEM;
2106
2107 /* mlock limit tests */
2108 if (vma->vm_flags & VM_LOCKED) {
2109 unsigned long locked;
2110 unsigned long limit;
2111 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002112 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002113 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 if (locked > limit && !capable(CAP_IPC_LOCK))
2115 return -ENOMEM;
2116 }
2117
Adam Litke0d59a012007-01-30 14:35:39 -08002118 /* Check to ensure the stack will not grow into a hugetlb-only region */
2119 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2120 vma->vm_end - size;
2121 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2122 return -EFAULT;
2123
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 /*
2125 * Overcommit.. This must be the final test, as it will
2126 * update security statistics.
2127 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002128 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 return -ENOMEM;
2130
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 return 0;
2132}
2133
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002134#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002136 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2137 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002139int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140{
Oleg Nesterov09357812015-11-05 18:48:17 -08002141 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 int error;
2143
2144 if (!(vma->vm_flags & VM_GROWSUP))
2145 return -EFAULT;
2146
2147 /*
2148 * We must make sure the anon_vma is allocated
2149 * so that the anon_vma locking is not a noop.
2150 */
2151 if (unlikely(anon_vma_prepare(vma)))
2152 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002153 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
2155 /*
2156 * vma->vm_start/vm_end cannot change under us because the caller
2157 * is required to hold the mmap_sem in read mode. We need the
2158 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002159 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 */
Helge Deller06b32f32006-12-19 19:28:33 +01002161 if (address < PAGE_ALIGN(address+4))
2162 address = PAGE_ALIGN(address+4);
2163 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002164 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002165 return -ENOMEM;
2166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 error = 0;
2168
2169 /* Somebody else might have raced and expanded it already */
2170 if (address > vma->vm_end) {
2171 unsigned long size, grow;
2172
2173 size = address - vma->vm_start;
2174 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2175
Hugh Dickins42c36f62011-05-09 17:44:42 -07002176 error = -ENOMEM;
2177 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2178 error = acct_stack_growth(vma, size, grow);
2179 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002180 /*
2181 * vma_gap_update() doesn't support concurrent
2182 * updates, but we only hold a shared mmap_sem
2183 * lock here, so we need to protect against
2184 * concurrent vma expansions.
2185 * vma_lock_anon_vma() doesn't help here, as
2186 * we don't guarantee that all growable vmas
2187 * in a mm share the same root anon vma.
2188 * So, we reuse mm->page_table_lock to guard
2189 * against concurrent vma expansions.
2190 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002191 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002192 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002193 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002194 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002195 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002196 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002197 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002198 if (vma->vm_next)
2199 vma_gap_update(vma->vm_next);
2200 else
Oleg Nesterov09357812015-11-05 18:48:17 -08002201 mm->highest_vm_end = address;
2202 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002203
Hugh Dickins42c36f62011-05-09 17:44:42 -07002204 perf_event_mmap(vma);
2205 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002208 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002209 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002210 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 return error;
2212}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002213#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215/*
2216 * vma is the first one with address < vma->vm_start. Have to extend vma.
2217 */
Michal Hockod05f3162011-05-24 17:11:44 -07002218int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002219 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220{
Oleg Nesterov09357812015-11-05 18:48:17 -08002221 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 int error;
2223
2224 /*
2225 * We must make sure the anon_vma is allocated
2226 * so that the anon_vma locking is not a noop.
2227 */
2228 if (unlikely(anon_vma_prepare(vma)))
2229 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002230
2231 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002232 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002233 if (error)
2234 return error;
2235
Rik van Rielbb4a3402010-08-09 17:18:37 -07002236 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
2238 /*
2239 * vma->vm_start/vm_end cannot change under us because the caller
2240 * is required to hold the mmap_sem in read mode. We need the
2241 * anon_vma lock to serialize against concurrent expand_stacks.
2242 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
2244 /* Somebody else might have raced and expanded it already */
2245 if (address < vma->vm_start) {
2246 unsigned long size, grow;
2247
2248 size = vma->vm_end - address;
2249 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2250
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002251 error = -ENOMEM;
2252 if (grow <= vma->vm_pgoff) {
2253 error = acct_stack_growth(vma, size, grow);
2254 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002255 /*
2256 * vma_gap_update() doesn't support concurrent
2257 * updates, but we only hold a shared mmap_sem
2258 * lock here, so we need to protect against
2259 * concurrent vma expansions.
2260 * vma_lock_anon_vma() doesn't help here, as
2261 * we don't guarantee that all growable vmas
2262 * in a mm share the same root anon vma.
2263 * So, we reuse mm->page_table_lock to guard
2264 * against concurrent vma expansions.
2265 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002266 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002267 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002268 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002269 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002270 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002271 vma->vm_start = address;
2272 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002273 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002274 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002275 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002276
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002277 perf_event_mmap(vma);
2278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 }
2280 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002281 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002282 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002283 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 return error;
2285}
2286
Linus Torvalds09884962013-02-27 08:36:04 -08002287/*
2288 * Note how expand_stack() refuses to expand the stack all the way to
2289 * abut the next virtual mapping, *unless* that mapping itself is also
2290 * a stack mapping. We want to leave room for a guard page, after all
2291 * (the guard page itself is not added here, that is done by the
2292 * actual page faulting logic)
2293 *
2294 * This matches the behavior of the guard page logic (see mm/memory.c:
2295 * check_stack_guard_page()), which only allows the guard page to be
2296 * removed under these circumstances.
2297 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002298#ifdef CONFIG_STACK_GROWSUP
2299int expand_stack(struct vm_area_struct *vma, unsigned long address)
2300{
Linus Torvalds09884962013-02-27 08:36:04 -08002301 struct vm_area_struct *next;
2302
2303 address &= PAGE_MASK;
2304 next = vma->vm_next;
2305 if (next && next->vm_start == address + PAGE_SIZE) {
2306 if (!(next->vm_flags & VM_GROWSUP))
2307 return -ENOMEM;
2308 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002309 return expand_upwards(vma, address);
2310}
2311
2312struct vm_area_struct *
2313find_extend_vma(struct mm_struct *mm, unsigned long addr)
2314{
2315 struct vm_area_struct *vma, *prev;
2316
2317 addr &= PAGE_MASK;
2318 vma = find_vma_prev(mm, addr, &prev);
2319 if (vma && (vma->vm_start <= addr))
2320 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002321 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002322 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002323 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002324 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002325 return prev;
2326}
2327#else
2328int expand_stack(struct vm_area_struct *vma, unsigned long address)
2329{
Linus Torvalds09884962013-02-27 08:36:04 -08002330 struct vm_area_struct *prev;
2331
2332 address &= PAGE_MASK;
2333 prev = vma->vm_prev;
2334 if (prev && prev->vm_end == address) {
2335 if (!(prev->vm_flags & VM_GROWSDOWN))
2336 return -ENOMEM;
2337 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002338 return expand_downwards(vma, address);
2339}
2340
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002342find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343{
vishnu.pscc71aba2014-10-09 15:26:29 -07002344 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 unsigned long start;
2346
2347 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002348 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 if (!vma)
2350 return NULL;
2351 if (vma->vm_start <= addr)
2352 return vma;
2353 if (!(vma->vm_flags & VM_GROWSDOWN))
2354 return NULL;
2355 start = vma->vm_start;
2356 if (expand_stack(vma, addr))
2357 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002358 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002359 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 return vma;
2361}
2362#endif
2363
Jesse Barnese1d6d012014-12-12 16:55:27 -08002364EXPORT_SYMBOL_GPL(find_extend_vma);
2365
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002367 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002369 *
2370 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002372static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002374 unsigned long nr_accounted = 0;
2375
Hugh Dickins365e9c872005-10-29 18:16:18 -07002376 /* Update high watermark before we lower total_vm */
2377 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002379 long nrpages = vma_pages(vma);
2380
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002381 if (vma->vm_flags & VM_ACCOUNT)
2382 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002383 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002384 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002385 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002386 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 validate_mm(mm);
2388}
2389
2390/*
2391 * Get rid of page table information in the indicated region.
2392 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002393 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 */
2395static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002396 struct vm_area_struct *vma, struct vm_area_struct *prev,
2397 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398{
vishnu.pscc71aba2014-10-09 15:26:29 -07002399 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002400 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401
2402 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002403 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002404 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002405 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002406 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002407 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002408 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409}
2410
2411/*
2412 * Create a list of vma's touched by the unmap, removing them from the mm's
2413 * vma list as we go..
2414 */
2415static void
2416detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2417 struct vm_area_struct *prev, unsigned long end)
2418{
2419 struct vm_area_struct **insertion_point;
2420 struct vm_area_struct *tail_vma = NULL;
2421
2422 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002423 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002425 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 mm->map_count--;
2427 tail_vma = vma;
2428 vma = vma->vm_next;
2429 } while (vma && vma->vm_start < end);
2430 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002431 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002432 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002433 vma_gap_update(vma);
2434 } else
2435 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002437
2438 /* Kill the cache */
2439 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440}
2441
2442/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002443 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2444 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002446static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 unsigned long addr, int new_below)
2448{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002450 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
Andi Kleena5516432008-07-23 21:27:41 -07002452 if (is_vm_hugetlb_page(vma) && (addr &
2453 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 return -EINVAL;
2455
Christoph Lametere94b1762006-12-06 20:33:17 -08002456 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002458 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
2460 /* most fields are the same, copy all, and then fixup */
2461 *new = *vma;
2462
Rik van Riel5beb4932010-03-05 13:42:07 -08002463 INIT_LIST_HEAD(&new->anon_vma_chain);
2464
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (new_below)
2466 new->vm_end = addr;
2467 else {
2468 new->vm_start = addr;
2469 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2470 }
2471
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002472 err = vma_dup_policy(vma, new);
2473 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002474 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002476 err = anon_vma_clone(new, vma);
2477 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002478 goto out_free_mpol;
2479
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002480 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 get_file(new->vm_file);
2482
2483 if (new->vm_ops && new->vm_ops->open)
2484 new->vm_ops->open(new);
2485
2486 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002487 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2489 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002490 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Rik van Riel5beb4932010-03-05 13:42:07 -08002492 /* Success. */
2493 if (!err)
2494 return 0;
2495
2496 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002497 if (new->vm_ops && new->vm_ops->close)
2498 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002499 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002500 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002501 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002502 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002503 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002504 out_free_vma:
2505 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002506 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507}
2508
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002509/*
2510 * Split a vma into two pieces at address 'addr', a new vma is allocated
2511 * either for the first part or the tail.
2512 */
2513int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2514 unsigned long addr, int new_below)
2515{
2516 if (mm->map_count >= sysctl_max_map_count)
2517 return -ENOMEM;
2518
2519 return __split_vma(mm, vma, addr, new_below);
2520}
2521
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522/* Munmap is split into 2 main parts -- this part which finds
2523 * what needs doing, and the areas themselves, which do the
2524 * work. This now handles partial unmappings.
2525 * Jeremy Fitzhardinge <jeremy@goop.org>
2526 */
2527int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2528{
2529 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002530 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002532 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 return -EINVAL;
2534
vishnu.pscc71aba2014-10-09 15:26:29 -07002535 len = PAGE_ALIGN(len);
2536 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 return -EINVAL;
2538
2539 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002540 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002541 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002543 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002544 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
2546 /* if it doesn't overlap, we have nothing.. */
2547 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002548 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 return 0;
2550
2551 /*
2552 * If we need to split any vma, do it now to save pain later.
2553 *
2554 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2555 * unmapped vm_area_struct will remain in use: so lower split_vma
2556 * places tmp vma above, and higher split_vma places tmp vma below.
2557 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002558 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002559 int error;
2560
2561 /*
2562 * Make sure that map_count on return from munmap() will
2563 * not exceed its limit; but let map_count go just above
2564 * its limit temporarily, to help free resources as expected.
2565 */
2566 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2567 return -ENOMEM;
2568
2569 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 if (error)
2571 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002572 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 }
2574
2575 /* Does it split the last one? */
2576 last = find_vma(mm, end);
2577 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002578 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 if (error)
2580 return error;
2581 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002582 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583
2584 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002585 * unlock any mlock()ed ranges before detaching vmas
2586 */
2587 if (mm->locked_vm) {
2588 struct vm_area_struct *tmp = vma;
2589 while (tmp && tmp->vm_start < end) {
2590 if (tmp->vm_flags & VM_LOCKED) {
2591 mm->locked_vm -= vma_pages(tmp);
2592 munlock_vma_pages_all(tmp);
2593 }
2594 tmp = tmp->vm_next;
2595 }
2596 }
2597
2598 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 * Remove the vma's, and unmap the actual pages
2600 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002601 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2602 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
Dave Hansen1de4fa12014-11-14 07:18:31 -08002604 arch_unmap(mm, vma, start, end);
2605
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002607 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608
2609 return 0;
2610}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
Al Virobfce2812012-04-20 21:57:04 -04002612int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002613{
2614 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002615 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002616
2617 down_write(&mm->mmap_sem);
2618 ret = do_munmap(mm, start, len);
2619 up_write(&mm->mmap_sem);
2620 return ret;
2621}
2622EXPORT_SYMBOL(vm_munmap);
2623
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002624SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002627 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628}
2629
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002630
2631/*
2632 * Emulation of deprecated remap_file_pages() syscall.
2633 */
2634SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2635 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2636{
2637
2638 struct mm_struct *mm = current->mm;
2639 struct vm_area_struct *vma;
2640 unsigned long populate = 0;
2641 unsigned long ret = -EINVAL;
2642 struct file *file;
2643
2644 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
2645 "See Documentation/vm/remap_file_pages.txt.\n",
2646 current->comm, current->pid);
2647
2648 if (prot)
2649 return ret;
2650 start = start & PAGE_MASK;
2651 size = size & PAGE_MASK;
2652
2653 if (start + size <= start)
2654 return ret;
2655
2656 /* Does pgoff wrap? */
2657 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2658 return ret;
2659
2660 down_write(&mm->mmap_sem);
2661 vma = find_vma(mm, start);
2662
2663 if (!vma || !(vma->vm_flags & VM_SHARED))
2664 goto out;
2665
2666 if (start < vma->vm_start || start + size > vma->vm_end)
2667 goto out;
2668
2669 if (pgoff == linear_page_index(vma, start)) {
2670 ret = 0;
2671 goto out;
2672 }
2673
2674 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2675 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2676 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2677
2678 flags &= MAP_NONBLOCK;
2679 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2680 if (vma->vm_flags & VM_LOCKED) {
2681 flags |= MAP_LOCKED;
2682 /* drop PG_Mlocked flag for over-mapped range */
2683 munlock_vma_pages_range(vma, start, start + size);
2684 }
2685
2686 file = get_file(vma->vm_file);
2687 ret = do_mmap_pgoff(vma->vm_file, start, size,
2688 prot, flags, pgoff, &populate);
2689 fput(file);
2690out:
2691 up_write(&mm->mmap_sem);
2692 if (populate)
2693 mm_populate(ret, populate);
2694 if (!IS_ERR_VALUE(ret))
2695 ret = 0;
2696 return ret;
2697}
2698
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699static inline void verify_mm_writelocked(struct mm_struct *mm)
2700{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002701#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2703 WARN_ON(1);
2704 up_read(&mm->mmap_sem);
2705 }
2706#endif
2707}
2708
2709/*
2710 * this is really a simplified "do_mmap". it only handles
2711 * anonymous maps. eventually we may be able to do some
2712 * brk-specific accounting here.
2713 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002714static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715{
vishnu.pscc71aba2014-10-09 15:26:29 -07002716 struct mm_struct *mm = current->mm;
2717 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002719 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002721 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722
2723 len = PAGE_ALIGN(len);
2724 if (!len)
2725 return addr;
2726
Kirill Korotaev3a459752006-09-07 14:17:04 +04002727 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2728
Al Viro2c6a1012009-12-03 19:40:46 -05002729 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002730 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002731 return error;
2732
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002733 error = mlock_future_check(mm, mm->def_flags, len);
2734 if (error)
2735 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736
2737 /*
2738 * mm->mmap_sem is required to protect against another thread
2739 * changing the mappings in case we sleep.
2740 */
2741 verify_mm_writelocked(mm);
2742
2743 /*
2744 * Clear old maps. this also does some error checking for us
2745 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002746 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2747 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 if (do_munmap(mm, addr, len))
2749 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 }
2751
2752 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002753 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 return -ENOMEM;
2755
2756 if (mm->map_count > sysctl_max_map_count)
2757 return -ENOMEM;
2758
Al Viro191c5422012-02-13 03:58:52 +00002759 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 return -ENOMEM;
2761
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002763 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002764 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002765 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 goto out;
2767
2768 /*
2769 * create a vma struct for an anonymous mapping
2770 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002771 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 if (!vma) {
2773 vm_unacct_memory(len >> PAGE_SHIFT);
2774 return -ENOMEM;
2775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
Rik van Riel5beb4932010-03-05 13:42:07 -08002777 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 vma->vm_mm = mm;
2779 vma->vm_start = addr;
2780 vma->vm_end = addr + len;
2781 vma->vm_pgoff = pgoff;
2782 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002783 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 vma_link(mm, vma, prev, rb_link, rb_parent);
2785out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002786 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002788 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002789 if (flags & VM_LOCKED)
2790 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002791 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 return addr;
2793}
2794
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002795unsigned long vm_brk(unsigned long addr, unsigned long len)
2796{
2797 struct mm_struct *mm = current->mm;
2798 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002799 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002800
2801 down_write(&mm->mmap_sem);
2802 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002803 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002804 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002805 if (populate)
2806 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002807 return ret;
2808}
2809EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
2811/* Release all mmaps. */
2812void exit_mmap(struct mm_struct *mm)
2813{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002814 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002815 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 unsigned long nr_accounted = 0;
2817
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002818 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002819 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002820
Rik van Rielba470de2008-10-18 20:26:50 -07002821 if (mm->locked_vm) {
2822 vma = mm->mmap;
2823 while (vma) {
2824 if (vma->vm_flags & VM_LOCKED)
2825 munlock_vma_pages_all(vma);
2826 vma = vma->vm_next;
2827 }
2828 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002829
2830 arch_exit_mmap(mm);
2831
Rik van Rielba470de2008-10-18 20:26:50 -07002832 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002833 if (!vma) /* Can happen if dup_mmap() received an OOM */
2834 return;
2835
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002838 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002839 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002840 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002841 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002842
Hugh Dickins6ee86302013-04-29 15:07:44 -07002843 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002844 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002847 * Walk the list again, actually closing and freeing it,
2848 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002850 while (vma) {
2851 if (vma->vm_flags & VM_ACCOUNT)
2852 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002853 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002854 }
2855 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856}
2857
2858/* Insert vm structure into process list sorted by address
2859 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002860 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002862int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863{
Hugh Dickins6597d782012-10-08 16:29:07 -07002864 struct vm_area_struct *prev;
2865 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866
Chen Gangc9d13f52015-09-08 15:04:08 -07002867 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2868 &prev, &rb_link, &rb_parent))
2869 return -ENOMEM;
2870 if ((vma->vm_flags & VM_ACCOUNT) &&
2871 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2872 return -ENOMEM;
2873
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 /*
2875 * The vm_pgoff of a purely anonymous vma should be irrelevant
2876 * until its first write fault, when page's anon_vma and index
2877 * are set. But now set the vm_pgoff it will almost certainly
2878 * end up with (unless mremap moves it elsewhere before that
2879 * first wfault), so /proc/pid/maps tells a consistent story.
2880 *
2881 * By setting it to reflect the virtual start address of the
2882 * vma, merges and splits can happen in a seamless way, just
2883 * using the existing file pgoff checks and manipulations.
2884 * Similarly in do_mmap_pgoff and in do_brk.
2885 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002886 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 BUG_ON(vma->anon_vma);
2888 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2889 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302890
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 vma_link(mm, vma, prev, rb_link, rb_parent);
2892 return 0;
2893}
2894
2895/*
2896 * Copy the vma structure to a new location in the same mm,
2897 * prior to moving page table entries, to effect an mremap move.
2898 */
2899struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002900 unsigned long addr, unsigned long len, pgoff_t pgoff,
2901 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902{
2903 struct vm_area_struct *vma = *vmap;
2904 unsigned long vma_start = vma->vm_start;
2905 struct mm_struct *mm = vma->vm_mm;
2906 struct vm_area_struct *new_vma, *prev;
2907 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002908 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909
2910 /*
2911 * If anonymous vma has not yet been faulted, update new pgoff
2912 * to match new location, to increase its chance of merging.
2913 */
Oleg Nesterovce757992015-09-08 14:58:34 -07002914 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002916 faulted_in_anon_vma = false;
2917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918
Hugh Dickins6597d782012-10-08 16:29:07 -07002919 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2920 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002922 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2923 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 if (new_vma) {
2925 /*
2926 * Source vma may have been merged into new_vma
2927 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002928 if (unlikely(vma_start >= new_vma->vm_start &&
2929 vma_start < new_vma->vm_end)) {
2930 /*
2931 * The only way we can get a vma_merge with
2932 * self during an mremap is if the vma hasn't
2933 * been faulted in yet and we were allowed to
2934 * reset the dst vma->vm_pgoff to the
2935 * destination address of the mremap to allow
2936 * the merge to happen. mremap must change the
2937 * vm_pgoff linearity between src and dst vmas
2938 * (in turn preventing a vma_merge) to be
2939 * safe. It is only safe to keep the vm_pgoff
2940 * linear if there are no pages mapped yet.
2941 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002942 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002943 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002944 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002945 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002947 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07002948 if (!new_vma)
2949 goto out;
2950 *new_vma = *vma;
2951 new_vma->vm_start = addr;
2952 new_vma->vm_end = addr + len;
2953 new_vma->vm_pgoff = pgoff;
2954 if (vma_dup_policy(vma, new_vma))
2955 goto out_free_vma;
2956 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2957 if (anon_vma_clone(new_vma, vma))
2958 goto out_free_mempol;
2959 if (new_vma->vm_file)
2960 get_file(new_vma->vm_file);
2961 if (new_vma->vm_ops && new_vma->vm_ops->open)
2962 new_vma->vm_ops->open(new_vma);
2963 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2964 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 }
2966 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002967
Chen Gange3975892015-09-08 15:03:38 -07002968out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002969 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07002970out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08002971 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07002972out:
Rik van Riel5beb4932010-03-05 13:42:07 -08002973 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002975
2976/*
2977 * Return true if the calling process may expand its vm space by the passed
2978 * number of pages
2979 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002980bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07002981{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002982 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
2983 return false;
2984
2985 if ((flags & (VM_WRITE | VM_SHARED | (VM_STACK_FLAGS &
2986 (VM_GROWSUP | VM_GROWSDOWN)))) == VM_WRITE)
2987 return mm->data_vm + npages <= rlimit(RLIMIT_DATA);
2988
2989 return true;
2990}
2991
2992void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
2993{
2994 mm->total_vm += npages;
2995
2996 if ((flags & (VM_EXEC | VM_WRITE)) == VM_EXEC)
2997 mm->exec_vm += npages;
2998 else if (flags & (VM_STACK_FLAGS & (VM_GROWSUP | VM_GROWSDOWN)))
2999 mm->stack_vm += npages;
3000 else if ((flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
3001 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003002}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003003
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003004static int special_mapping_fault(struct vm_area_struct *vma,
3005 struct vm_fault *vmf);
3006
3007/*
3008 * Having a close hook prevents vma merging regardless of flags.
3009 */
3010static void special_mapping_close(struct vm_area_struct *vma)
3011{
3012}
3013
3014static const char *special_mapping_name(struct vm_area_struct *vma)
3015{
3016 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3017}
3018
3019static const struct vm_operations_struct special_mapping_vmops = {
3020 .close = special_mapping_close,
3021 .fault = special_mapping_fault,
3022 .name = special_mapping_name,
3023};
3024
3025static const struct vm_operations_struct legacy_special_mapping_vmops = {
3026 .close = special_mapping_close,
3027 .fault = special_mapping_fault,
3028};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003029
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003030static int special_mapping_fault(struct vm_area_struct *vma,
3031 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003032{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003033 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003034 struct page **pages;
3035
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003036 if (vma->vm_ops == &legacy_special_mapping_vmops)
3037 pages = vma->vm_private_data;
3038 else
3039 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
3040 pages;
3041
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003042 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003043 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003044
3045 if (*pages) {
3046 struct page *page = *pages;
3047 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003048 vmf->page = page;
3049 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003050 }
3051
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003052 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003053}
3054
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003055static struct vm_area_struct *__install_special_mapping(
3056 struct mm_struct *mm,
3057 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003058 unsigned long vm_flags, void *priv,
3059 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003060{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003061 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003062 struct vm_area_struct *vma;
3063
3064 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3065 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003066 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003067
Rik van Riel5beb4932010-03-05 13:42:07 -08003068 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003069 vma->vm_mm = mm;
3070 vma->vm_start = addr;
3071 vma->vm_end = addr + len;
3072
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003073 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003074 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003075
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003076 vma->vm_ops = ops;
3077 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003078
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003079 ret = insert_vm_struct(mm, vma);
3080 if (ret)
3081 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003082
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003083 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003084
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003085 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003086
Stefani Seibold3935ed62014-03-17 23:22:02 +01003087 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003088
3089out:
3090 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003091 return ERR_PTR(ret);
3092}
3093
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003094/*
3095 * Called with mm->mmap_sem held for writing.
3096 * Insert a new vma covering the given region, with the given flags.
3097 * Its pages are supplied by the given array of struct page *.
3098 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3099 * The region past the last page supplied will always produce SIGBUS.
3100 * The array pointer and the pages it points to are assumed to stay alive
3101 * for as long as this mapping might exist.
3102 */
3103struct vm_area_struct *_install_special_mapping(
3104 struct mm_struct *mm,
3105 unsigned long addr, unsigned long len,
3106 unsigned long vm_flags, const struct vm_special_mapping *spec)
3107{
Chen Gang27f28b92015-11-05 18:48:41 -08003108 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3109 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003110}
3111
Stefani Seibold3935ed62014-03-17 23:22:02 +01003112int install_special_mapping(struct mm_struct *mm,
3113 unsigned long addr, unsigned long len,
3114 unsigned long vm_flags, struct page **pages)
3115{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003116 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003117 mm, addr, len, vm_flags, (void *)pages,
3118 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003119
Duan Jiong14bd5b42014-06-04 16:07:05 -07003120 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003121}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003122
3123static DEFINE_MUTEX(mm_all_locks_mutex);
3124
Peter Zijlstra454ed842008-08-11 09:30:25 +02003125static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003126{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003127 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003128 /*
3129 * The LSB of head.next can't change from under us
3130 * because we hold the mm_all_locks_mutex.
3131 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003132 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003133 /*
3134 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003135 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003136 * the same anon_vma we won't take it again.
3137 *
3138 * No need of atomic instructions here, head.next
3139 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003140 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003141 */
3142 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003143 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003144 BUG();
3145 }
3146}
3147
Peter Zijlstra454ed842008-08-11 09:30:25 +02003148static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003149{
3150 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3151 /*
3152 * AS_MM_ALL_LOCKS can't change from under us because
3153 * we hold the mm_all_locks_mutex.
3154 *
3155 * Operations on ->flags have to be atomic because
3156 * even if AS_MM_ALL_LOCKS is stable thanks to the
3157 * mm_all_locks_mutex, there may be other cpus
3158 * changing other bitflags in parallel to us.
3159 */
3160 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3161 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003162 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003163 }
3164}
3165
3166/*
3167 * This operation locks against the VM for all pte/vma/mm related
3168 * operations that could ever happen on a certain mm. This includes
3169 * vmtruncate, try_to_unmap, and all page faults.
3170 *
3171 * The caller must take the mmap_sem in write mode before calling
3172 * mm_take_all_locks(). The caller isn't allowed to release the
3173 * mmap_sem until mm_drop_all_locks() returns.
3174 *
3175 * mmap_sem in write mode is required in order to block all operations
3176 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003177 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003178 * anon_vmas to be associated with existing vmas.
3179 *
3180 * A single task can't take more than one mm_take_all_locks() in a row
3181 * or it would deadlock.
3182 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003183 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003184 * mapping->flags avoid to take the same lock twice, if more than one
3185 * vma in this mm is backed by the same anon_vma or address_space.
3186 *
3187 * We can take all the locks in random order because the VM code
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003188 * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003189 * takes more than one of them in a row. Secondly we're protected
3190 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3191 *
3192 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3193 * that may have to take thousand of locks.
3194 *
3195 * mm_take_all_locks() can fail if it's interrupted by signals.
3196 */
3197int mm_take_all_locks(struct mm_struct *mm)
3198{
3199 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003200 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003201
3202 BUG_ON(down_read_trylock(&mm->mmap_sem));
3203
3204 mutex_lock(&mm_all_locks_mutex);
3205
3206 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3207 if (signal_pending(current))
3208 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003209 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003210 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003211 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003212
3213 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3214 if (signal_pending(current))
3215 goto out_unlock;
3216 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003217 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3218 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003219 }
3220
Kautuk Consul584cff52011-10-31 17:08:59 -07003221 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003222
3223out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003224 mm_drop_all_locks(mm);
3225 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003226}
3227
3228static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3229{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003230 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003231 /*
3232 * The LSB of head.next can't change to 0 from under
3233 * us because we hold the mm_all_locks_mutex.
3234 *
3235 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003236 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003237 * never see our bitflag.
3238 *
3239 * No need of atomic instructions here, head.next
3240 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003241 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003242 */
3243 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003244 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003245 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003246 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003247 }
3248}
3249
3250static void vm_unlock_mapping(struct address_space *mapping)
3251{
3252 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3253 /*
3254 * AS_MM_ALL_LOCKS can't change to 0 from under us
3255 * because we hold the mm_all_locks_mutex.
3256 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003257 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003258 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3259 &mapping->flags))
3260 BUG();
3261 }
3262}
3263
3264/*
3265 * The mmap_sem cannot be released by the caller until
3266 * mm_drop_all_locks() returns.
3267 */
3268void mm_drop_all_locks(struct mm_struct *mm)
3269{
3270 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003271 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003272
3273 BUG_ON(down_read_trylock(&mm->mmap_sem));
3274 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3275
3276 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3277 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003278 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3279 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003280 if (vma->vm_file && vma->vm_file->f_mapping)
3281 vm_unlock_mapping(vma->vm_file->f_mapping);
3282 }
3283
3284 mutex_unlock(&mm_all_locks_mutex);
3285}
David Howells8feae132009-01-08 12:04:47 +00003286
3287/*
3288 * initialise the VMA slab
3289 */
3290void __init mmap_init(void)
3291{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003292 int ret;
3293
Tejun Heo908c7f12014-09-08 09:51:29 +09003294 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003295 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003296}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003297
3298/*
3299 * Initialise sysctl_user_reserve_kbytes.
3300 *
3301 * This is intended to prevent a user from starting a single memory hogging
3302 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3303 * mode.
3304 *
3305 * The default value is min(3% of free memory, 128MB)
3306 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3307 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003308static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003309{
3310 unsigned long free_kbytes;
3311
3312 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3313
3314 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3315 return 0;
3316}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003317subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003318
3319/*
3320 * Initialise sysctl_admin_reserve_kbytes.
3321 *
3322 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3323 * to log in and kill a memory hogging process.
3324 *
3325 * Systems with more than 256MB will reserve 8MB, enough to recover
3326 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3327 * only reserve 3% of free pages by default.
3328 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003329static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003330{
3331 unsigned long free_kbytes;
3332
3333 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3334
3335 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3336 return 0;
3337}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003338subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003339
3340/*
3341 * Reinititalise user and admin reserves if memory is added or removed.
3342 *
3343 * The default user reserve max is 128MB, and the default max for the
3344 * admin reserve is 8MB. These are usually, but not always, enough to
3345 * enable recovery from a memory hogging process using login/sshd, a shell,
3346 * and tools like top. It may make sense to increase or even disable the
3347 * reserve depending on the existence of swap or variations in the recovery
3348 * tools. So, the admin may have changed them.
3349 *
3350 * If memory is added and the reserves have been eliminated or increased above
3351 * the default max, then we'll trust the admin.
3352 *
3353 * If memory is removed and there isn't enough free memory, then we
3354 * need to reset the reserves.
3355 *
3356 * Otherwise keep the reserve set by the admin.
3357 */
3358static int reserve_mem_notifier(struct notifier_block *nb,
3359 unsigned long action, void *data)
3360{
3361 unsigned long tmp, free_kbytes;
3362
3363 switch (action) {
3364 case MEM_ONLINE:
3365 /* Default max is 128MB. Leave alone if modified by operator. */
3366 tmp = sysctl_user_reserve_kbytes;
3367 if (0 < tmp && tmp < (1UL << 17))
3368 init_user_reserve();
3369
3370 /* Default max is 8MB. Leave alone if modified by operator. */
3371 tmp = sysctl_admin_reserve_kbytes;
3372 if (0 < tmp && tmp < (1UL << 13))
3373 init_admin_reserve();
3374
3375 break;
3376 case MEM_OFFLINE:
3377 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3378
3379 if (sysctl_user_reserve_kbytes > free_kbytes) {
3380 init_user_reserve();
3381 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3382 sysctl_user_reserve_kbytes);
3383 }
3384
3385 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3386 init_admin_reserve();
3387 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3388 sysctl_admin_reserve_kbytes);
3389 }
3390 break;
3391 default:
3392 break;
3393 }
3394 return NOTIFY_OK;
3395}
3396
3397static struct notifier_block reserve_mem_nb = {
3398 .notifier_call = reserve_mem_notifier,
3399};
3400
3401static int __meminit init_reserve_notifier(void)
3402{
3403 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003404 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003405
3406 return 0;
3407}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003408subsys_initcall(init_reserve_notifier);