blob: 79bcc9f92e482de9047c3927e068ff392db5c1bc [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
Hugh Dickinse0da3822005-04-19 13:29:15 -070061static void unmap_region(struct mm_struct *mm,
62 struct vm_area_struct *vma, struct vm_area_struct *prev,
63 unsigned long start, unsigned long end);
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/* description of effects of mapping type and prot in current implementation.
66 * this is due to the limited x86 page protection hardware. The expected
67 * behavior is in parens:
68 *
69 * map_type prot
70 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
71 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
72 * w: (no) no w: (no) no w: (yes) yes w: (no) no
73 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070074 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
76 * w: (no) no w: (no) no w: (copy) copy w: (no) no
77 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
78 *
79 */
80pgprot_t protection_map[16] = {
81 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
82 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
83};
84
Hugh Dickins804af2c2006-07-26 21:39:49 +010085pgprot_t vm_get_page_prot(unsigned long vm_flags)
86{
Dave Kleikampb845f312008-07-08 00:28:51 +100087 return __pgprot(pgprot_val(protection_map[vm_flags &
88 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
89 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010090}
91EXPORT_SYMBOL(vm_get_page_prot);
92
Peter Feiner64e45502014-10-13 15:55:46 -070093static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
94{
95 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
96}
97
98/* Update vma->vm_page_prot to reflect vma->vm_flags. */
99void vma_set_page_prot(struct vm_area_struct *vma)
100{
101 unsigned long vm_flags = vma->vm_flags;
102
103 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
104 if (vma_wants_writenotify(vma)) {
105 vm_flags &= ~VM_SHARED;
106 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
107 vm_flags);
108 }
109}
110
111
Shaohua Li34679d72011-05-24 17:11:18 -0700112int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
113int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -0800114unsigned long sysctl_overcommit_kbytes __read_mostly;
Christoph Lameterc3d8c142005-09-06 15:16:33 -0700115int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700116unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700117unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -0700118/*
119 * Make sure vm_committed_as in one cacheline and not cacheline shared with
120 * other variables. It can be updated by several CPUs frequently.
121 */
122struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -0800125 * The global memory commitment made in the system can be a metric
126 * that can be used to drive ballooning decisions when Linux is hosted
127 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
128 * balancing memory across competing virtual machines that are hosted.
129 * Several metrics drive this policy engine including the guest reported
130 * memory commitment.
131 */
132unsigned long vm_memory_committed(void)
133{
134 return percpu_counter_read_positive(&vm_committed_as);
135}
136EXPORT_SYMBOL_GPL(vm_memory_committed);
137
138/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 * Check that a process has enough memory to allocate a new virtual
140 * mapping. 0 means there is enough memory for the allocation to
141 * succeed and -ENOMEM implies there is not.
142 *
143 * We currently support three overcommit policies, which are set via the
144 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
145 *
146 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
147 * Additional code 2002 Jul 20 by Robert Love.
148 *
149 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
150 *
151 * Note this is a helper function intended to be used by LSMs which
152 * wish to use this logic.
153 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700154int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
Roman Gushchin5703b082015-02-11 15:28:39 -0800156 long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -0700158 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
159 -(s64)vm_committed_as_batch * num_online_cpus(),
160 "memory commitment underflow");
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 vm_acct_memory(pages);
163
164 /*
165 * Sometimes we want to use more memory than we have
166 */
167 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
168 return 0;
169
170 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700171 free = global_page_state(NR_FREE_PAGES);
172 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Dmitry Finkc15bef32011-07-25 17:12:19 -0700174 /*
175 * shmem pages shouldn't be counted as free in this
176 * case, they can't be purged, only swapped out, and
177 * that won't affect the overall amount of available
178 * memory in the system.
179 */
180 free -= global_page_state(NR_SHMEM);
181
Shaohua Liec8acf22013-02-22 16:34:38 -0800182 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184 /*
185 * Any slabs which are created with the
186 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
187 * which are reclaimable, under pressure. The dentry
188 * cache and most inode caches should fall into this
189 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700190 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
192 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700193 * Leave reserved pages. The pages are not for anonymous pages.
194 */
195 if (free <= totalreserve_pages)
196 goto error;
197 else
198 free -= totalreserve_pages;
199
200 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700201 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 */
203 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700204 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206 if (free > pages)
207 return 0;
208
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700209 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
211
Jerome Marchand00619bc2013-11-12 15:08:31 -0800212 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700214 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
216 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700217 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700219 /*
220 * Don't let a single process grow so big a user can't recover
221 */
222 if (mm) {
223 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
Roman Gushchin5703b082015-02-11 15:28:39 -0800224 allowed -= min_t(long, mm->total_vm / 32, reserve);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700227 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700229error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 vm_unacct_memory(pages);
231
232 return -ENOMEM;
233}
234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800236 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 */
238static void __remove_shared_vm_struct(struct vm_area_struct *vma,
239 struct file *file, struct address_space *mapping)
240{
241 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500242 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700244 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800247 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 flush_dcache_mmap_unlock(mapping);
249}
250
251/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700252 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700253 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700255void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
257 struct file *file = vma->vm_file;
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 if (file) {
260 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800261 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800263 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700265}
266
267/*
268 * Close a vm structure and free it, returning the next.
269 */
270static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
271{
272 struct vm_area_struct *next = vma->vm_next;
273
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700274 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (vma->vm_ops && vma->vm_ops->close)
276 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700277 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700278 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700279 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700281 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282}
283
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700284static unsigned long do_brk(unsigned long addr, unsigned long len);
285
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100286SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700288 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 unsigned long newbrk, oldbrk;
290 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700291 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800292 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
294 down_write(&mm->mmap_sem);
295
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700296#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800297 /*
298 * CONFIG_COMPAT_BRK can still be overridden by setting
299 * randomize_va_space to 2, which will still cause mm->start_brk
300 * to be arbitrarily shifted
301 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700302 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800303 min_brk = mm->start_brk;
304 else
305 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700306#else
307 min_brk = mm->start_brk;
308#endif
309 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700311
312 /*
313 * Check against rlimit here. If this check is done later after the test
314 * of oldbrk with newbrk then it can escape the test and let the data
315 * segment grow beyond its set limit the in case where the limit is
316 * not page aligned -Ram Gupta
317 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700318 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
319 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700320 goto out;
321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 newbrk = PAGE_ALIGN(brk);
323 oldbrk = PAGE_ALIGN(mm->brk);
324 if (oldbrk == newbrk)
325 goto set_brk;
326
327 /* Always allow shrinking brk. */
328 if (brk <= mm->brk) {
329 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
330 goto set_brk;
331 goto out;
332 }
333
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 /* Check against existing mmap mappings. */
335 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
336 goto out;
337
338 /* Ok, looks good - let it rip. */
339 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
340 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342set_brk:
343 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800344 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
345 up_write(&mm->mmap_sem);
346 if (populate)
347 mm_populate(oldbrk, newbrk - oldbrk);
348 return brk;
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350out:
351 retval = mm->brk;
352 up_write(&mm->mmap_sem);
353 return retval;
354}
355
Michel Lespinassed3737182012-12-11 16:01:38 -0800356static long vma_compute_subtree_gap(struct vm_area_struct *vma)
357{
358 unsigned long max, subtree_gap;
359 max = vma->vm_start;
360 if (vma->vm_prev)
361 max -= vma->vm_prev->vm_end;
362 if (vma->vm_rb.rb_left) {
363 subtree_gap = rb_entry(vma->vm_rb.rb_left,
364 struct vm_area_struct, vm_rb)->rb_subtree_gap;
365 if (subtree_gap > max)
366 max = subtree_gap;
367 }
368 if (vma->vm_rb.rb_right) {
369 subtree_gap = rb_entry(vma->vm_rb.rb_right,
370 struct vm_area_struct, vm_rb)->rb_subtree_gap;
371 if (subtree_gap > max)
372 max = subtree_gap;
373 }
374 return max;
375}
376
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700377#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378static int browse_rb(struct rb_root *root)
379{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800380 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 struct rb_node *nd, *pn = NULL;
382 unsigned long prev = 0, pend = 0;
383
384 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
385 struct vm_area_struct *vma;
386 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800387 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700388 pr_emerg("vm_start %lx < prev %lx\n",
389 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800390 bug = 1;
391 }
392 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700393 pr_emerg("vm_start %lx < pend %lx\n",
394 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800395 bug = 1;
396 }
397 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700398 pr_emerg("vm_start %lx > vm_end %lx\n",
399 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800400 bug = 1;
401 }
402 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700403 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800404 vma->rb_subtree_gap,
405 vma_compute_subtree_gap(vma));
406 bug = 1;
407 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 i++;
409 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800410 prev = vma->vm_start;
411 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 }
413 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800414 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800416 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700417 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800418 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800420 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421}
422
Michel Lespinassed3737182012-12-11 16:01:38 -0800423static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
424{
425 struct rb_node *nd;
426
427 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
428 struct vm_area_struct *vma;
429 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700430 VM_BUG_ON_VMA(vma != ignore &&
431 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
432 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700436static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 int bug = 0;
439 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800440 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700441 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700442
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700443 while (vma) {
444 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700445
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800446 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700447 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
448 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800449 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800450 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700451 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 i++;
453 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800454 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700455 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800456 bug = 1;
457 }
458 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700459 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700460 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800461 bug = 1;
462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800464 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700465 if (i != -1)
466 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800467 bug = 1;
468 }
Sasha Levin96dad672014-10-09 15:28:39 -0700469 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470}
471#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800472#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473#define validate_mm(mm) do { } while (0)
474#endif
475
Michel Lespinassed3737182012-12-11 16:01:38 -0800476RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
477 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
478
479/*
480 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
481 * vma->vm_prev->vm_end values changed, without modifying the vma's position
482 * in the rbtree.
483 */
484static void vma_gap_update(struct vm_area_struct *vma)
485{
486 /*
487 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
488 * function that does exacltly what we want.
489 */
490 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
491}
492
493static inline void vma_rb_insert(struct vm_area_struct *vma,
494 struct rb_root *root)
495{
496 /* All rb_subtree_gap values must be consistent prior to insertion */
497 validate_mm_rb(root, NULL);
498
499 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
500}
501
502static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
503{
504 /*
505 * All rb_subtree_gap values must be consistent prior to erase,
506 * with the possible exception of the vma being erased.
507 */
508 validate_mm_rb(root, vma);
509
510 /*
511 * Note rb_erase_augmented is a fairly large inline function,
512 * so make sure we instantiate it only once with our desired
513 * augmented rbtree callbacks.
514 */
515 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
516}
517
Michel Lespinassebf181b92012-10-08 16:31:39 -0700518/*
519 * vma has some anon_vma assigned, and is already inserted on that
520 * anon_vma's interval trees.
521 *
522 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
523 * vma must be removed from the anon_vma's interval trees using
524 * anon_vma_interval_tree_pre_update_vma().
525 *
526 * After the update, the vma will be reinserted using
527 * anon_vma_interval_tree_post_update_vma().
528 *
529 * The entire update must be protected by exclusive mmap_sem and by
530 * the root anon_vma's mutex.
531 */
532static inline void
533anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
534{
535 struct anon_vma_chain *avc;
536
537 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
538 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
539}
540
541static inline void
542anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
543{
544 struct anon_vma_chain *avc;
545
546 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
547 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
548}
549
Hugh Dickins6597d782012-10-08 16:29:07 -0700550static int find_vma_links(struct mm_struct *mm, unsigned long addr,
551 unsigned long end, struct vm_area_struct **pprev,
552 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553{
Hugh Dickins6597d782012-10-08 16:29:07 -0700554 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
556 __rb_link = &mm->mm_rb.rb_node;
557 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559 while (*__rb_link) {
560 struct vm_area_struct *vma_tmp;
561
562 __rb_parent = *__rb_link;
563 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
564
565 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700566 /* Fail if an existing vma overlaps the area */
567 if (vma_tmp->vm_start < end)
568 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 __rb_link = &__rb_parent->rb_left;
570 } else {
571 rb_prev = __rb_parent;
572 __rb_link = &__rb_parent->rb_right;
573 }
574 }
575
576 *pprev = NULL;
577 if (rb_prev)
578 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
579 *rb_link = __rb_link;
580 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700581 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582}
583
Cyril Hrubise8420a82013-04-29 15:08:33 -0700584static unsigned long count_vma_pages_range(struct mm_struct *mm,
585 unsigned long addr, unsigned long end)
586{
587 unsigned long nr_pages = 0;
588 struct vm_area_struct *vma;
589
590 /* Find first overlaping mapping */
591 vma = find_vma_intersection(mm, addr, end);
592 if (!vma)
593 return 0;
594
595 nr_pages = (min(end, vma->vm_end) -
596 max(addr, vma->vm_start)) >> PAGE_SHIFT;
597
598 /* Iterate over the rest of the overlaps */
599 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
600 unsigned long overlap_len;
601
602 if (vma->vm_start > end)
603 break;
604
605 overlap_len = min(end, vma->vm_end) - vma->vm_start;
606 nr_pages += overlap_len >> PAGE_SHIFT;
607 }
608
609 return nr_pages;
610}
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
613 struct rb_node **rb_link, struct rb_node *rb_parent)
614{
Michel Lespinassed3737182012-12-11 16:01:38 -0800615 /* Update tracking information for the gap following the new vma. */
616 if (vma->vm_next)
617 vma_gap_update(vma->vm_next);
618 else
619 mm->highest_vm_end = vma->vm_end;
620
621 /*
622 * vma->vm_prev wasn't known when we followed the rbtree to find the
623 * correct insertion point for that vma. As a result, we could not
624 * update the vma vm_rb parents rb_subtree_gap values on the way down.
625 * So, we first insert the vma with a zero rb_subtree_gap value
626 * (to be consistent with what we did on the way down), and then
627 * immediately update the gap to the correct value. Finally we
628 * rebalance the rbtree after all augmented values have been set.
629 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800631 vma->rb_subtree_gap = 0;
632 vma_gap_update(vma);
633 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}
635
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700636static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
ZhenwenXu48aae422009-01-06 14:40:21 -0800638 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
640 file = vma->vm_file;
641 if (file) {
642 struct address_space *mapping = file->f_mapping;
643
644 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500645 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700647 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800650 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 flush_dcache_mmap_unlock(mapping);
652 }
653}
654
655static void
656__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
657 struct vm_area_struct *prev, struct rb_node **rb_link,
658 struct rb_node *rb_parent)
659{
660 __vma_link_list(mm, vma, prev, rb_parent);
661 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662}
663
664static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
665 struct vm_area_struct *prev, struct rb_node **rb_link,
666 struct rb_node *rb_parent)
667{
668 struct address_space *mapping = NULL;
669
Huang Shijie64ac4942014-06-04 16:07:33 -0700670 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800672 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 __vma_link(mm, vma, prev, rb_link, rb_parent);
676 __vma_link_file(vma);
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800679 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681 mm->map_count++;
682 validate_mm(mm);
683}
684
685/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700686 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700687 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800689static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Hugh Dickins6597d782012-10-08 16:29:07 -0700691 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800692 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Hugh Dickins6597d782012-10-08 16:29:07 -0700694 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
695 &prev, &rb_link, &rb_parent))
696 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 __vma_link(mm, vma, prev, rb_link, rb_parent);
698 mm->map_count++;
699}
700
701static inline void
702__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
703 struct vm_area_struct *prev)
704{
Michel Lespinassed3737182012-12-11 16:01:38 -0800705 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700706
Michel Lespinassed3737182012-12-11 16:01:38 -0800707 vma_rb_erase(vma, &mm->mm_rb);
708 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700709 if (next)
710 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700711
712 /* Kill the cache */
713 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714}
715
716/*
717 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
718 * is already present in an i_mmap tree without adjusting the tree.
719 * The following helper function should be used when such adjustments
720 * are necessary. The "insert" vma (if any) is to be inserted
721 * before we drop the necessary locks.
722 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800723int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
725{
726 struct mm_struct *mm = vma->vm_mm;
727 struct vm_area_struct *next = vma->vm_next;
728 struct vm_area_struct *importer = NULL;
729 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700730 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700731 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800733 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 long adjust_next = 0;
735 int remove_next = 0;
736
737 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700738 struct vm_area_struct *exporter = NULL;
739
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 if (end >= next->vm_end) {
741 /*
742 * vma expands, overlapping all the next, and
743 * perhaps the one after too (mprotect case 6).
744 */
745again: remove_next = 1 + (end > next->vm_end);
746 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700747 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 importer = vma;
749 } else if (end > next->vm_start) {
750 /*
751 * vma expands, overlapping part of the next:
752 * mprotect case 5 shifting the boundary up.
753 */
754 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700755 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 importer = vma;
757 } else if (end < vma->vm_end) {
758 /*
759 * vma shrinks, and !insert tells it's not
760 * split_vma inserting another: so it must be
761 * mprotect case 4 shifting the boundary down.
762 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700763 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700764 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 importer = next;
766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Rik van Riel5beb4932010-03-05 13:42:07 -0800768 /*
769 * Easily overlooked: when mprotect shifts the boundary,
770 * make sure the expanding vma has anon_vma set if the
771 * shrinking vma had, to cover any anon pages imported.
772 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700773 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800774 int error;
775
Linus Torvalds287d97a2010-04-10 15:22:30 -0700776 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300777 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700778 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300779 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800780 }
781 }
782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (file) {
784 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800785 root = &mapping->i_mmap;
786 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530787
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800788 if (adjust_next)
789 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530790
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800791 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700794 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 * are visible to arm/parisc __flush_dcache_page
796 * throughout; but we cannot insert into address
797 * space until vma start or end is updated.
798 */
799 __vma_link_file(insert);
800 }
801 }
802
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800803 vma_adjust_trans_huge(vma, start, end, adjust_next);
804
Michel Lespinassebf181b92012-10-08 16:31:39 -0700805 anon_vma = vma->anon_vma;
806 if (!anon_vma && adjust_next)
807 anon_vma = next->anon_vma;
808 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700809 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
810 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000811 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700812 anon_vma_interval_tree_pre_update_vma(vma);
813 if (adjust_next)
814 anon_vma_interval_tree_pre_update_vma(next);
815 }
Rik van Riel012f18002010-08-09 17:18:40 -0700816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 if (root) {
818 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700819 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700821 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
823
Michel Lespinassed3737182012-12-11 16:01:38 -0800824 if (start != vma->vm_start) {
825 vma->vm_start = start;
826 start_changed = true;
827 }
828 if (end != vma->vm_end) {
829 vma->vm_end = end;
830 end_changed = true;
831 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 vma->vm_pgoff = pgoff;
833 if (adjust_next) {
834 next->vm_start += adjust_next << PAGE_SHIFT;
835 next->vm_pgoff += adjust_next;
836 }
837
838 if (root) {
839 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700840 vma_interval_tree_insert(next, root);
841 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 flush_dcache_mmap_unlock(mapping);
843 }
844
845 if (remove_next) {
846 /*
847 * vma_merge has merged next into vma, and needs
848 * us to remove next before dropping the locks.
849 */
850 __vma_unlink(mm, next, vma);
851 if (file)
852 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 } else if (insert) {
854 /*
855 * split_vma has split insert from vma, and needs
856 * us to insert it before dropping the locks
857 * (it may either follow vma or precede it).
858 */
859 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800860 } else {
861 if (start_changed)
862 vma_gap_update(vma);
863 if (end_changed) {
864 if (!next)
865 mm->highest_vm_end = end;
866 else if (!adjust_next)
867 vma_gap_update(next);
868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 }
870
Michel Lespinassebf181b92012-10-08 16:31:39 -0700871 if (anon_vma) {
872 anon_vma_interval_tree_post_update_vma(vma);
873 if (adjust_next)
874 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800875 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800878 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530880 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100881 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530882
883 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100884 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530885 }
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700888 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530889 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700891 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800892 if (next->anon_vma)
893 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700895 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 kmem_cache_free(vm_area_cachep, next);
897 /*
898 * In mprotect's case 6 (see comments on vma_merge),
899 * we must remove another next too. It would clutter
900 * up the code too much to do both in one go.
901 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800902 next = vma->vm_next;
903 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800905 else if (next)
906 vma_gap_update(next);
907 else
908 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530910 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100911 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800914
915 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916}
917
918/*
919 * If the vma has a ->close operation then the driver probably needs to release
920 * per-vma resources, so we don't attempt to merge those.
921 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700923 struct file *file, unsigned long vm_flags,
924 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800926 /*
927 * VM_SOFTDIRTY should not prevent from VMA merging, if we
928 * match the flags but dirty bit -- the caller should mark
929 * merged VMA as dirty. If dirty bit won't be excluded from
930 * comparison, we increase pressue on the memory system forcing
931 * the kernel to generate new VMAs when old one could be
932 * extended instead.
933 */
934 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 return 0;
936 if (vma->vm_file != file)
937 return 0;
938 if (vma->vm_ops && vma->vm_ops->close)
939 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700940 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
941 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 return 1;
943}
944
945static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700946 struct anon_vma *anon_vma2,
947 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948{
Shaohua Li965f55d2011-05-24 17:11:20 -0700949 /*
950 * The list_is_singular() test is to avoid merging VMA cloned from
951 * parents. This can improve scalability caused by anon_vma lock.
952 */
953 if ((!anon_vma1 || !anon_vma2) && (!vma ||
954 list_is_singular(&vma->anon_vma_chain)))
955 return 1;
956 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
959/*
960 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
961 * in front of (at a lower virtual address and file offset than) the vma.
962 *
963 * We cannot merge two vmas if they have differently assigned (non-NULL)
964 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
965 *
966 * We don't check here for the merged mmap wrapping around the end of pagecache
967 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
968 * wrap, nor mmaps which cover the final page at index -1UL.
969 */
970static int
971can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700972 struct anon_vma *anon_vma, struct file *file,
973 pgoff_t vm_pgoff,
974 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700976 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700977 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (vma->vm_pgoff == vm_pgoff)
979 return 1;
980 }
981 return 0;
982}
983
984/*
985 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
986 * beyond (at a higher virtual address and file offset than) the vma.
987 *
988 * We cannot merge two vmas if they have differently assigned (non-NULL)
989 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
990 */
991static int
992can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700993 struct anon_vma *anon_vma, struct file *file,
994 pgoff_t vm_pgoff,
995 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700997 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700998 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001000 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1002 return 1;
1003 }
1004 return 0;
1005}
1006
1007/*
1008 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1009 * whether that can be merged with its predecessor or its successor.
1010 * Or both (it neatly fills a hole).
1011 *
1012 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1013 * certain not to be mapped by the time vma_merge is called; but when
1014 * called for mprotect, it is certain to be already mapped (either at
1015 * an offset within prev, or at the start of next), and the flags of
1016 * this area are about to be changed to vm_flags - and the no-change
1017 * case has already been eliminated.
1018 *
1019 * The following mprotect cases have to be considered, where AAAA is
1020 * the area passed down from mprotect_fixup, never extending beyond one
1021 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1022 *
1023 * AAAA AAAA AAAA AAAA
1024 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1025 * cannot merge might become might become might become
1026 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1027 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1028 * mremap move: PPPPNNNNNNNN 8
1029 * AAAA
1030 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1031 * might become case 1 below case 2 below case 3 below
1032 *
1033 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1034 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1035 */
1036struct vm_area_struct *vma_merge(struct mm_struct *mm,
1037 struct vm_area_struct *prev, unsigned long addr,
1038 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001039 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001040 pgoff_t pgoff, struct mempolicy *policy,
1041 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042{
1043 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1044 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001045 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
1047 /*
1048 * We later require that vma->vm_flags == vm_flags,
1049 * so this tests vma->vm_flags & VM_SPECIAL, too.
1050 */
1051 if (vm_flags & VM_SPECIAL)
1052 return NULL;
1053
1054 if (prev)
1055 next = prev->vm_next;
1056 else
1057 next = mm->mmap;
1058 area = next;
1059 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1060 next = next->vm_next;
1061
1062 /*
1063 * Can it merge with the predecessor?
1064 */
1065 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001066 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001068 anon_vma, file, pgoff,
1069 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 /*
1071 * OK, it can. Can we now merge in the successor as well?
1072 */
1073 if (next && end == next->vm_start &&
1074 mpol_equal(policy, vma_policy(next)) &&
1075 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001076 anon_vma, file,
1077 pgoff+pglen,
1078 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001080 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001082 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 next->vm_end, prev->vm_pgoff, NULL);
1084 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001085 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001087 if (err)
1088 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001089 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 return prev;
1091 }
1092
1093 /*
1094 * Can this new request be merged in front of next?
1095 */
1096 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001097 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001099 anon_vma, file, pgoff+pglen,
1100 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001102 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 addr, prev->vm_pgoff, NULL);
1104 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001105 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001107 if (err)
1108 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001109 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 return area;
1111 }
1112
1113 return NULL;
1114}
1115
1116/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001117 * Rough compatbility check to quickly see if it's even worth looking
1118 * at sharing an anon_vma.
1119 *
1120 * They need to have the same vm_file, and the flags can only differ
1121 * in things that mprotect may change.
1122 *
1123 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1124 * we can merge the two vma's. For example, we refuse to merge a vma if
1125 * there is a vm_ops->close() function, because that indicates that the
1126 * driver is doing some kind of reference counting. But that doesn't
1127 * really matter for the anon_vma sharing case.
1128 */
1129static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1130{
1131 return a->vm_end == b->vm_start &&
1132 mpol_equal(vma_policy(a), vma_policy(b)) &&
1133 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001134 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001135 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1136}
1137
1138/*
1139 * Do some basic sanity checking to see if we can re-use the anon_vma
1140 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1141 * the same as 'old', the other will be the new one that is trying
1142 * to share the anon_vma.
1143 *
1144 * NOTE! This runs with mm_sem held for reading, so it is possible that
1145 * the anon_vma of 'old' is concurrently in the process of being set up
1146 * by another page fault trying to merge _that_. But that's ok: if it
1147 * is being set up, that automatically means that it will be a singleton
1148 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001149 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001150 *
1151 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1152 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1153 * is to return an anon_vma that is "complex" due to having gone through
1154 * a fork).
1155 *
1156 * We also make sure that the two vma's are compatible (adjacent,
1157 * and with the same memory policies). That's all stable, even with just
1158 * a read lock on the mm_sem.
1159 */
1160static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1161{
1162 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001163 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001164
1165 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1166 return anon_vma;
1167 }
1168 return NULL;
1169}
1170
1171/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1173 * neighbouring vmas for a suitable anon_vma, before it goes off
1174 * to allocate a new anon_vma. It checks because a repetitive
1175 * sequence of mprotects and faults may otherwise lead to distinct
1176 * anon_vmas being allocated, preventing vma merge in subsequent
1177 * mprotect.
1178 */
1179struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1180{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001181 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184 near = vma->vm_next;
1185 if (!near)
1186 goto try_prev;
1187
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001188 anon_vma = reusable_anon_vma(near, vma, near);
1189 if (anon_vma)
1190 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001192 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 if (!near)
1194 goto none;
1195
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001196 anon_vma = reusable_anon_vma(near, near, vma);
1197 if (anon_vma)
1198 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199none:
1200 /*
1201 * There's no absolute need to look only at touching neighbours:
1202 * we could search further afield for "compatible" anon_vmas.
1203 * But it would probably just be a waste of time searching,
1204 * or lead to too many vmas hanging off the same anon_vma.
1205 * We're trying to allow mprotect remerging later on,
1206 * not trying to minimize memory used for anon_vmas.
1207 */
1208 return NULL;
1209}
1210
1211#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001212void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 struct file *file, long pages)
1214{
1215 const unsigned long stack_flags
1216 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1217
Huang Shijie44de9d02012-07-31 16:41:49 -07001218 mm->total_vm += pages;
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 if (file) {
1221 mm->shared_vm += pages;
1222 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1223 mm->exec_vm += pages;
1224 } else if (flags & stack_flags)
1225 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
1227#endif /* CONFIG_PROC_FS */
1228
1229/*
Al Viro40401532012-02-13 03:58:52 +00001230 * If a hint addr is less than mmap_min_addr change hint to be as
1231 * low as possible but still greater than mmap_min_addr
1232 */
1233static inline unsigned long round_hint_to_min(unsigned long hint)
1234{
1235 hint &= PAGE_MASK;
1236 if (((void *)hint != NULL) &&
1237 (hint < mmap_min_addr))
1238 return PAGE_ALIGN(mmap_min_addr);
1239 return hint;
1240}
1241
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001242static inline int mlock_future_check(struct mm_struct *mm,
1243 unsigned long flags,
1244 unsigned long len)
1245{
1246 unsigned long locked, lock_limit;
1247
1248 /* mlock MCL_FUTURE? */
1249 if (flags & VM_LOCKED) {
1250 locked = len >> PAGE_SHIFT;
1251 locked += mm->locked_vm;
1252 lock_limit = rlimit(RLIMIT_MEMLOCK);
1253 lock_limit >>= PAGE_SHIFT;
1254 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1255 return -EAGAIN;
1256 }
1257 return 0;
1258}
1259
Al Viro40401532012-02-13 03:58:52 +00001260/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001261 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001263unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001265 unsigned long flags, vm_flags_t vm_flags,
1266 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
vishnu.pscc71aba2014-10-09 15:26:29 -07001268 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
Michel Lespinasse41badc12013-02-22 16:32:47 -08001270 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001271
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001272 if (!len)
1273 return -EINVAL;
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 /*
1276 * Does the application expect PROT_READ to imply PROT_EXEC?
1277 *
1278 * (the exception is when the underlying filesystem is noexec
1279 * mounted, in which case we dont add PROT_EXEC.)
1280 */
1281 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001282 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 prot |= PROT_EXEC;
1284
Eric Paris7cd94142007-11-26 18:47:40 -05001285 if (!(flags & MAP_FIXED))
1286 addr = round_hint_to_min(addr);
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 /* Careful about overflows.. */
1289 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001290 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 return -ENOMEM;
1292
1293 /* offset overflow? */
1294 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001295 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297 /* Too many mappings? */
1298 if (mm->map_count > sysctl_max_map_count)
1299 return -ENOMEM;
1300
1301 /* Obtain the address to map to. we verify (or select) it and ensure
1302 * that it represents a valid section of the address space.
1303 */
1304 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1305 if (addr & ~PAGE_MASK)
1306 return addr;
1307
1308 /* Do simple checking here so the lower-level routines won't have
1309 * to. we assume access permissions have been handled by the open
1310 * of the memory object, so we don't do any here.
1311 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001312 vm_flags |= calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1314
Huang Shijiecdf7b342009-09-21 17:03:36 -07001315 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if (!can_do_mlock())
1317 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001318
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001319 if (mlock_future_check(mm, vm_flags, len))
1320 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001323 struct inode *inode = file_inode(file);
1324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 switch (flags & MAP_TYPE) {
1326 case MAP_SHARED:
1327 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1328 return -EACCES;
1329
1330 /*
1331 * Make sure we don't allow writing to an append-only
1332 * file..
1333 */
1334 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1335 return -EACCES;
1336
1337 /*
1338 * Make sure there are no mandatory locks on the file.
1339 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001340 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 return -EAGAIN;
1342
1343 vm_flags |= VM_SHARED | VM_MAYSHARE;
1344 if (!(file->f_mode & FMODE_WRITE))
1345 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1346
1347 /* fall through */
1348 case MAP_PRIVATE:
1349 if (!(file->f_mode & FMODE_READ))
1350 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001351 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001352 if (vm_flags & VM_EXEC)
1353 return -EPERM;
1354 vm_flags &= ~VM_MAYEXEC;
1355 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001356
Al Viro72c2d532013-09-22 16:27:52 -04001357 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001358 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001359 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1360 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 break;
1362
1363 default:
1364 return -EINVAL;
1365 }
1366 } else {
1367 switch (flags & MAP_TYPE) {
1368 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001369 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1370 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001371 /*
1372 * Ignore pgoff.
1373 */
1374 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 vm_flags |= VM_SHARED | VM_MAYSHARE;
1376 break;
1377 case MAP_PRIVATE:
1378 /*
1379 * Set pgoff according to addr for anon_vma.
1380 */
1381 pgoff = addr >> PAGE_SHIFT;
1382 break;
1383 default:
1384 return -EINVAL;
1385 }
1386 }
1387
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001388 /*
1389 * Set 'VM_NORESERVE' if we should not account for the
1390 * memory use of this mapping.
1391 */
1392 if (flags & MAP_NORESERVE) {
1393 /* We honor MAP_NORESERVE if allowed to overcommit */
1394 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1395 vm_flags |= VM_NORESERVE;
1396
1397 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1398 if (file && is_file_hugepages(file))
1399 vm_flags |= VM_NORESERVE;
1400 }
1401
1402 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001403 if (!IS_ERR_VALUE(addr) &&
1404 ((vm_flags & VM_LOCKED) ||
1405 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001406 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001407 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001408}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001409
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001410SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1411 unsigned long, prot, unsigned long, flags,
1412 unsigned long, fd, unsigned long, pgoff)
1413{
1414 struct file *file = NULL;
1415 unsigned long retval = -EBADF;
1416
1417 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001418 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001419 file = fget(fd);
1420 if (!file)
1421 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001422 if (is_file_hugepages(file))
1423 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001424 retval = -EINVAL;
1425 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1426 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001427 } else if (flags & MAP_HUGETLB) {
1428 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001429 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001430
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001431 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001432 if (!hs)
1433 return -EINVAL;
1434
1435 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001436 /*
1437 * VM_NORESERVE is used because the reservations will be
1438 * taken when vm_ops->mmap() is called
1439 * A dummy user value is used because we are not locking
1440 * memory so no accounting is necessary
1441 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001442 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001443 VM_NORESERVE,
1444 &user, HUGETLB_ANONHUGE_INODE,
1445 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001446 if (IS_ERR(file))
1447 return PTR_ERR(file);
1448 }
1449
1450 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1451
Al Viroeb36c582012-05-30 20:17:35 -04001452 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001453out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001454 if (file)
1455 fput(file);
1456out:
1457 return retval;
1458}
1459
Christoph Hellwiga4679372010-03-10 15:21:15 -08001460#ifdef __ARCH_WANT_SYS_OLD_MMAP
1461struct mmap_arg_struct {
1462 unsigned long addr;
1463 unsigned long len;
1464 unsigned long prot;
1465 unsigned long flags;
1466 unsigned long fd;
1467 unsigned long offset;
1468};
1469
1470SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1471{
1472 struct mmap_arg_struct a;
1473
1474 if (copy_from_user(&a, arg, sizeof(a)))
1475 return -EFAULT;
1476 if (a.offset & ~PAGE_MASK)
1477 return -EINVAL;
1478
1479 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1480 a.offset >> PAGE_SHIFT);
1481}
1482#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1483
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001484/*
1485 * Some shared mappigns will want the pages marked read-only
1486 * to track write events. If so, we'll downgrade vm_page_prot
1487 * to the private version (using protection_map[] without the
1488 * VM_SHARED bit).
1489 */
1490int vma_wants_writenotify(struct vm_area_struct *vma)
1491{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001492 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001493 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001494
1495 /* If it was private or non-writable, the write bit is already clear */
1496 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1497 return 0;
1498
1499 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001500 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001501 return 1;
1502
Peter Feiner64e45502014-10-13 15:55:46 -07001503 /* The open routine did something to the protections that pgprot_modify
1504 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001505 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001506 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001507 return 0;
1508
Peter Feiner64e45502014-10-13 15:55:46 -07001509 /* Do we need to track softdirty? */
1510 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1511 return 1;
1512
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001513 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001514 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001515 return 0;
1516
1517 /* Can the mapping track the dirty pages? */
1518 return vma->vm_file && vma->vm_file->f_mapping &&
1519 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1520}
1521
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001522/*
1523 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001524 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001525 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001526static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001527{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001528 /*
1529 * hugetlb has its own accounting separate from the core VM
1530 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1531 */
1532 if (file && is_file_hugepages(file))
1533 return 0;
1534
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001535 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1536}
1537
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001538unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001539 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001540{
1541 struct mm_struct *mm = current->mm;
1542 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001543 int error;
1544 struct rb_node **rb_link, *rb_parent;
1545 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001546
Cyril Hrubise8420a82013-04-29 15:08:33 -07001547 /* Check against address space limit. */
1548 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1549 unsigned long nr_pages;
1550
1551 /*
1552 * MAP_FIXED may remove pages of mappings that intersects with
1553 * requested mapping. Account for the pages it would unmap.
1554 */
1555 if (!(vm_flags & MAP_FIXED))
1556 return -ENOMEM;
1557
1558 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1559
1560 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1561 return -ENOMEM;
1562 }
1563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 /* Clear old maps */
1565 error = -ENOMEM;
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001566 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1567 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 if (do_munmap(mm, addr, len))
1569 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
1571
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001572 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001573 * Private writable mapping: check memory availability
1574 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001575 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001576 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001577 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001578 return -ENOMEM;
1579 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581
1582 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001583 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001585 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1586 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001587 if (vma)
1588 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
1590 /*
1591 * Determine the object being mapped and call the appropriate
1592 * specific mapper. the address has already been validated, but
1593 * not unmapped, but the maps are removed from the list.
1594 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001595 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 if (!vma) {
1597 error = -ENOMEM;
1598 goto unacct_error;
1599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 vma->vm_mm = mm;
1602 vma->vm_start = addr;
1603 vma->vm_end = addr + len;
1604 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001605 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001607 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
1609 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 if (vm_flags & VM_DENYWRITE) {
1611 error = deny_write_access(file);
1612 if (error)
1613 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001615 if (vm_flags & VM_SHARED) {
1616 error = mapping_map_writable(file->f_mapping);
1617 if (error)
1618 goto allow_write_and_free_vma;
1619 }
1620
1621 /* ->mmap() can change vma->vm_file, but must guarantee that
1622 * vma_link() below can deny write-access if VM_DENYWRITE is set
1623 * and map writably if VM_SHARED is set. This usually means the
1624 * new file must not have been exposed to user-space, yet.
1625 */
Al Virocb0942b2012-08-27 14:48:26 -04001626 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 error = file->f_op->mmap(file, vma);
1628 if (error)
1629 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001630
1631 /* Can addr have changed??
1632 *
1633 * Answer: Yes, several device drivers can do it in their
1634 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001635 * Bug: If addr is changed, prev, rb_link, rb_parent should
1636 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001637 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001638 WARN_ON_ONCE(addr != vma->vm_start);
1639
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001640 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001641 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 } else if (vm_flags & VM_SHARED) {
1643 error = shmem_zero_setup(vma);
1644 if (error)
1645 goto free_vma;
1646 }
1647
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001648 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001649 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001650 if (file) {
1651 if (vm_flags & VM_SHARED)
1652 mapping_unmap_writable(file->f_mapping);
1653 if (vm_flags & VM_DENYWRITE)
1654 allow_write_access(file);
1655 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001656 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001657out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001658 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001659
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001660 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001662 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1663 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001664 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001665 else
1666 vma->vm_flags &= ~VM_LOCKED;
1667 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301668
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001669 if (file)
1670 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301671
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001672 /*
1673 * New (or expanded) vma always get soft dirty status.
1674 * Otherwise user-space soft-dirty page tracker won't
1675 * be able to distinguish situation when vma area unmapped,
1676 * then new mapped in-place (which must be aimed as
1677 * a completely new data area).
1678 */
1679 vma->vm_flags |= VM_SOFTDIRTY;
1680
Peter Feiner64e45502014-10-13 15:55:46 -07001681 vma_set_page_prot(vma);
1682
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 return addr;
1684
1685unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 vma->vm_file = NULL;
1687 fput(file);
1688
1689 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001690 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1691 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001692 if (vm_flags & VM_SHARED)
1693 mapping_unmap_writable(file->f_mapping);
1694allow_write_and_free_vma:
1695 if (vm_flags & VM_DENYWRITE)
1696 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697free_vma:
1698 kmem_cache_free(vm_area_cachep, vma);
1699unacct_error:
1700 if (charged)
1701 vm_unacct_memory(charged);
1702 return error;
1703}
1704
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001705unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1706{
1707 /*
1708 * We implement the search by looking for an rbtree node that
1709 * immediately follows a suitable gap. That is,
1710 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1711 * - gap_end = vma->vm_start >= info->low_limit + length;
1712 * - gap_end - gap_start >= length
1713 */
1714
1715 struct mm_struct *mm = current->mm;
1716 struct vm_area_struct *vma;
1717 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1718
1719 /* Adjust search length to account for worst case alignment overhead */
1720 length = info->length + info->align_mask;
1721 if (length < info->length)
1722 return -ENOMEM;
1723
1724 /* Adjust search limits by the desired length */
1725 if (info->high_limit < length)
1726 return -ENOMEM;
1727 high_limit = info->high_limit - length;
1728
1729 if (info->low_limit > high_limit)
1730 return -ENOMEM;
1731 low_limit = info->low_limit + length;
1732
1733 /* Check if rbtree root looks promising */
1734 if (RB_EMPTY_ROOT(&mm->mm_rb))
1735 goto check_highest;
1736 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1737 if (vma->rb_subtree_gap < length)
1738 goto check_highest;
1739
1740 while (true) {
1741 /* Visit left subtree if it looks promising */
1742 gap_end = vma->vm_start;
1743 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1744 struct vm_area_struct *left =
1745 rb_entry(vma->vm_rb.rb_left,
1746 struct vm_area_struct, vm_rb);
1747 if (left->rb_subtree_gap >= length) {
1748 vma = left;
1749 continue;
1750 }
1751 }
1752
1753 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1754check_current:
1755 /* Check if current node has a suitable gap */
1756 if (gap_start > high_limit)
1757 return -ENOMEM;
1758 if (gap_end >= low_limit && gap_end - gap_start >= length)
1759 goto found;
1760
1761 /* Visit right subtree if it looks promising */
1762 if (vma->vm_rb.rb_right) {
1763 struct vm_area_struct *right =
1764 rb_entry(vma->vm_rb.rb_right,
1765 struct vm_area_struct, vm_rb);
1766 if (right->rb_subtree_gap >= length) {
1767 vma = right;
1768 continue;
1769 }
1770 }
1771
1772 /* Go back up the rbtree to find next candidate node */
1773 while (true) {
1774 struct rb_node *prev = &vma->vm_rb;
1775 if (!rb_parent(prev))
1776 goto check_highest;
1777 vma = rb_entry(rb_parent(prev),
1778 struct vm_area_struct, vm_rb);
1779 if (prev == vma->vm_rb.rb_left) {
1780 gap_start = vma->vm_prev->vm_end;
1781 gap_end = vma->vm_start;
1782 goto check_current;
1783 }
1784 }
1785 }
1786
1787check_highest:
1788 /* Check highest gap, which does not precede any rbtree node */
1789 gap_start = mm->highest_vm_end;
1790 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1791 if (gap_start > high_limit)
1792 return -ENOMEM;
1793
1794found:
1795 /* We found a suitable gap. Clip it with the original low_limit. */
1796 if (gap_start < info->low_limit)
1797 gap_start = info->low_limit;
1798
1799 /* Adjust gap address to the desired alignment */
1800 gap_start += (info->align_offset - gap_start) & info->align_mask;
1801
1802 VM_BUG_ON(gap_start + info->length > info->high_limit);
1803 VM_BUG_ON(gap_start + info->length > gap_end);
1804 return gap_start;
1805}
1806
1807unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1808{
1809 struct mm_struct *mm = current->mm;
1810 struct vm_area_struct *vma;
1811 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1812
1813 /* Adjust search length to account for worst case alignment overhead */
1814 length = info->length + info->align_mask;
1815 if (length < info->length)
1816 return -ENOMEM;
1817
1818 /*
1819 * Adjust search limits by the desired length.
1820 * See implementation comment at top of unmapped_area().
1821 */
1822 gap_end = info->high_limit;
1823 if (gap_end < length)
1824 return -ENOMEM;
1825 high_limit = gap_end - length;
1826
1827 if (info->low_limit > high_limit)
1828 return -ENOMEM;
1829 low_limit = info->low_limit + length;
1830
1831 /* Check highest gap, which does not precede any rbtree node */
1832 gap_start = mm->highest_vm_end;
1833 if (gap_start <= high_limit)
1834 goto found_highest;
1835
1836 /* Check if rbtree root looks promising */
1837 if (RB_EMPTY_ROOT(&mm->mm_rb))
1838 return -ENOMEM;
1839 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1840 if (vma->rb_subtree_gap < length)
1841 return -ENOMEM;
1842
1843 while (true) {
1844 /* Visit right subtree if it looks promising */
1845 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1846 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1847 struct vm_area_struct *right =
1848 rb_entry(vma->vm_rb.rb_right,
1849 struct vm_area_struct, vm_rb);
1850 if (right->rb_subtree_gap >= length) {
1851 vma = right;
1852 continue;
1853 }
1854 }
1855
1856check_current:
1857 /* Check if current node has a suitable gap */
1858 gap_end = vma->vm_start;
1859 if (gap_end < low_limit)
1860 return -ENOMEM;
1861 if (gap_start <= high_limit && gap_end - gap_start >= length)
1862 goto found;
1863
1864 /* Visit left subtree if it looks promising */
1865 if (vma->vm_rb.rb_left) {
1866 struct vm_area_struct *left =
1867 rb_entry(vma->vm_rb.rb_left,
1868 struct vm_area_struct, vm_rb);
1869 if (left->rb_subtree_gap >= length) {
1870 vma = left;
1871 continue;
1872 }
1873 }
1874
1875 /* Go back up the rbtree to find next candidate node */
1876 while (true) {
1877 struct rb_node *prev = &vma->vm_rb;
1878 if (!rb_parent(prev))
1879 return -ENOMEM;
1880 vma = rb_entry(rb_parent(prev),
1881 struct vm_area_struct, vm_rb);
1882 if (prev == vma->vm_rb.rb_right) {
1883 gap_start = vma->vm_prev ?
1884 vma->vm_prev->vm_end : 0;
1885 goto check_current;
1886 }
1887 }
1888 }
1889
1890found:
1891 /* We found a suitable gap. Clip it with the original high_limit. */
1892 if (gap_end > info->high_limit)
1893 gap_end = info->high_limit;
1894
1895found_highest:
1896 /* Compute highest gap address at the desired alignment */
1897 gap_end -= info->length;
1898 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1899
1900 VM_BUG_ON(gap_end < info->low_limit);
1901 VM_BUG_ON(gap_end < gap_start);
1902 return gap_end;
1903}
1904
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905/* Get an address range which is currently unmapped.
1906 * For shmat() with addr=0.
1907 *
1908 * Ugly calling convention alert:
1909 * Return value with the low bits set means error value,
1910 * ie
1911 * if (ret & ~PAGE_MASK)
1912 * error = ret;
1913 *
1914 * This function "knows" that -ENOMEM has the bits set.
1915 */
1916#ifndef HAVE_ARCH_UNMAPPED_AREA
1917unsigned long
1918arch_get_unmapped_area(struct file *filp, unsigned long addr,
1919 unsigned long len, unsigned long pgoff, unsigned long flags)
1920{
1921 struct mm_struct *mm = current->mm;
1922 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001923 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001925 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 return -ENOMEM;
1927
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001928 if (flags & MAP_FIXED)
1929 return addr;
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 if (addr) {
1932 addr = PAGE_ALIGN(addr);
1933 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001934 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 (!vma || addr + len <= vma->vm_start))
1936 return addr;
1937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001939 info.flags = 0;
1940 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001941 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001942 info.high_limit = TASK_SIZE;
1943 info.align_mask = 0;
1944 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945}
vishnu.pscc71aba2014-10-09 15:26:29 -07001946#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948/*
1949 * This mmap-allocator allocates new areas top-down from below the
1950 * stack's low limit (the base):
1951 */
1952#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1953unsigned long
1954arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1955 const unsigned long len, const unsigned long pgoff,
1956 const unsigned long flags)
1957{
1958 struct vm_area_struct *vma;
1959 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001960 unsigned long addr = addr0;
1961 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001964 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return -ENOMEM;
1966
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001967 if (flags & MAP_FIXED)
1968 return addr;
1969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 /* requesting a specific address */
1971 if (addr) {
1972 addr = PAGE_ALIGN(addr);
1973 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001974 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 (!vma || addr + len <= vma->vm_start))
1976 return addr;
1977 }
1978
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001979 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1980 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001981 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001982 info.high_limit = mm->mmap_base;
1983 info.align_mask = 0;
1984 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 /*
1987 * A failed mmap() very likely causes application failure,
1988 * so fall back to the bottom-up function here. This scenario
1989 * can happen with large stack limits and large mmap()
1990 * allocations.
1991 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001992 if (addr & ~PAGE_MASK) {
1993 VM_BUG_ON(addr != -ENOMEM);
1994 info.flags = 0;
1995 info.low_limit = TASK_UNMAPPED_BASE;
1996 info.high_limit = TASK_SIZE;
1997 addr = vm_unmapped_area(&info);
1998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 return addr;
2001}
2002#endif
2003
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004unsigned long
2005get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2006 unsigned long pgoff, unsigned long flags)
2007{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002008 unsigned long (*get_area)(struct file *, unsigned long,
2009 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
Al Viro9206de92009-12-03 15:23:11 -05002011 unsigned long error = arch_mmap_check(addr, len, flags);
2012 if (error)
2013 return error;
2014
2015 /* Careful about overflows.. */
2016 if (len > TASK_SIZE)
2017 return -ENOMEM;
2018
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002019 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04002020 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002021 get_area = file->f_op->get_unmapped_area;
2022 addr = get_area(file, addr, len, pgoff, flags);
2023 if (IS_ERR_VALUE(addr))
2024 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002026 if (addr > TASK_SIZE - len)
2027 return -ENOMEM;
2028 if (addr & ~PAGE_MASK)
2029 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002030
Al Viro9ac4ed42012-05-30 17:13:15 -04002031 addr = arch_rebalance_pgtables(addr, len);
2032 error = security_mmap_addr(addr);
2033 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034}
2035
2036EXPORT_SYMBOL(get_unmapped_area);
2037
2038/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002039struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002041 struct rb_node *rb_node;
2042 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002044 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002045 vma = vmacache_find(mm, addr);
2046 if (likely(vma))
2047 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002049 rb_node = mm->mm_rb.rb_node;
2050 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002052 while (rb_node) {
2053 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002055 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002056
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002057 if (tmp->vm_end > addr) {
2058 vma = tmp;
2059 if (tmp->vm_start <= addr)
2060 break;
2061 rb_node = rb_node->rb_left;
2062 } else
2063 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002065
2066 if (vma)
2067 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 return vma;
2069}
2070
2071EXPORT_SYMBOL(find_vma);
2072
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002073/*
2074 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002075 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076struct vm_area_struct *
2077find_vma_prev(struct mm_struct *mm, unsigned long addr,
2078 struct vm_area_struct **pprev)
2079{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002080 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002082 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002083 if (vma) {
2084 *pprev = vma->vm_prev;
2085 } else {
2086 struct rb_node *rb_node = mm->mm_rb.rb_node;
2087 *pprev = NULL;
2088 while (rb_node) {
2089 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2090 rb_node = rb_node->rb_right;
2091 }
2092 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002093 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094}
2095
2096/*
2097 * Verify that the stack growth is acceptable and
2098 * update accounting. This is shared with both the
2099 * grow-up and grow-down cases.
2100 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002101static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102{
2103 struct mm_struct *mm = vma->vm_mm;
2104 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08002105 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106
2107 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002108 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 return -ENOMEM;
2110
2111 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08002112 actual_size = size;
2113 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2114 actual_size -= PAGE_SIZE;
Jason Low4db0c3c2015-04-15 16:14:08 -07002115 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 return -ENOMEM;
2117
2118 /* mlock limit tests */
2119 if (vma->vm_flags & VM_LOCKED) {
2120 unsigned long locked;
2121 unsigned long limit;
2122 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002123 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002124 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 if (locked > limit && !capable(CAP_IPC_LOCK))
2126 return -ENOMEM;
2127 }
2128
Adam Litke0d59a012007-01-30 14:35:39 -08002129 /* Check to ensure the stack will not grow into a hugetlb-only region */
2130 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2131 vma->vm_end - size;
2132 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2133 return -EFAULT;
2134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 /*
2136 * Overcommit.. This must be the final test, as it will
2137 * update security statistics.
2138 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002139 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 return -ENOMEM;
2141
2142 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 if (vma->vm_flags & VM_LOCKED)
2144 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002145 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 return 0;
2147}
2148
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002149#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002151 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2152 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002154int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155{
2156 int error;
2157
2158 if (!(vma->vm_flags & VM_GROWSUP))
2159 return -EFAULT;
2160
2161 /*
2162 * We must make sure the anon_vma is allocated
2163 * so that the anon_vma locking is not a noop.
2164 */
2165 if (unlikely(anon_vma_prepare(vma)))
2166 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002167 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168
2169 /*
2170 * vma->vm_start/vm_end cannot change under us because the caller
2171 * is required to hold the mmap_sem in read mode. We need the
2172 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002173 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 */
Helge Deller06b32f32006-12-19 19:28:33 +01002175 if (address < PAGE_ALIGN(address+4))
2176 address = PAGE_ALIGN(address+4);
2177 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002178 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002179 return -ENOMEM;
2180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 error = 0;
2182
2183 /* Somebody else might have raced and expanded it already */
2184 if (address > vma->vm_end) {
2185 unsigned long size, grow;
2186
2187 size = address - vma->vm_start;
2188 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2189
Hugh Dickins42c36f62011-05-09 17:44:42 -07002190 error = -ENOMEM;
2191 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2192 error = acct_stack_growth(vma, size, grow);
2193 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002194 /*
2195 * vma_gap_update() doesn't support concurrent
2196 * updates, but we only hold a shared mmap_sem
2197 * lock here, so we need to protect against
2198 * concurrent vma expansions.
2199 * vma_lock_anon_vma() doesn't help here, as
2200 * we don't guarantee that all growable vmas
2201 * in a mm share the same root anon vma.
2202 * So, we reuse mm->page_table_lock to guard
2203 * against concurrent vma expansions.
2204 */
2205 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002206 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002207 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002208 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002209 if (vma->vm_next)
2210 vma_gap_update(vma->vm_next);
2211 else
2212 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002213 spin_unlock(&vma->vm_mm->page_table_lock);
2214
Hugh Dickins42c36f62011-05-09 17:44:42 -07002215 perf_event_mmap(vma);
2216 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002219 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002220 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002221 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 return error;
2223}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002224#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2225
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226/*
2227 * vma is the first one with address < vma->vm_start. Have to extend vma.
2228 */
Michal Hockod05f3162011-05-24 17:11:44 -07002229int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002230 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231{
2232 int error;
2233
2234 /*
2235 * We must make sure the anon_vma is allocated
2236 * so that the anon_vma locking is not a noop.
2237 */
2238 if (unlikely(anon_vma_prepare(vma)))
2239 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002240
2241 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002242 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002243 if (error)
2244 return error;
2245
Rik van Rielbb4a3402010-08-09 17:18:37 -07002246 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 /*
2249 * vma->vm_start/vm_end cannot change under us because the caller
2250 * is required to hold the mmap_sem in read mode. We need the
2251 * anon_vma lock to serialize against concurrent expand_stacks.
2252 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
2254 /* Somebody else might have raced and expanded it already */
2255 if (address < vma->vm_start) {
2256 unsigned long size, grow;
2257
2258 size = vma->vm_end - address;
2259 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2260
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002261 error = -ENOMEM;
2262 if (grow <= vma->vm_pgoff) {
2263 error = acct_stack_growth(vma, size, grow);
2264 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002265 /*
2266 * vma_gap_update() doesn't support concurrent
2267 * updates, but we only hold a shared mmap_sem
2268 * lock here, so we need to protect against
2269 * concurrent vma expansions.
2270 * vma_lock_anon_vma() doesn't help here, as
2271 * we don't guarantee that all growable vmas
2272 * in a mm share the same root anon vma.
2273 * So, we reuse mm->page_table_lock to guard
2274 * against concurrent vma expansions.
2275 */
2276 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002277 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002278 vma->vm_start = address;
2279 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002280 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002281 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002282 spin_unlock(&vma->vm_mm->page_table_lock);
2283
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002284 perf_event_mmap(vma);
2285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 }
2287 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002288 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002289 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002290 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 return error;
2292}
2293
Linus Torvalds09884962013-02-27 08:36:04 -08002294/*
2295 * Note how expand_stack() refuses to expand the stack all the way to
2296 * abut the next virtual mapping, *unless* that mapping itself is also
2297 * a stack mapping. We want to leave room for a guard page, after all
2298 * (the guard page itself is not added here, that is done by the
2299 * actual page faulting logic)
2300 *
2301 * This matches the behavior of the guard page logic (see mm/memory.c:
2302 * check_stack_guard_page()), which only allows the guard page to be
2303 * removed under these circumstances.
2304 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002305#ifdef CONFIG_STACK_GROWSUP
2306int expand_stack(struct vm_area_struct *vma, unsigned long address)
2307{
Linus Torvalds09884962013-02-27 08:36:04 -08002308 struct vm_area_struct *next;
2309
2310 address &= PAGE_MASK;
2311 next = vma->vm_next;
2312 if (next && next->vm_start == address + PAGE_SIZE) {
2313 if (!(next->vm_flags & VM_GROWSUP))
2314 return -ENOMEM;
2315 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002316 return expand_upwards(vma, address);
2317}
2318
2319struct vm_area_struct *
2320find_extend_vma(struct mm_struct *mm, unsigned long addr)
2321{
2322 struct vm_area_struct *vma, *prev;
2323
2324 addr &= PAGE_MASK;
2325 vma = find_vma_prev(mm, addr, &prev);
2326 if (vma && (vma->vm_start <= addr))
2327 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002328 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002329 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002330 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002331 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002332 return prev;
2333}
2334#else
2335int expand_stack(struct vm_area_struct *vma, unsigned long address)
2336{
Linus Torvalds09884962013-02-27 08:36:04 -08002337 struct vm_area_struct *prev;
2338
2339 address &= PAGE_MASK;
2340 prev = vma->vm_prev;
2341 if (prev && prev->vm_end == address) {
2342 if (!(prev->vm_flags & VM_GROWSDOWN))
2343 return -ENOMEM;
2344 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002345 return expand_downwards(vma, address);
2346}
2347
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002349find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350{
vishnu.pscc71aba2014-10-09 15:26:29 -07002351 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 unsigned long start;
2353
2354 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002355 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 if (!vma)
2357 return NULL;
2358 if (vma->vm_start <= addr)
2359 return vma;
2360 if (!(vma->vm_flags & VM_GROWSDOWN))
2361 return NULL;
2362 start = vma->vm_start;
2363 if (expand_stack(vma, addr))
2364 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002365 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002366 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 return vma;
2368}
2369#endif
2370
Jesse Barnese1d6d012014-12-12 16:55:27 -08002371EXPORT_SYMBOL_GPL(find_extend_vma);
2372
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002374 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002376 *
2377 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002379static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002381 unsigned long nr_accounted = 0;
2382
Hugh Dickins365e9c872005-10-29 18:16:18 -07002383 /* Update high watermark before we lower total_vm */
2384 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002386 long nrpages = vma_pages(vma);
2387
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002388 if (vma->vm_flags & VM_ACCOUNT)
2389 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002390 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002391 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002392 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002393 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 validate_mm(mm);
2395}
2396
2397/*
2398 * Get rid of page table information in the indicated region.
2399 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002400 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 */
2402static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002403 struct vm_area_struct *vma, struct vm_area_struct *prev,
2404 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405{
vishnu.pscc71aba2014-10-09 15:26:29 -07002406 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002407 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408
2409 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002410 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002411 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002412 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002413 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002414 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002415 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416}
2417
2418/*
2419 * Create a list of vma's touched by the unmap, removing them from the mm's
2420 * vma list as we go..
2421 */
2422static void
2423detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2424 struct vm_area_struct *prev, unsigned long end)
2425{
2426 struct vm_area_struct **insertion_point;
2427 struct vm_area_struct *tail_vma = NULL;
2428
2429 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002430 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002432 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 mm->map_count--;
2434 tail_vma = vma;
2435 vma = vma->vm_next;
2436 } while (vma && vma->vm_start < end);
2437 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002438 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002439 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002440 vma_gap_update(vma);
2441 } else
2442 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002444
2445 /* Kill the cache */
2446 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447}
2448
2449/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002450 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2451 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002453static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 unsigned long addr, int new_below)
2455{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002457 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
Andi Kleena5516432008-07-23 21:27:41 -07002459 if (is_vm_hugetlb_page(vma) && (addr &
2460 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 return -EINVAL;
2462
Christoph Lametere94b1762006-12-06 20:33:17 -08002463 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002465 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
2467 /* most fields are the same, copy all, and then fixup */
2468 *new = *vma;
2469
Rik van Riel5beb4932010-03-05 13:42:07 -08002470 INIT_LIST_HEAD(&new->anon_vma_chain);
2471
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 if (new_below)
2473 new->vm_end = addr;
2474 else {
2475 new->vm_start = addr;
2476 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2477 }
2478
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002479 err = vma_dup_policy(vma, new);
2480 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002481 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002483 err = anon_vma_clone(new, vma);
2484 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002485 goto out_free_mpol;
2486
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002487 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 get_file(new->vm_file);
2489
2490 if (new->vm_ops && new->vm_ops->open)
2491 new->vm_ops->open(new);
2492
2493 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002494 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2496 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002497 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498
Rik van Riel5beb4932010-03-05 13:42:07 -08002499 /* Success. */
2500 if (!err)
2501 return 0;
2502
2503 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002504 if (new->vm_ops && new->vm_ops->close)
2505 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002506 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002507 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002508 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002509 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002510 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002511 out_free_vma:
2512 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002513 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514}
2515
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002516/*
2517 * Split a vma into two pieces at address 'addr', a new vma is allocated
2518 * either for the first part or the tail.
2519 */
2520int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2521 unsigned long addr, int new_below)
2522{
2523 if (mm->map_count >= sysctl_max_map_count)
2524 return -ENOMEM;
2525
2526 return __split_vma(mm, vma, addr, new_below);
2527}
2528
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529/* Munmap is split into 2 main parts -- this part which finds
2530 * what needs doing, and the areas themselves, which do the
2531 * work. This now handles partial unmappings.
2532 * Jeremy Fitzhardinge <jeremy@goop.org>
2533 */
2534int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2535{
2536 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002537 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
2539 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2540 return -EINVAL;
2541
vishnu.pscc71aba2014-10-09 15:26:29 -07002542 len = PAGE_ALIGN(len);
2543 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 return -EINVAL;
2545
2546 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002547 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002548 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002550 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002551 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552
2553 /* if it doesn't overlap, we have nothing.. */
2554 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002555 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 return 0;
2557
2558 /*
2559 * If we need to split any vma, do it now to save pain later.
2560 *
2561 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2562 * unmapped vm_area_struct will remain in use: so lower split_vma
2563 * places tmp vma above, and higher split_vma places tmp vma below.
2564 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002565 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002566 int error;
2567
2568 /*
2569 * Make sure that map_count on return from munmap() will
2570 * not exceed its limit; but let map_count go just above
2571 * its limit temporarily, to help free resources as expected.
2572 */
2573 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2574 return -ENOMEM;
2575
2576 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 if (error)
2578 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002579 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 }
2581
2582 /* Does it split the last one? */
2583 last = find_vma(mm, end);
2584 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002585 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 if (error)
2587 return error;
2588 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002589 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590
2591 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002592 * unlock any mlock()ed ranges before detaching vmas
2593 */
2594 if (mm->locked_vm) {
2595 struct vm_area_struct *tmp = vma;
2596 while (tmp && tmp->vm_start < end) {
2597 if (tmp->vm_flags & VM_LOCKED) {
2598 mm->locked_vm -= vma_pages(tmp);
2599 munlock_vma_pages_all(tmp);
2600 }
2601 tmp = tmp->vm_next;
2602 }
2603 }
2604
2605 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 * Remove the vma's, and unmap the actual pages
2607 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002608 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2609 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Dave Hansen1de4fa12014-11-14 07:18:31 -08002611 arch_unmap(mm, vma, start, end);
2612
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002614 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615
2616 return 0;
2617}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Al Virobfce2812012-04-20 21:57:04 -04002619int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002620{
2621 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002622 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002623
2624 down_write(&mm->mmap_sem);
2625 ret = do_munmap(mm, start, len);
2626 up_write(&mm->mmap_sem);
2627 return ret;
2628}
2629EXPORT_SYMBOL(vm_munmap);
2630
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002631SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002634 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635}
2636
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002637
2638/*
2639 * Emulation of deprecated remap_file_pages() syscall.
2640 */
2641SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2642 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2643{
2644
2645 struct mm_struct *mm = current->mm;
2646 struct vm_area_struct *vma;
2647 unsigned long populate = 0;
2648 unsigned long ret = -EINVAL;
2649 struct file *file;
2650
2651 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
2652 "See Documentation/vm/remap_file_pages.txt.\n",
2653 current->comm, current->pid);
2654
2655 if (prot)
2656 return ret;
2657 start = start & PAGE_MASK;
2658 size = size & PAGE_MASK;
2659
2660 if (start + size <= start)
2661 return ret;
2662
2663 /* Does pgoff wrap? */
2664 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2665 return ret;
2666
2667 down_write(&mm->mmap_sem);
2668 vma = find_vma(mm, start);
2669
2670 if (!vma || !(vma->vm_flags & VM_SHARED))
2671 goto out;
2672
2673 if (start < vma->vm_start || start + size > vma->vm_end)
2674 goto out;
2675
2676 if (pgoff == linear_page_index(vma, start)) {
2677 ret = 0;
2678 goto out;
2679 }
2680
2681 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2682 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2683 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2684
2685 flags &= MAP_NONBLOCK;
2686 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2687 if (vma->vm_flags & VM_LOCKED) {
2688 flags |= MAP_LOCKED;
2689 /* drop PG_Mlocked flag for over-mapped range */
2690 munlock_vma_pages_range(vma, start, start + size);
2691 }
2692
2693 file = get_file(vma->vm_file);
2694 ret = do_mmap_pgoff(vma->vm_file, start, size,
2695 prot, flags, pgoff, &populate);
2696 fput(file);
2697out:
2698 up_write(&mm->mmap_sem);
2699 if (populate)
2700 mm_populate(ret, populate);
2701 if (!IS_ERR_VALUE(ret))
2702 ret = 0;
2703 return ret;
2704}
2705
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706static inline void verify_mm_writelocked(struct mm_struct *mm)
2707{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002708#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2710 WARN_ON(1);
2711 up_read(&mm->mmap_sem);
2712 }
2713#endif
2714}
2715
2716/*
2717 * this is really a simplified "do_mmap". it only handles
2718 * anonymous maps. eventually we may be able to do some
2719 * brk-specific accounting here.
2720 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002721static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722{
vishnu.pscc71aba2014-10-09 15:26:29 -07002723 struct mm_struct *mm = current->mm;
2724 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002726 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002728 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
2730 len = PAGE_ALIGN(len);
2731 if (!len)
2732 return addr;
2733
Kirill Korotaev3a459752006-09-07 14:17:04 +04002734 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2735
Al Viro2c6a1012009-12-03 19:40:46 -05002736 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2737 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002738 return error;
2739
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002740 error = mlock_future_check(mm, mm->def_flags, len);
2741 if (error)
2742 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743
2744 /*
2745 * mm->mmap_sem is required to protect against another thread
2746 * changing the mappings in case we sleep.
2747 */
2748 verify_mm_writelocked(mm);
2749
2750 /*
2751 * Clear old maps. this also does some error checking for us
2752 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002753 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2754 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 if (do_munmap(mm, addr, len))
2756 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 }
2758
2759 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002760 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 return -ENOMEM;
2762
2763 if (mm->map_count > sysctl_max_map_count)
2764 return -ENOMEM;
2765
Al Viro191c5422012-02-13 03:58:52 +00002766 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 return -ENOMEM;
2768
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002770 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002771 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002772 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 goto out;
2774
2775 /*
2776 * create a vma struct for an anonymous mapping
2777 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002778 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 if (!vma) {
2780 vm_unacct_memory(len >> PAGE_SHIFT);
2781 return -ENOMEM;
2782 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783
Rik van Riel5beb4932010-03-05 13:42:07 -08002784 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 vma->vm_mm = mm;
2786 vma->vm_start = addr;
2787 vma->vm_end = addr + len;
2788 vma->vm_pgoff = pgoff;
2789 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002790 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 vma_link(mm, vma, prev, rb_link, rb_parent);
2792out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002793 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002795 if (flags & VM_LOCKED)
2796 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002797 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 return addr;
2799}
2800
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002801unsigned long vm_brk(unsigned long addr, unsigned long len)
2802{
2803 struct mm_struct *mm = current->mm;
2804 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002805 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002806
2807 down_write(&mm->mmap_sem);
2808 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002809 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002810 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002811 if (populate)
2812 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002813 return ret;
2814}
2815EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816
2817/* Release all mmaps. */
2818void exit_mmap(struct mm_struct *mm)
2819{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002820 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002821 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 unsigned long nr_accounted = 0;
2823
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002824 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002825 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002826
Rik van Rielba470de2008-10-18 20:26:50 -07002827 if (mm->locked_vm) {
2828 vma = mm->mmap;
2829 while (vma) {
2830 if (vma->vm_flags & VM_LOCKED)
2831 munlock_vma_pages_all(vma);
2832 vma = vma->vm_next;
2833 }
2834 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002835
2836 arch_exit_mmap(mm);
2837
Rik van Rielba470de2008-10-18 20:26:50 -07002838 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002839 if (!vma) /* Can happen if dup_mmap() received an OOM */
2840 return;
2841
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002844 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002845 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002846 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002847 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002848
Hugh Dickins6ee86302013-04-29 15:07:44 -07002849 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002850 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002853 * Walk the list again, actually closing and freeing it,
2854 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002856 while (vma) {
2857 if (vma->vm_flags & VM_ACCOUNT)
2858 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002859 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002860 }
2861 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862}
2863
2864/* Insert vm structure into process list sorted by address
2865 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002866 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002868int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869{
Hugh Dickins6597d782012-10-08 16:29:07 -07002870 struct vm_area_struct *prev;
2871 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872
Chen Gangc9d13f52015-09-08 15:04:08 -07002873 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2874 &prev, &rb_link, &rb_parent))
2875 return -ENOMEM;
2876 if ((vma->vm_flags & VM_ACCOUNT) &&
2877 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2878 return -ENOMEM;
2879
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 /*
2881 * The vm_pgoff of a purely anonymous vma should be irrelevant
2882 * until its first write fault, when page's anon_vma and index
2883 * are set. But now set the vm_pgoff it will almost certainly
2884 * end up with (unless mremap moves it elsewhere before that
2885 * first wfault), so /proc/pid/maps tells a consistent story.
2886 *
2887 * By setting it to reflect the virtual start address of the
2888 * vma, merges and splits can happen in a seamless way, just
2889 * using the existing file pgoff checks and manipulations.
2890 * Similarly in do_mmap_pgoff and in do_brk.
2891 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002892 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 BUG_ON(vma->anon_vma);
2894 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2895 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302896
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 vma_link(mm, vma, prev, rb_link, rb_parent);
2898 return 0;
2899}
2900
2901/*
2902 * Copy the vma structure to a new location in the same mm,
2903 * prior to moving page table entries, to effect an mremap move.
2904 */
2905struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002906 unsigned long addr, unsigned long len, pgoff_t pgoff,
2907 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908{
2909 struct vm_area_struct *vma = *vmap;
2910 unsigned long vma_start = vma->vm_start;
2911 struct mm_struct *mm = vma->vm_mm;
2912 struct vm_area_struct *new_vma, *prev;
2913 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002914 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915
2916 /*
2917 * If anonymous vma has not yet been faulted, update new pgoff
2918 * to match new location, to increase its chance of merging.
2919 */
Oleg Nesterovce757992015-09-08 14:58:34 -07002920 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002922 faulted_in_anon_vma = false;
2923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924
Hugh Dickins6597d782012-10-08 16:29:07 -07002925 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2926 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002928 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2929 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 if (new_vma) {
2931 /*
2932 * Source vma may have been merged into new_vma
2933 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002934 if (unlikely(vma_start >= new_vma->vm_start &&
2935 vma_start < new_vma->vm_end)) {
2936 /*
2937 * The only way we can get a vma_merge with
2938 * self during an mremap is if the vma hasn't
2939 * been faulted in yet and we were allowed to
2940 * reset the dst vma->vm_pgoff to the
2941 * destination address of the mremap to allow
2942 * the merge to happen. mremap must change the
2943 * vm_pgoff linearity between src and dst vmas
2944 * (in turn preventing a vma_merge) to be
2945 * safe. It is only safe to keep the vm_pgoff
2946 * linear if there are no pages mapped yet.
2947 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002948 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002949 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002950 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002951 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002953 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07002954 if (!new_vma)
2955 goto out;
2956 *new_vma = *vma;
2957 new_vma->vm_start = addr;
2958 new_vma->vm_end = addr + len;
2959 new_vma->vm_pgoff = pgoff;
2960 if (vma_dup_policy(vma, new_vma))
2961 goto out_free_vma;
2962 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2963 if (anon_vma_clone(new_vma, vma))
2964 goto out_free_mempol;
2965 if (new_vma->vm_file)
2966 get_file(new_vma->vm_file);
2967 if (new_vma->vm_ops && new_vma->vm_ops->open)
2968 new_vma->vm_ops->open(new_vma);
2969 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2970 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 }
2972 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002973
Chen Gange3975892015-09-08 15:03:38 -07002974out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002975 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07002976out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08002977 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07002978out:
Rik van Riel5beb4932010-03-05 13:42:07 -08002979 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002981
2982/*
2983 * Return true if the calling process may expand its vm space by the passed
2984 * number of pages
2985 */
2986int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2987{
2988 unsigned long cur = mm->total_vm; /* pages */
2989 unsigned long lim;
2990
Jiri Slaby59e99e52010-03-05 13:41:44 -08002991 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002992
2993 if (cur + npages > lim)
2994 return 0;
2995 return 1;
2996}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002997
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002998static int special_mapping_fault(struct vm_area_struct *vma,
2999 struct vm_fault *vmf);
3000
3001/*
3002 * Having a close hook prevents vma merging regardless of flags.
3003 */
3004static void special_mapping_close(struct vm_area_struct *vma)
3005{
3006}
3007
3008static const char *special_mapping_name(struct vm_area_struct *vma)
3009{
3010 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3011}
3012
3013static const struct vm_operations_struct special_mapping_vmops = {
3014 .close = special_mapping_close,
3015 .fault = special_mapping_fault,
3016 .name = special_mapping_name,
3017};
3018
3019static const struct vm_operations_struct legacy_special_mapping_vmops = {
3020 .close = special_mapping_close,
3021 .fault = special_mapping_fault,
3022};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003023
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003024static int special_mapping_fault(struct vm_area_struct *vma,
3025 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003026{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003027 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003028 struct page **pages;
3029
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003030 if (vma->vm_ops == &legacy_special_mapping_vmops)
3031 pages = vma->vm_private_data;
3032 else
3033 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
3034 pages;
3035
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003036 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003037 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003038
3039 if (*pages) {
3040 struct page *page = *pages;
3041 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003042 vmf->page = page;
3043 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003044 }
3045
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003046 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003047}
3048
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003049static struct vm_area_struct *__install_special_mapping(
3050 struct mm_struct *mm,
3051 unsigned long addr, unsigned long len,
3052 unsigned long vm_flags, const struct vm_operations_struct *ops,
3053 void *priv)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003054{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003055 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003056 struct vm_area_struct *vma;
3057
3058 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3059 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003060 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003061
Rik van Riel5beb4932010-03-05 13:42:07 -08003062 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003063 vma->vm_mm = mm;
3064 vma->vm_start = addr;
3065 vma->vm_end = addr + len;
3066
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003067 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003068 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003069
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003070 vma->vm_ops = ops;
3071 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003072
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003073 ret = insert_vm_struct(mm, vma);
3074 if (ret)
3075 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003076
3077 mm->total_vm += len >> PAGE_SHIFT;
3078
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003079 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003080
Stefani Seibold3935ed62014-03-17 23:22:02 +01003081 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003082
3083out:
3084 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003085 return ERR_PTR(ret);
3086}
3087
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003088/*
3089 * Called with mm->mmap_sem held for writing.
3090 * Insert a new vma covering the given region, with the given flags.
3091 * Its pages are supplied by the given array of struct page *.
3092 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3093 * The region past the last page supplied will always produce SIGBUS.
3094 * The array pointer and the pages it points to are assumed to stay alive
3095 * for as long as this mapping might exist.
3096 */
3097struct vm_area_struct *_install_special_mapping(
3098 struct mm_struct *mm,
3099 unsigned long addr, unsigned long len,
3100 unsigned long vm_flags, const struct vm_special_mapping *spec)
3101{
3102 return __install_special_mapping(mm, addr, len, vm_flags,
3103 &special_mapping_vmops, (void *)spec);
3104}
3105
Stefani Seibold3935ed62014-03-17 23:22:02 +01003106int install_special_mapping(struct mm_struct *mm,
3107 unsigned long addr, unsigned long len,
3108 unsigned long vm_flags, struct page **pages)
3109{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003110 struct vm_area_struct *vma = __install_special_mapping(
3111 mm, addr, len, vm_flags, &legacy_special_mapping_vmops,
3112 (void *)pages);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003113
Duan Jiong14bd5b42014-06-04 16:07:05 -07003114 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003115}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003116
3117static DEFINE_MUTEX(mm_all_locks_mutex);
3118
Peter Zijlstra454ed842008-08-11 09:30:25 +02003119static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003120{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003121 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003122 /*
3123 * The LSB of head.next can't change from under us
3124 * because we hold the mm_all_locks_mutex.
3125 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003126 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003127 /*
3128 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003129 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003130 * the same anon_vma we won't take it again.
3131 *
3132 * No need of atomic instructions here, head.next
3133 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003134 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003135 */
3136 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003137 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003138 BUG();
3139 }
3140}
3141
Peter Zijlstra454ed842008-08-11 09:30:25 +02003142static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003143{
3144 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3145 /*
3146 * AS_MM_ALL_LOCKS can't change from under us because
3147 * we hold the mm_all_locks_mutex.
3148 *
3149 * Operations on ->flags have to be atomic because
3150 * even if AS_MM_ALL_LOCKS is stable thanks to the
3151 * mm_all_locks_mutex, there may be other cpus
3152 * changing other bitflags in parallel to us.
3153 */
3154 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3155 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003156 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003157 }
3158}
3159
3160/*
3161 * This operation locks against the VM for all pte/vma/mm related
3162 * operations that could ever happen on a certain mm. This includes
3163 * vmtruncate, try_to_unmap, and all page faults.
3164 *
3165 * The caller must take the mmap_sem in write mode before calling
3166 * mm_take_all_locks(). The caller isn't allowed to release the
3167 * mmap_sem until mm_drop_all_locks() returns.
3168 *
3169 * mmap_sem in write mode is required in order to block all operations
3170 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003171 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003172 * anon_vmas to be associated with existing vmas.
3173 *
3174 * A single task can't take more than one mm_take_all_locks() in a row
3175 * or it would deadlock.
3176 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003177 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003178 * mapping->flags avoid to take the same lock twice, if more than one
3179 * vma in this mm is backed by the same anon_vma or address_space.
3180 *
3181 * We can take all the locks in random order because the VM code
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003182 * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003183 * takes more than one of them in a row. Secondly we're protected
3184 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3185 *
3186 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3187 * that may have to take thousand of locks.
3188 *
3189 * mm_take_all_locks() can fail if it's interrupted by signals.
3190 */
3191int mm_take_all_locks(struct mm_struct *mm)
3192{
3193 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003194 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003195
3196 BUG_ON(down_read_trylock(&mm->mmap_sem));
3197
3198 mutex_lock(&mm_all_locks_mutex);
3199
3200 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3201 if (signal_pending(current))
3202 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003203 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003204 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003205 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003206
3207 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3208 if (signal_pending(current))
3209 goto out_unlock;
3210 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003211 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3212 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003213 }
3214
Kautuk Consul584cff52011-10-31 17:08:59 -07003215 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003216
3217out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003218 mm_drop_all_locks(mm);
3219 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003220}
3221
3222static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3223{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003224 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003225 /*
3226 * The LSB of head.next can't change to 0 from under
3227 * us because we hold the mm_all_locks_mutex.
3228 *
3229 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003230 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003231 * never see our bitflag.
3232 *
3233 * No need of atomic instructions here, head.next
3234 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003235 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003236 */
3237 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003238 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003239 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003240 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003241 }
3242}
3243
3244static void vm_unlock_mapping(struct address_space *mapping)
3245{
3246 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3247 /*
3248 * AS_MM_ALL_LOCKS can't change to 0 from under us
3249 * because we hold the mm_all_locks_mutex.
3250 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003251 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003252 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3253 &mapping->flags))
3254 BUG();
3255 }
3256}
3257
3258/*
3259 * The mmap_sem cannot be released by the caller until
3260 * mm_drop_all_locks() returns.
3261 */
3262void mm_drop_all_locks(struct mm_struct *mm)
3263{
3264 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003265 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003266
3267 BUG_ON(down_read_trylock(&mm->mmap_sem));
3268 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3269
3270 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3271 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003272 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3273 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003274 if (vma->vm_file && vma->vm_file->f_mapping)
3275 vm_unlock_mapping(vma->vm_file->f_mapping);
3276 }
3277
3278 mutex_unlock(&mm_all_locks_mutex);
3279}
David Howells8feae132009-01-08 12:04:47 +00003280
3281/*
3282 * initialise the VMA slab
3283 */
3284void __init mmap_init(void)
3285{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003286 int ret;
3287
Tejun Heo908c7f12014-09-08 09:51:29 +09003288 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003289 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003290}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003291
3292/*
3293 * Initialise sysctl_user_reserve_kbytes.
3294 *
3295 * This is intended to prevent a user from starting a single memory hogging
3296 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3297 * mode.
3298 *
3299 * The default value is min(3% of free memory, 128MB)
3300 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3301 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003302static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003303{
3304 unsigned long free_kbytes;
3305
3306 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3307
3308 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3309 return 0;
3310}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003311subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003312
3313/*
3314 * Initialise sysctl_admin_reserve_kbytes.
3315 *
3316 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3317 * to log in and kill a memory hogging process.
3318 *
3319 * Systems with more than 256MB will reserve 8MB, enough to recover
3320 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3321 * only reserve 3% of free pages by default.
3322 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003323static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003324{
3325 unsigned long free_kbytes;
3326
3327 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3328
3329 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3330 return 0;
3331}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003332subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003333
3334/*
3335 * Reinititalise user and admin reserves if memory is added or removed.
3336 *
3337 * The default user reserve max is 128MB, and the default max for the
3338 * admin reserve is 8MB. These are usually, but not always, enough to
3339 * enable recovery from a memory hogging process using login/sshd, a shell,
3340 * and tools like top. It may make sense to increase or even disable the
3341 * reserve depending on the existence of swap or variations in the recovery
3342 * tools. So, the admin may have changed them.
3343 *
3344 * If memory is added and the reserves have been eliminated or increased above
3345 * the default max, then we'll trust the admin.
3346 *
3347 * If memory is removed and there isn't enough free memory, then we
3348 * need to reset the reserves.
3349 *
3350 * Otherwise keep the reserve set by the admin.
3351 */
3352static int reserve_mem_notifier(struct notifier_block *nb,
3353 unsigned long action, void *data)
3354{
3355 unsigned long tmp, free_kbytes;
3356
3357 switch (action) {
3358 case MEM_ONLINE:
3359 /* Default max is 128MB. Leave alone if modified by operator. */
3360 tmp = sysctl_user_reserve_kbytes;
3361 if (0 < tmp && tmp < (1UL << 17))
3362 init_user_reserve();
3363
3364 /* Default max is 8MB. Leave alone if modified by operator. */
3365 tmp = sysctl_admin_reserve_kbytes;
3366 if (0 < tmp && tmp < (1UL << 13))
3367 init_admin_reserve();
3368
3369 break;
3370 case MEM_OFFLINE:
3371 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3372
3373 if (sysctl_user_reserve_kbytes > free_kbytes) {
3374 init_user_reserve();
3375 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3376 sysctl_user_reserve_kbytes);
3377 }
3378
3379 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3380 init_admin_reserve();
3381 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3382 sysctl_admin_reserve_kbytes);
3383 }
3384 break;
3385 default:
3386 break;
3387 }
3388 return NOTIFY_OK;
3389}
3390
3391static struct notifier_block reserve_mem_nb = {
3392 .notifier_call = reserve_mem_notifier,
3393};
3394
3395static int __meminit init_reserve_notifier(void)
3396{
3397 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003398 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003399
3400 return 0;
3401}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003402subsys_initcall(init_reserve_notifier);