blob: b6be3249f0a923cf1735405726f17cc4de8085a7 [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 */
1263
Al Viroe3fc6292012-05-30 20:08:42 -04001264unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001266 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001267 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268{
vishnu.pscc71aba2014-10-09 15:26:29 -07001269 struct mm_struct *mm = current->mm;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001270 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Michel Lespinasse41badc12013-02-22 16:32:47 -08001272 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001273
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001274 if (!len)
1275 return -EINVAL;
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 /*
1278 * Does the application expect PROT_READ to imply PROT_EXEC?
1279 *
1280 * (the exception is when the underlying filesystem is noexec
1281 * mounted, in which case we dont add PROT_EXEC.)
1282 */
1283 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001284 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 prot |= PROT_EXEC;
1286
Eric Paris7cd94142007-11-26 18:47:40 -05001287 if (!(flags & MAP_FIXED))
1288 addr = round_hint_to_min(addr);
1289
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 /* Careful about overflows.. */
1291 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001292 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 return -ENOMEM;
1294
1295 /* offset overflow? */
1296 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001297 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
1299 /* Too many mappings? */
1300 if (mm->map_count > sysctl_max_map_count)
1301 return -ENOMEM;
1302
1303 /* Obtain the address to map to. we verify (or select) it and ensure
1304 * that it represents a valid section of the address space.
1305 */
1306 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1307 if (addr & ~PAGE_MASK)
1308 return addr;
1309
1310 /* Do simple checking here so the lower-level routines won't have
1311 * to. we assume access permissions have been handled by the open
1312 * of the memory object, so we don't do any here.
1313 */
1314 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1315 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1316
Huang Shijiecdf7b342009-09-21 17:03:36 -07001317 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 if (!can_do_mlock())
1319 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001320
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001321 if (mlock_future_check(mm, vm_flags, len))
1322 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001325 struct inode *inode = file_inode(file);
1326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 switch (flags & MAP_TYPE) {
1328 case MAP_SHARED:
1329 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1330 return -EACCES;
1331
1332 /*
1333 * Make sure we don't allow writing to an append-only
1334 * file..
1335 */
1336 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1337 return -EACCES;
1338
1339 /*
1340 * Make sure there are no mandatory locks on the file.
1341 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001342 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 return -EAGAIN;
1344
1345 vm_flags |= VM_SHARED | VM_MAYSHARE;
1346 if (!(file->f_mode & FMODE_WRITE))
1347 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1348
1349 /* fall through */
1350 case MAP_PRIVATE:
1351 if (!(file->f_mode & FMODE_READ))
1352 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001353 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001354 if (vm_flags & VM_EXEC)
1355 return -EPERM;
1356 vm_flags &= ~VM_MAYEXEC;
1357 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001358
Al Viro72c2d532013-09-22 16:27:52 -04001359 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001360 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001361 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1362 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 break;
1364
1365 default:
1366 return -EINVAL;
1367 }
1368 } else {
1369 switch (flags & MAP_TYPE) {
1370 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001371 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1372 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001373 /*
1374 * Ignore pgoff.
1375 */
1376 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 vm_flags |= VM_SHARED | VM_MAYSHARE;
1378 break;
1379 case MAP_PRIVATE:
1380 /*
1381 * Set pgoff according to addr for anon_vma.
1382 */
1383 pgoff = addr >> PAGE_SHIFT;
1384 break;
1385 default:
1386 return -EINVAL;
1387 }
1388 }
1389
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001390 /*
1391 * Set 'VM_NORESERVE' if we should not account for the
1392 * memory use of this mapping.
1393 */
1394 if (flags & MAP_NORESERVE) {
1395 /* We honor MAP_NORESERVE if allowed to overcommit */
1396 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1397 vm_flags |= VM_NORESERVE;
1398
1399 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1400 if (file && is_file_hugepages(file))
1401 vm_flags |= VM_NORESERVE;
1402 }
1403
1404 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001405 if (!IS_ERR_VALUE(addr) &&
1406 ((vm_flags & VM_LOCKED) ||
1407 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001408 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001409 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001410}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001411
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001412SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1413 unsigned long, prot, unsigned long, flags,
1414 unsigned long, fd, unsigned long, pgoff)
1415{
1416 struct file *file = NULL;
1417 unsigned long retval = -EBADF;
1418
1419 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001420 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001421 file = fget(fd);
1422 if (!file)
1423 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001424 if (is_file_hugepages(file))
1425 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001426 retval = -EINVAL;
1427 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1428 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001429 } else if (flags & MAP_HUGETLB) {
1430 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001431 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001432
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001433 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001434 if (!hs)
1435 return -EINVAL;
1436
1437 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001438 /*
1439 * VM_NORESERVE is used because the reservations will be
1440 * taken when vm_ops->mmap() is called
1441 * A dummy user value is used because we are not locking
1442 * memory so no accounting is necessary
1443 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001444 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001445 VM_NORESERVE,
1446 &user, HUGETLB_ANONHUGE_INODE,
1447 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001448 if (IS_ERR(file))
1449 return PTR_ERR(file);
1450 }
1451
1452 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1453
Al Viroeb36c582012-05-30 20:17:35 -04001454 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001455out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001456 if (file)
1457 fput(file);
1458out:
1459 return retval;
1460}
1461
Christoph Hellwiga4679372010-03-10 15:21:15 -08001462#ifdef __ARCH_WANT_SYS_OLD_MMAP
1463struct mmap_arg_struct {
1464 unsigned long addr;
1465 unsigned long len;
1466 unsigned long prot;
1467 unsigned long flags;
1468 unsigned long fd;
1469 unsigned long offset;
1470};
1471
1472SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1473{
1474 struct mmap_arg_struct a;
1475
1476 if (copy_from_user(&a, arg, sizeof(a)))
1477 return -EFAULT;
1478 if (a.offset & ~PAGE_MASK)
1479 return -EINVAL;
1480
1481 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1482 a.offset >> PAGE_SHIFT);
1483}
1484#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1485
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001486/*
1487 * Some shared mappigns will want the pages marked read-only
1488 * to track write events. If so, we'll downgrade vm_page_prot
1489 * to the private version (using protection_map[] without the
1490 * VM_SHARED bit).
1491 */
1492int vma_wants_writenotify(struct vm_area_struct *vma)
1493{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001494 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001495
1496 /* If it was private or non-writable, the write bit is already clear */
1497 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1498 return 0;
1499
1500 /* The backer wishes to know when pages are first written to? */
1501 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1502 return 1;
1503
Peter Feiner64e45502014-10-13 15:55:46 -07001504 /* The open routine did something to the protections that pgprot_modify
1505 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001506 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001507 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001508 return 0;
1509
Peter Feiner64e45502014-10-13 15:55:46 -07001510 /* Do we need to track softdirty? */
1511 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1512 return 1;
1513
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001514 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001515 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001516 return 0;
1517
1518 /* Can the mapping track the dirty pages? */
1519 return vma->vm_file && vma->vm_file->f_mapping &&
1520 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1521}
1522
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001523/*
1524 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001525 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001526 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001527static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001528{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001529 /*
1530 * hugetlb has its own accounting separate from the core VM
1531 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1532 */
1533 if (file && is_file_hugepages(file))
1534 return 0;
1535
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001536 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1537}
1538
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001539unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001540 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001541{
1542 struct mm_struct *mm = current->mm;
1543 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001544 int error;
1545 struct rb_node **rb_link, *rb_parent;
1546 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001547
Cyril Hrubise8420a82013-04-29 15:08:33 -07001548 /* Check against address space limit. */
1549 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1550 unsigned long nr_pages;
1551
1552 /*
1553 * MAP_FIXED may remove pages of mappings that intersects with
1554 * requested mapping. Account for the pages it would unmap.
1555 */
1556 if (!(vm_flags & MAP_FIXED))
1557 return -ENOMEM;
1558
1559 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1560
1561 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1562 return -ENOMEM;
1563 }
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 /* Clear old maps */
1566 error = -ENOMEM;
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001567 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1568 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 if (do_munmap(mm, addr, len))
1570 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
1572
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001573 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001574 * Private writable mapping: check memory availability
1575 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001576 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001577 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001578 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001579 return -ENOMEM;
1580 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 }
1582
1583 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001584 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001586 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1587 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001588 if (vma)
1589 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 /*
1592 * Determine the object being mapped and call the appropriate
1593 * specific mapper. the address has already been validated, but
1594 * not unmapped, but the maps are removed from the list.
1595 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001596 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 if (!vma) {
1598 error = -ENOMEM;
1599 goto unacct_error;
1600 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
1602 vma->vm_mm = mm;
1603 vma->vm_start = addr;
1604 vma->vm_end = addr + len;
1605 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001606 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001608 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
1610 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 if (vm_flags & VM_DENYWRITE) {
1612 error = deny_write_access(file);
1613 if (error)
1614 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001616 if (vm_flags & VM_SHARED) {
1617 error = mapping_map_writable(file->f_mapping);
1618 if (error)
1619 goto allow_write_and_free_vma;
1620 }
1621
1622 /* ->mmap() can change vma->vm_file, but must guarantee that
1623 * vma_link() below can deny write-access if VM_DENYWRITE is set
1624 * and map writably if VM_SHARED is set. This usually means the
1625 * new file must not have been exposed to user-space, yet.
1626 */
Al Virocb0942b2012-08-27 14:48:26 -04001627 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 error = file->f_op->mmap(file, vma);
1629 if (error)
1630 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001631
1632 /* Can addr have changed??
1633 *
1634 * Answer: Yes, several device drivers can do it in their
1635 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001636 * Bug: If addr is changed, prev, rb_link, rb_parent should
1637 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001638 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001639 WARN_ON_ONCE(addr != vma->vm_start);
1640
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001641 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001642 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 } else if (vm_flags & VM_SHARED) {
1644 error = shmem_zero_setup(vma);
1645 if (error)
1646 goto free_vma;
1647 }
1648
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001649 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001650 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001651 if (file) {
1652 if (vm_flags & VM_SHARED)
1653 mapping_unmap_writable(file->f_mapping);
1654 if (vm_flags & VM_DENYWRITE)
1655 allow_write_access(file);
1656 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001657 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001658out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001659 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001660
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001661 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001663 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1664 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001665 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001666 else
1667 vma->vm_flags &= ~VM_LOCKED;
1668 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301669
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001670 if (file)
1671 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301672
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001673 /*
1674 * New (or expanded) vma always get soft dirty status.
1675 * Otherwise user-space soft-dirty page tracker won't
1676 * be able to distinguish situation when vma area unmapped,
1677 * then new mapped in-place (which must be aimed as
1678 * a completely new data area).
1679 */
1680 vma->vm_flags |= VM_SOFTDIRTY;
1681
Peter Feiner64e45502014-10-13 15:55:46 -07001682 vma_set_page_prot(vma);
1683
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 return addr;
1685
1686unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 vma->vm_file = NULL;
1688 fput(file);
1689
1690 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001691 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1692 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001693 if (vm_flags & VM_SHARED)
1694 mapping_unmap_writable(file->f_mapping);
1695allow_write_and_free_vma:
1696 if (vm_flags & VM_DENYWRITE)
1697 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698free_vma:
1699 kmem_cache_free(vm_area_cachep, vma);
1700unacct_error:
1701 if (charged)
1702 vm_unacct_memory(charged);
1703 return error;
1704}
1705
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001706unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1707{
1708 /*
1709 * We implement the search by looking for an rbtree node that
1710 * immediately follows a suitable gap. That is,
1711 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1712 * - gap_end = vma->vm_start >= info->low_limit + length;
1713 * - gap_end - gap_start >= length
1714 */
1715
1716 struct mm_struct *mm = current->mm;
1717 struct vm_area_struct *vma;
1718 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1719
1720 /* Adjust search length to account for worst case alignment overhead */
1721 length = info->length + info->align_mask;
1722 if (length < info->length)
1723 return -ENOMEM;
1724
1725 /* Adjust search limits by the desired length */
1726 if (info->high_limit < length)
1727 return -ENOMEM;
1728 high_limit = info->high_limit - length;
1729
1730 if (info->low_limit > high_limit)
1731 return -ENOMEM;
1732 low_limit = info->low_limit + length;
1733
1734 /* Check if rbtree root looks promising */
1735 if (RB_EMPTY_ROOT(&mm->mm_rb))
1736 goto check_highest;
1737 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1738 if (vma->rb_subtree_gap < length)
1739 goto check_highest;
1740
1741 while (true) {
1742 /* Visit left subtree if it looks promising */
1743 gap_end = vma->vm_start;
1744 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1745 struct vm_area_struct *left =
1746 rb_entry(vma->vm_rb.rb_left,
1747 struct vm_area_struct, vm_rb);
1748 if (left->rb_subtree_gap >= length) {
1749 vma = left;
1750 continue;
1751 }
1752 }
1753
1754 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1755check_current:
1756 /* Check if current node has a suitable gap */
1757 if (gap_start > high_limit)
1758 return -ENOMEM;
1759 if (gap_end >= low_limit && gap_end - gap_start >= length)
1760 goto found;
1761
1762 /* Visit right subtree if it looks promising */
1763 if (vma->vm_rb.rb_right) {
1764 struct vm_area_struct *right =
1765 rb_entry(vma->vm_rb.rb_right,
1766 struct vm_area_struct, vm_rb);
1767 if (right->rb_subtree_gap >= length) {
1768 vma = right;
1769 continue;
1770 }
1771 }
1772
1773 /* Go back up the rbtree to find next candidate node */
1774 while (true) {
1775 struct rb_node *prev = &vma->vm_rb;
1776 if (!rb_parent(prev))
1777 goto check_highest;
1778 vma = rb_entry(rb_parent(prev),
1779 struct vm_area_struct, vm_rb);
1780 if (prev == vma->vm_rb.rb_left) {
1781 gap_start = vma->vm_prev->vm_end;
1782 gap_end = vma->vm_start;
1783 goto check_current;
1784 }
1785 }
1786 }
1787
1788check_highest:
1789 /* Check highest gap, which does not precede any rbtree node */
1790 gap_start = mm->highest_vm_end;
1791 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1792 if (gap_start > high_limit)
1793 return -ENOMEM;
1794
1795found:
1796 /* We found a suitable gap. Clip it with the original low_limit. */
1797 if (gap_start < info->low_limit)
1798 gap_start = info->low_limit;
1799
1800 /* Adjust gap address to the desired alignment */
1801 gap_start += (info->align_offset - gap_start) & info->align_mask;
1802
1803 VM_BUG_ON(gap_start + info->length > info->high_limit);
1804 VM_BUG_ON(gap_start + info->length > gap_end);
1805 return gap_start;
1806}
1807
1808unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1809{
1810 struct mm_struct *mm = current->mm;
1811 struct vm_area_struct *vma;
1812 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1813
1814 /* Adjust search length to account for worst case alignment overhead */
1815 length = info->length + info->align_mask;
1816 if (length < info->length)
1817 return -ENOMEM;
1818
1819 /*
1820 * Adjust search limits by the desired length.
1821 * See implementation comment at top of unmapped_area().
1822 */
1823 gap_end = info->high_limit;
1824 if (gap_end < length)
1825 return -ENOMEM;
1826 high_limit = gap_end - length;
1827
1828 if (info->low_limit > high_limit)
1829 return -ENOMEM;
1830 low_limit = info->low_limit + length;
1831
1832 /* Check highest gap, which does not precede any rbtree node */
1833 gap_start = mm->highest_vm_end;
1834 if (gap_start <= high_limit)
1835 goto found_highest;
1836
1837 /* Check if rbtree root looks promising */
1838 if (RB_EMPTY_ROOT(&mm->mm_rb))
1839 return -ENOMEM;
1840 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1841 if (vma->rb_subtree_gap < length)
1842 return -ENOMEM;
1843
1844 while (true) {
1845 /* Visit right subtree if it looks promising */
1846 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1847 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1848 struct vm_area_struct *right =
1849 rb_entry(vma->vm_rb.rb_right,
1850 struct vm_area_struct, vm_rb);
1851 if (right->rb_subtree_gap >= length) {
1852 vma = right;
1853 continue;
1854 }
1855 }
1856
1857check_current:
1858 /* Check if current node has a suitable gap */
1859 gap_end = vma->vm_start;
1860 if (gap_end < low_limit)
1861 return -ENOMEM;
1862 if (gap_start <= high_limit && gap_end - gap_start >= length)
1863 goto found;
1864
1865 /* Visit left subtree if it looks promising */
1866 if (vma->vm_rb.rb_left) {
1867 struct vm_area_struct *left =
1868 rb_entry(vma->vm_rb.rb_left,
1869 struct vm_area_struct, vm_rb);
1870 if (left->rb_subtree_gap >= length) {
1871 vma = left;
1872 continue;
1873 }
1874 }
1875
1876 /* Go back up the rbtree to find next candidate node */
1877 while (true) {
1878 struct rb_node *prev = &vma->vm_rb;
1879 if (!rb_parent(prev))
1880 return -ENOMEM;
1881 vma = rb_entry(rb_parent(prev),
1882 struct vm_area_struct, vm_rb);
1883 if (prev == vma->vm_rb.rb_right) {
1884 gap_start = vma->vm_prev ?
1885 vma->vm_prev->vm_end : 0;
1886 goto check_current;
1887 }
1888 }
1889 }
1890
1891found:
1892 /* We found a suitable gap. Clip it with the original high_limit. */
1893 if (gap_end > info->high_limit)
1894 gap_end = info->high_limit;
1895
1896found_highest:
1897 /* Compute highest gap address at the desired alignment */
1898 gap_end -= info->length;
1899 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1900
1901 VM_BUG_ON(gap_end < info->low_limit);
1902 VM_BUG_ON(gap_end < gap_start);
1903 return gap_end;
1904}
1905
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906/* Get an address range which is currently unmapped.
1907 * For shmat() with addr=0.
1908 *
1909 * Ugly calling convention alert:
1910 * Return value with the low bits set means error value,
1911 * ie
1912 * if (ret & ~PAGE_MASK)
1913 * error = ret;
1914 *
1915 * This function "knows" that -ENOMEM has the bits set.
1916 */
1917#ifndef HAVE_ARCH_UNMAPPED_AREA
1918unsigned long
1919arch_get_unmapped_area(struct file *filp, unsigned long addr,
1920 unsigned long len, unsigned long pgoff, unsigned long flags)
1921{
1922 struct mm_struct *mm = current->mm;
1923 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001924 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001926 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 return -ENOMEM;
1928
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001929 if (flags & MAP_FIXED)
1930 return addr;
1931
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 if (addr) {
1933 addr = PAGE_ALIGN(addr);
1934 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001935 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 (!vma || addr + len <= vma->vm_start))
1937 return addr;
1938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001940 info.flags = 0;
1941 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001942 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001943 info.high_limit = TASK_SIZE;
1944 info.align_mask = 0;
1945 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946}
vishnu.pscc71aba2014-10-09 15:26:29 -07001947#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949/*
1950 * This mmap-allocator allocates new areas top-down from below the
1951 * stack's low limit (the base):
1952 */
1953#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1954unsigned long
1955arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1956 const unsigned long len, const unsigned long pgoff,
1957 const unsigned long flags)
1958{
1959 struct vm_area_struct *vma;
1960 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001961 unsigned long addr = addr0;
1962 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
1964 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001965 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 return -ENOMEM;
1967
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001968 if (flags & MAP_FIXED)
1969 return addr;
1970
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 /* requesting a specific address */
1972 if (addr) {
1973 addr = PAGE_ALIGN(addr);
1974 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001975 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 (!vma || addr + len <= vma->vm_start))
1977 return addr;
1978 }
1979
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001980 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1981 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001982 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001983 info.high_limit = mm->mmap_base;
1984 info.align_mask = 0;
1985 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 /*
1988 * A failed mmap() very likely causes application failure,
1989 * so fall back to the bottom-up function here. This scenario
1990 * can happen with large stack limits and large mmap()
1991 * allocations.
1992 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001993 if (addr & ~PAGE_MASK) {
1994 VM_BUG_ON(addr != -ENOMEM);
1995 info.flags = 0;
1996 info.low_limit = TASK_UNMAPPED_BASE;
1997 info.high_limit = TASK_SIZE;
1998 addr = vm_unmapped_area(&info);
1999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
2001 return addr;
2002}
2003#endif
2004
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005unsigned long
2006get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2007 unsigned long pgoff, unsigned long flags)
2008{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002009 unsigned long (*get_area)(struct file *, unsigned long,
2010 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Al Viro9206de92009-12-03 15:23:11 -05002012 unsigned long error = arch_mmap_check(addr, len, flags);
2013 if (error)
2014 return error;
2015
2016 /* Careful about overflows.. */
2017 if (len > TASK_SIZE)
2018 return -ENOMEM;
2019
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002020 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04002021 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002022 get_area = file->f_op->get_unmapped_area;
2023 addr = get_area(file, addr, len, pgoff, flags);
2024 if (IS_ERR_VALUE(addr))
2025 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002027 if (addr > TASK_SIZE - len)
2028 return -ENOMEM;
2029 if (addr & ~PAGE_MASK)
2030 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002031
Al Viro9ac4ed42012-05-30 17:13:15 -04002032 addr = arch_rebalance_pgtables(addr, len);
2033 error = security_mmap_addr(addr);
2034 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035}
2036
2037EXPORT_SYMBOL(get_unmapped_area);
2038
2039/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002040struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002042 struct rb_node *rb_node;
2043 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002045 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002046 vma = vmacache_find(mm, addr);
2047 if (likely(vma))
2048 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002050 rb_node = mm->mm_rb.rb_node;
2051 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002053 while (rb_node) {
2054 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002056 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002057
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002058 if (tmp->vm_end > addr) {
2059 vma = tmp;
2060 if (tmp->vm_start <= addr)
2061 break;
2062 rb_node = rb_node->rb_left;
2063 } else
2064 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002066
2067 if (vma)
2068 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 return vma;
2070}
2071
2072EXPORT_SYMBOL(find_vma);
2073
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002074/*
2075 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002076 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077struct vm_area_struct *
2078find_vma_prev(struct mm_struct *mm, unsigned long addr,
2079 struct vm_area_struct **pprev)
2080{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002081 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002083 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002084 if (vma) {
2085 *pprev = vma->vm_prev;
2086 } else {
2087 struct rb_node *rb_node = mm->mm_rb.rb_node;
2088 *pprev = NULL;
2089 while (rb_node) {
2090 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2091 rb_node = rb_node->rb_right;
2092 }
2093 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002094 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095}
2096
2097/*
2098 * Verify that the stack growth is acceptable and
2099 * update accounting. This is shared with both the
2100 * grow-up and grow-down cases.
2101 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002102static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
2104 struct mm_struct *mm = vma->vm_mm;
2105 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08002106 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
2108 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002109 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 return -ENOMEM;
2111
2112 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08002113 actual_size = size;
2114 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2115 actual_size -= PAGE_SIZE;
Jason Low4db0c3c2015-04-15 16:14:08 -07002116 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 return -ENOMEM;
2118
2119 /* mlock limit tests */
2120 if (vma->vm_flags & VM_LOCKED) {
2121 unsigned long locked;
2122 unsigned long limit;
2123 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002124 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002125 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 if (locked > limit && !capable(CAP_IPC_LOCK))
2127 return -ENOMEM;
2128 }
2129
Adam Litke0d59a012007-01-30 14:35:39 -08002130 /* Check to ensure the stack will not grow into a hugetlb-only region */
2131 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2132 vma->vm_end - size;
2133 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2134 return -EFAULT;
2135
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 /*
2137 * Overcommit.. This must be the final test, as it will
2138 * update security statistics.
2139 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002140 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 return -ENOMEM;
2142
2143 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (vma->vm_flags & VM_LOCKED)
2145 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002146 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 return 0;
2148}
2149
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002150#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002152 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2153 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002155int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156{
2157 int error;
2158
2159 if (!(vma->vm_flags & VM_GROWSUP))
2160 return -EFAULT;
2161
2162 /*
2163 * We must make sure the anon_vma is allocated
2164 * so that the anon_vma locking is not a noop.
2165 */
2166 if (unlikely(anon_vma_prepare(vma)))
2167 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002168 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
2170 /*
2171 * vma->vm_start/vm_end cannot change under us because the caller
2172 * is required to hold the mmap_sem in read mode. We need the
2173 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002174 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 */
Helge Deller06b32f32006-12-19 19:28:33 +01002176 if (address < PAGE_ALIGN(address+4))
2177 address = PAGE_ALIGN(address+4);
2178 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002179 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002180 return -ENOMEM;
2181 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 error = 0;
2183
2184 /* Somebody else might have raced and expanded it already */
2185 if (address > vma->vm_end) {
2186 unsigned long size, grow;
2187
2188 size = address - vma->vm_start;
2189 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2190
Hugh Dickins42c36f62011-05-09 17:44:42 -07002191 error = -ENOMEM;
2192 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2193 error = acct_stack_growth(vma, size, grow);
2194 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002195 /*
2196 * vma_gap_update() doesn't support concurrent
2197 * updates, but we only hold a shared mmap_sem
2198 * lock here, so we need to protect against
2199 * concurrent vma expansions.
2200 * vma_lock_anon_vma() doesn't help here, as
2201 * we don't guarantee that all growable vmas
2202 * in a mm share the same root anon vma.
2203 * So, we reuse mm->page_table_lock to guard
2204 * against concurrent vma expansions.
2205 */
2206 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002207 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002208 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002209 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002210 if (vma->vm_next)
2211 vma_gap_update(vma->vm_next);
2212 else
2213 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002214 spin_unlock(&vma->vm_mm->page_table_lock);
2215
Hugh Dickins42c36f62011-05-09 17:44:42 -07002216 perf_event_mmap(vma);
2217 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002220 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002221 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002222 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 return error;
2224}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002225#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2226
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227/*
2228 * vma is the first one with address < vma->vm_start. Have to extend vma.
2229 */
Michal Hockod05f3162011-05-24 17:11:44 -07002230int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002231 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232{
2233 int error;
2234
2235 /*
2236 * We must make sure the anon_vma is allocated
2237 * so that the anon_vma locking is not a noop.
2238 */
2239 if (unlikely(anon_vma_prepare(vma)))
2240 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002241
2242 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002243 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002244 if (error)
2245 return error;
2246
Rik van Rielbb4a3402010-08-09 17:18:37 -07002247 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
2249 /*
2250 * vma->vm_start/vm_end cannot change under us because the caller
2251 * is required to hold the mmap_sem in read mode. We need the
2252 * anon_vma lock to serialize against concurrent expand_stacks.
2253 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254
2255 /* Somebody else might have raced and expanded it already */
2256 if (address < vma->vm_start) {
2257 unsigned long size, grow;
2258
2259 size = vma->vm_end - address;
2260 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2261
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002262 error = -ENOMEM;
2263 if (grow <= vma->vm_pgoff) {
2264 error = acct_stack_growth(vma, size, grow);
2265 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002266 /*
2267 * vma_gap_update() doesn't support concurrent
2268 * updates, but we only hold a shared mmap_sem
2269 * lock here, so we need to protect against
2270 * concurrent vma expansions.
2271 * vma_lock_anon_vma() doesn't help here, as
2272 * we don't guarantee that all growable vmas
2273 * in a mm share the same root anon vma.
2274 * So, we reuse mm->page_table_lock to guard
2275 * against concurrent vma expansions.
2276 */
2277 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002278 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002279 vma->vm_start = address;
2280 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002281 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002282 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002283 spin_unlock(&vma->vm_mm->page_table_lock);
2284
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002285 perf_event_mmap(vma);
2286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 }
2288 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002289 vma_unlock_anon_vma(vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002290 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002291 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 return error;
2293}
2294
Linus Torvalds09884962013-02-27 08:36:04 -08002295/*
2296 * Note how expand_stack() refuses to expand the stack all the way to
2297 * abut the next virtual mapping, *unless* that mapping itself is also
2298 * a stack mapping. We want to leave room for a guard page, after all
2299 * (the guard page itself is not added here, that is done by the
2300 * actual page faulting logic)
2301 *
2302 * This matches the behavior of the guard page logic (see mm/memory.c:
2303 * check_stack_guard_page()), which only allows the guard page to be
2304 * removed under these circumstances.
2305 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002306#ifdef CONFIG_STACK_GROWSUP
2307int expand_stack(struct vm_area_struct *vma, unsigned long address)
2308{
Linus Torvalds09884962013-02-27 08:36:04 -08002309 struct vm_area_struct *next;
2310
2311 address &= PAGE_MASK;
2312 next = vma->vm_next;
2313 if (next && next->vm_start == address + PAGE_SIZE) {
2314 if (!(next->vm_flags & VM_GROWSUP))
2315 return -ENOMEM;
2316 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002317 return expand_upwards(vma, address);
2318}
2319
2320struct vm_area_struct *
2321find_extend_vma(struct mm_struct *mm, unsigned long addr)
2322{
2323 struct vm_area_struct *vma, *prev;
2324
2325 addr &= PAGE_MASK;
2326 vma = find_vma_prev(mm, addr, &prev);
2327 if (vma && (vma->vm_start <= addr))
2328 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002329 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002330 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002331 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002332 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002333 return prev;
2334}
2335#else
2336int expand_stack(struct vm_area_struct *vma, unsigned long address)
2337{
Linus Torvalds09884962013-02-27 08:36:04 -08002338 struct vm_area_struct *prev;
2339
2340 address &= PAGE_MASK;
2341 prev = vma->vm_prev;
2342 if (prev && prev->vm_end == address) {
2343 if (!(prev->vm_flags & VM_GROWSDOWN))
2344 return -ENOMEM;
2345 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002346 return expand_downwards(vma, address);
2347}
2348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002350find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351{
vishnu.pscc71aba2014-10-09 15:26:29 -07002352 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 unsigned long start;
2354
2355 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002356 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 if (!vma)
2358 return NULL;
2359 if (vma->vm_start <= addr)
2360 return vma;
2361 if (!(vma->vm_flags & VM_GROWSDOWN))
2362 return NULL;
2363 start = vma->vm_start;
2364 if (expand_stack(vma, addr))
2365 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002366 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002367 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 return vma;
2369}
2370#endif
2371
Jesse Barnese1d6d012014-12-12 16:55:27 -08002372EXPORT_SYMBOL_GPL(find_extend_vma);
2373
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002375 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002377 *
2378 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002380static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002382 unsigned long nr_accounted = 0;
2383
Hugh Dickins365e9c872005-10-29 18:16:18 -07002384 /* Update high watermark before we lower total_vm */
2385 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002387 long nrpages = vma_pages(vma);
2388
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002389 if (vma->vm_flags & VM_ACCOUNT)
2390 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002391 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002392 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002393 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002394 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 validate_mm(mm);
2396}
2397
2398/*
2399 * Get rid of page table information in the indicated region.
2400 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002401 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 */
2403static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002404 struct vm_area_struct *vma, struct vm_area_struct *prev,
2405 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406{
vishnu.pscc71aba2014-10-09 15:26:29 -07002407 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002408 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
2410 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002411 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002412 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002413 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002414 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002415 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002416 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417}
2418
2419/*
2420 * Create a list of vma's touched by the unmap, removing them from the mm's
2421 * vma list as we go..
2422 */
2423static void
2424detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2425 struct vm_area_struct *prev, unsigned long end)
2426{
2427 struct vm_area_struct **insertion_point;
2428 struct vm_area_struct *tail_vma = NULL;
2429
2430 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002431 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002433 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 mm->map_count--;
2435 tail_vma = vma;
2436 vma = vma->vm_next;
2437 } while (vma && vma->vm_start < end);
2438 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002439 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002440 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002441 vma_gap_update(vma);
2442 } else
2443 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002445
2446 /* Kill the cache */
2447 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448}
2449
2450/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002451 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2452 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002454static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 unsigned long addr, int new_below)
2456{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002458 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
Andi Kleena5516432008-07-23 21:27:41 -07002460 if (is_vm_hugetlb_page(vma) && (addr &
2461 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 return -EINVAL;
2463
Christoph Lametere94b1762006-12-06 20:33:17 -08002464 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002466 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
2468 /* most fields are the same, copy all, and then fixup */
2469 *new = *vma;
2470
Rik van Riel5beb4932010-03-05 13:42:07 -08002471 INIT_LIST_HEAD(&new->anon_vma_chain);
2472
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 if (new_below)
2474 new->vm_end = addr;
2475 else {
2476 new->vm_start = addr;
2477 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2478 }
2479
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002480 err = vma_dup_policy(vma, new);
2481 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002482 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002484 err = anon_vma_clone(new, vma);
2485 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002486 goto out_free_mpol;
2487
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002488 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 get_file(new->vm_file);
2490
2491 if (new->vm_ops && new->vm_ops->open)
2492 new->vm_ops->open(new);
2493
2494 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002495 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2497 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002498 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499
Rik van Riel5beb4932010-03-05 13:42:07 -08002500 /* Success. */
2501 if (!err)
2502 return 0;
2503
2504 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002505 if (new->vm_ops && new->vm_ops->close)
2506 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002507 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002508 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002509 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002510 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002511 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002512 out_free_vma:
2513 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002514 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515}
2516
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002517/*
2518 * Split a vma into two pieces at address 'addr', a new vma is allocated
2519 * either for the first part or the tail.
2520 */
2521int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2522 unsigned long addr, int new_below)
2523{
2524 if (mm->map_count >= sysctl_max_map_count)
2525 return -ENOMEM;
2526
2527 return __split_vma(mm, vma, addr, new_below);
2528}
2529
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530/* Munmap is split into 2 main parts -- this part which finds
2531 * what needs doing, and the areas themselves, which do the
2532 * work. This now handles partial unmappings.
2533 * Jeremy Fitzhardinge <jeremy@goop.org>
2534 */
2535int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2536{
2537 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002538 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
2540 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2541 return -EINVAL;
2542
vishnu.pscc71aba2014-10-09 15:26:29 -07002543 len = PAGE_ALIGN(len);
2544 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 return -EINVAL;
2546
2547 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002548 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002549 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002551 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002552 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
2554 /* if it doesn't overlap, we have nothing.. */
2555 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002556 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 return 0;
2558
2559 /*
2560 * If we need to split any vma, do it now to save pain later.
2561 *
2562 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2563 * unmapped vm_area_struct will remain in use: so lower split_vma
2564 * places tmp vma above, and higher split_vma places tmp vma below.
2565 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002566 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002567 int error;
2568
2569 /*
2570 * Make sure that map_count on return from munmap() will
2571 * not exceed its limit; but let map_count go just above
2572 * its limit temporarily, to help free resources as expected.
2573 */
2574 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2575 return -ENOMEM;
2576
2577 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 if (error)
2579 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002580 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 }
2582
2583 /* Does it split the last one? */
2584 last = find_vma(mm, end);
2585 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002586 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 if (error)
2588 return error;
2589 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002590 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
2592 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002593 * unlock any mlock()ed ranges before detaching vmas
2594 */
2595 if (mm->locked_vm) {
2596 struct vm_area_struct *tmp = vma;
2597 while (tmp && tmp->vm_start < end) {
2598 if (tmp->vm_flags & VM_LOCKED) {
2599 mm->locked_vm -= vma_pages(tmp);
2600 munlock_vma_pages_all(tmp);
2601 }
2602 tmp = tmp->vm_next;
2603 }
2604 }
2605
2606 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 * Remove the vma's, and unmap the actual pages
2608 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002609 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2610 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
Dave Hansen1de4fa12014-11-14 07:18:31 -08002612 arch_unmap(mm, vma, start, end);
2613
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002615 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
2617 return 0;
2618}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
Al Virobfce2812012-04-20 21:57:04 -04002620int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002621{
2622 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002623 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002624
2625 down_write(&mm->mmap_sem);
2626 ret = do_munmap(mm, start, len);
2627 up_write(&mm->mmap_sem);
2628 return ret;
2629}
2630EXPORT_SYMBOL(vm_munmap);
2631
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002632SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002635 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636}
2637
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002638
2639/*
2640 * Emulation of deprecated remap_file_pages() syscall.
2641 */
2642SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2643 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2644{
2645
2646 struct mm_struct *mm = current->mm;
2647 struct vm_area_struct *vma;
2648 unsigned long populate = 0;
2649 unsigned long ret = -EINVAL;
2650 struct file *file;
2651
2652 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
2653 "See Documentation/vm/remap_file_pages.txt.\n",
2654 current->comm, current->pid);
2655
2656 if (prot)
2657 return ret;
2658 start = start & PAGE_MASK;
2659 size = size & PAGE_MASK;
2660
2661 if (start + size <= start)
2662 return ret;
2663
2664 /* Does pgoff wrap? */
2665 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2666 return ret;
2667
2668 down_write(&mm->mmap_sem);
2669 vma = find_vma(mm, start);
2670
2671 if (!vma || !(vma->vm_flags & VM_SHARED))
2672 goto out;
2673
2674 if (start < vma->vm_start || start + size > vma->vm_end)
2675 goto out;
2676
2677 if (pgoff == linear_page_index(vma, start)) {
2678 ret = 0;
2679 goto out;
2680 }
2681
2682 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2683 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2684 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2685
2686 flags &= MAP_NONBLOCK;
2687 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2688 if (vma->vm_flags & VM_LOCKED) {
2689 flags |= MAP_LOCKED;
2690 /* drop PG_Mlocked flag for over-mapped range */
2691 munlock_vma_pages_range(vma, start, start + size);
2692 }
2693
2694 file = get_file(vma->vm_file);
2695 ret = do_mmap_pgoff(vma->vm_file, start, size,
2696 prot, flags, pgoff, &populate);
2697 fput(file);
2698out:
2699 up_write(&mm->mmap_sem);
2700 if (populate)
2701 mm_populate(ret, populate);
2702 if (!IS_ERR_VALUE(ret))
2703 ret = 0;
2704 return ret;
2705}
2706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707static inline void verify_mm_writelocked(struct mm_struct *mm)
2708{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002709#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2711 WARN_ON(1);
2712 up_read(&mm->mmap_sem);
2713 }
2714#endif
2715}
2716
2717/*
2718 * this is really a simplified "do_mmap". it only handles
2719 * anonymous maps. eventually we may be able to do some
2720 * brk-specific accounting here.
2721 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002722static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723{
vishnu.pscc71aba2014-10-09 15:26:29 -07002724 struct mm_struct *mm = current->mm;
2725 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002727 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002729 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730
2731 len = PAGE_ALIGN(len);
2732 if (!len)
2733 return addr;
2734
Kirill Korotaev3a459752006-09-07 14:17:04 +04002735 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2736
Al Viro2c6a1012009-12-03 19:40:46 -05002737 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2738 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002739 return error;
2740
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002741 error = mlock_future_check(mm, mm->def_flags, len);
2742 if (error)
2743 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744
2745 /*
2746 * mm->mmap_sem is required to protect against another thread
2747 * changing the mappings in case we sleep.
2748 */
2749 verify_mm_writelocked(mm);
2750
2751 /*
2752 * Clear old maps. this also does some error checking for us
2753 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002754 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2755 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 if (do_munmap(mm, addr, len))
2757 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 }
2759
2760 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002761 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 return -ENOMEM;
2763
2764 if (mm->map_count > sysctl_max_map_count)
2765 return -ENOMEM;
2766
Al Viro191c5422012-02-13 03:58:52 +00002767 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 return -ENOMEM;
2769
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002771 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002772 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002773 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 goto out;
2775
2776 /*
2777 * create a vma struct for an anonymous mapping
2778 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002779 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 if (!vma) {
2781 vm_unacct_memory(len >> PAGE_SHIFT);
2782 return -ENOMEM;
2783 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784
Rik van Riel5beb4932010-03-05 13:42:07 -08002785 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 vma->vm_mm = mm;
2787 vma->vm_start = addr;
2788 vma->vm_end = addr + len;
2789 vma->vm_pgoff = pgoff;
2790 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002791 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 vma_link(mm, vma, prev, rb_link, rb_parent);
2793out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002794 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002796 if (flags & VM_LOCKED)
2797 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002798 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 return addr;
2800}
2801
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002802unsigned long vm_brk(unsigned long addr, unsigned long len)
2803{
2804 struct mm_struct *mm = current->mm;
2805 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002806 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002807
2808 down_write(&mm->mmap_sem);
2809 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002810 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002811 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002812 if (populate)
2813 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002814 return ret;
2815}
2816EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817
2818/* Release all mmaps. */
2819void exit_mmap(struct mm_struct *mm)
2820{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002821 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002822 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 unsigned long nr_accounted = 0;
2824
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002825 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002826 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002827
Rik van Rielba470de2008-10-18 20:26:50 -07002828 if (mm->locked_vm) {
2829 vma = mm->mmap;
2830 while (vma) {
2831 if (vma->vm_flags & VM_LOCKED)
2832 munlock_vma_pages_all(vma);
2833 vma = vma->vm_next;
2834 }
2835 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002836
2837 arch_exit_mmap(mm);
2838
Rik van Rielba470de2008-10-18 20:26:50 -07002839 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002840 if (!vma) /* Can happen if dup_mmap() received an OOM */
2841 return;
2842
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002845 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002846 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002847 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002848 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002849
Hugh Dickins6ee86302013-04-29 15:07:44 -07002850 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002851 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002854 * Walk the list again, actually closing and freeing it,
2855 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002857 while (vma) {
2858 if (vma->vm_flags & VM_ACCOUNT)
2859 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002860 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002861 }
2862 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863}
2864
2865/* Insert vm structure into process list sorted by address
2866 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002867 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002869int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870{
Hugh Dickins6597d782012-10-08 16:29:07 -07002871 struct vm_area_struct *prev;
2872 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873
Chen Gangc9d13f52015-09-08 15:04:08 -07002874 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2875 &prev, &rb_link, &rb_parent))
2876 return -ENOMEM;
2877 if ((vma->vm_flags & VM_ACCOUNT) &&
2878 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2879 return -ENOMEM;
2880
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 /*
2882 * The vm_pgoff of a purely anonymous vma should be irrelevant
2883 * until its first write fault, when page's anon_vma and index
2884 * are set. But now set the vm_pgoff it will almost certainly
2885 * end up with (unless mremap moves it elsewhere before that
2886 * first wfault), so /proc/pid/maps tells a consistent story.
2887 *
2888 * By setting it to reflect the virtual start address of the
2889 * vma, merges and splits can happen in a seamless way, just
2890 * using the existing file pgoff checks and manipulations.
2891 * Similarly in do_mmap_pgoff and in do_brk.
2892 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002893 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 BUG_ON(vma->anon_vma);
2895 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2896 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302897
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 vma_link(mm, vma, prev, rb_link, rb_parent);
2899 return 0;
2900}
2901
2902/*
2903 * Copy the vma structure to a new location in the same mm,
2904 * prior to moving page table entries, to effect an mremap move.
2905 */
2906struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002907 unsigned long addr, unsigned long len, pgoff_t pgoff,
2908 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909{
2910 struct vm_area_struct *vma = *vmap;
2911 unsigned long vma_start = vma->vm_start;
2912 struct mm_struct *mm = vma->vm_mm;
2913 struct vm_area_struct *new_vma, *prev;
2914 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002915 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916
2917 /*
2918 * If anonymous vma has not yet been faulted, update new pgoff
2919 * to match new location, to increase its chance of merging.
2920 */
Oleg Nesterovce757992015-09-08 14:58:34 -07002921 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002923 faulted_in_anon_vma = false;
2924 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925
Hugh Dickins6597d782012-10-08 16:29:07 -07002926 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2927 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002929 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2930 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 if (new_vma) {
2932 /*
2933 * Source vma may have been merged into new_vma
2934 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002935 if (unlikely(vma_start >= new_vma->vm_start &&
2936 vma_start < new_vma->vm_end)) {
2937 /*
2938 * The only way we can get a vma_merge with
2939 * self during an mremap is if the vma hasn't
2940 * been faulted in yet and we were allowed to
2941 * reset the dst vma->vm_pgoff to the
2942 * destination address of the mremap to allow
2943 * the merge to happen. mremap must change the
2944 * vm_pgoff linearity between src and dst vmas
2945 * (in turn preventing a vma_merge) to be
2946 * safe. It is only safe to keep the vm_pgoff
2947 * linear if there are no pages mapped yet.
2948 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002949 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002950 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002951 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002952 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002954 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07002955 if (!new_vma)
2956 goto out;
2957 *new_vma = *vma;
2958 new_vma->vm_start = addr;
2959 new_vma->vm_end = addr + len;
2960 new_vma->vm_pgoff = pgoff;
2961 if (vma_dup_policy(vma, new_vma))
2962 goto out_free_vma;
2963 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2964 if (anon_vma_clone(new_vma, vma))
2965 goto out_free_mempol;
2966 if (new_vma->vm_file)
2967 get_file(new_vma->vm_file);
2968 if (new_vma->vm_ops && new_vma->vm_ops->open)
2969 new_vma->vm_ops->open(new_vma);
2970 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2971 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 }
2973 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002974
Chen Gange3975892015-09-08 15:03:38 -07002975out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002976 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07002977out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08002978 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07002979out:
Rik van Riel5beb4932010-03-05 13:42:07 -08002980 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002982
2983/*
2984 * Return true if the calling process may expand its vm space by the passed
2985 * number of pages
2986 */
2987int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2988{
2989 unsigned long cur = mm->total_vm; /* pages */
2990 unsigned long lim;
2991
Jiri Slaby59e99e52010-03-05 13:41:44 -08002992 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002993
2994 if (cur + npages > lim)
2995 return 0;
2996 return 1;
2997}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002998
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002999static int special_mapping_fault(struct vm_area_struct *vma,
3000 struct vm_fault *vmf);
3001
3002/*
3003 * Having a close hook prevents vma merging regardless of flags.
3004 */
3005static void special_mapping_close(struct vm_area_struct *vma)
3006{
3007}
3008
3009static const char *special_mapping_name(struct vm_area_struct *vma)
3010{
3011 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3012}
3013
3014static const struct vm_operations_struct special_mapping_vmops = {
3015 .close = special_mapping_close,
3016 .fault = special_mapping_fault,
3017 .name = special_mapping_name,
3018};
3019
3020static const struct vm_operations_struct legacy_special_mapping_vmops = {
3021 .close = special_mapping_close,
3022 .fault = special_mapping_fault,
3023};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003024
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003025static int special_mapping_fault(struct vm_area_struct *vma,
3026 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003027{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003028 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003029 struct page **pages;
3030
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003031 if (vma->vm_ops == &legacy_special_mapping_vmops)
3032 pages = vma->vm_private_data;
3033 else
3034 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
3035 pages;
3036
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003037 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003038 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003039
3040 if (*pages) {
3041 struct page *page = *pages;
3042 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003043 vmf->page = page;
3044 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003045 }
3046
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003047 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003048}
3049
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003050static struct vm_area_struct *__install_special_mapping(
3051 struct mm_struct *mm,
3052 unsigned long addr, unsigned long len,
3053 unsigned long vm_flags, const struct vm_operations_struct *ops,
3054 void *priv)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003055{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003056 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003057 struct vm_area_struct *vma;
3058
3059 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3060 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003061 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003062
Rik van Riel5beb4932010-03-05 13:42:07 -08003063 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003064 vma->vm_mm = mm;
3065 vma->vm_start = addr;
3066 vma->vm_end = addr + len;
3067
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003068 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003069 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003070
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003071 vma->vm_ops = ops;
3072 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003073
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003074 ret = insert_vm_struct(mm, vma);
3075 if (ret)
3076 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003077
3078 mm->total_vm += len >> PAGE_SHIFT;
3079
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003080 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003081
Stefani Seibold3935ed62014-03-17 23:22:02 +01003082 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003083
3084out:
3085 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003086 return ERR_PTR(ret);
3087}
3088
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003089/*
3090 * Called with mm->mmap_sem held for writing.
3091 * Insert a new vma covering the given region, with the given flags.
3092 * Its pages are supplied by the given array of struct page *.
3093 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3094 * The region past the last page supplied will always produce SIGBUS.
3095 * The array pointer and the pages it points to are assumed to stay alive
3096 * for as long as this mapping might exist.
3097 */
3098struct vm_area_struct *_install_special_mapping(
3099 struct mm_struct *mm,
3100 unsigned long addr, unsigned long len,
3101 unsigned long vm_flags, const struct vm_special_mapping *spec)
3102{
3103 return __install_special_mapping(mm, addr, len, vm_flags,
3104 &special_mapping_vmops, (void *)spec);
3105}
3106
Stefani Seibold3935ed62014-03-17 23:22:02 +01003107int install_special_mapping(struct mm_struct *mm,
3108 unsigned long addr, unsigned long len,
3109 unsigned long vm_flags, struct page **pages)
3110{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003111 struct vm_area_struct *vma = __install_special_mapping(
3112 mm, addr, len, vm_flags, &legacy_special_mapping_vmops,
3113 (void *)pages);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003114
Duan Jiong14bd5b42014-06-04 16:07:05 -07003115 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003116}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003117
3118static DEFINE_MUTEX(mm_all_locks_mutex);
3119
Peter Zijlstra454ed842008-08-11 09:30:25 +02003120static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003121{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003122 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003123 /*
3124 * The LSB of head.next can't change from under us
3125 * because we hold the mm_all_locks_mutex.
3126 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003127 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003128 /*
3129 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003130 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003131 * the same anon_vma we won't take it again.
3132 *
3133 * No need of atomic instructions here, head.next
3134 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003135 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003136 */
3137 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003138 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003139 BUG();
3140 }
3141}
3142
Peter Zijlstra454ed842008-08-11 09:30:25 +02003143static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003144{
3145 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3146 /*
3147 * AS_MM_ALL_LOCKS can't change from under us because
3148 * we hold the mm_all_locks_mutex.
3149 *
3150 * Operations on ->flags have to be atomic because
3151 * even if AS_MM_ALL_LOCKS is stable thanks to the
3152 * mm_all_locks_mutex, there may be other cpus
3153 * changing other bitflags in parallel to us.
3154 */
3155 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3156 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003157 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003158 }
3159}
3160
3161/*
3162 * This operation locks against the VM for all pte/vma/mm related
3163 * operations that could ever happen on a certain mm. This includes
3164 * vmtruncate, try_to_unmap, and all page faults.
3165 *
3166 * The caller must take the mmap_sem in write mode before calling
3167 * mm_take_all_locks(). The caller isn't allowed to release the
3168 * mmap_sem until mm_drop_all_locks() returns.
3169 *
3170 * mmap_sem in write mode is required in order to block all operations
3171 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003172 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003173 * anon_vmas to be associated with existing vmas.
3174 *
3175 * A single task can't take more than one mm_take_all_locks() in a row
3176 * or it would deadlock.
3177 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003178 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003179 * mapping->flags avoid to take the same lock twice, if more than one
3180 * vma in this mm is backed by the same anon_vma or address_space.
3181 *
3182 * We can take all the locks in random order because the VM code
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003183 * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003184 * takes more than one of them in a row. Secondly we're protected
3185 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3186 *
3187 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3188 * that may have to take thousand of locks.
3189 *
3190 * mm_take_all_locks() can fail if it's interrupted by signals.
3191 */
3192int mm_take_all_locks(struct mm_struct *mm)
3193{
3194 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003195 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003196
3197 BUG_ON(down_read_trylock(&mm->mmap_sem));
3198
3199 mutex_lock(&mm_all_locks_mutex);
3200
3201 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3202 if (signal_pending(current))
3203 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003204 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003205 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003206 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003207
3208 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3209 if (signal_pending(current))
3210 goto out_unlock;
3211 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003212 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3213 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003214 }
3215
Kautuk Consul584cff52011-10-31 17:08:59 -07003216 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003217
3218out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003219 mm_drop_all_locks(mm);
3220 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003221}
3222
3223static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3224{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003225 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003226 /*
3227 * The LSB of head.next can't change to 0 from under
3228 * us because we hold the mm_all_locks_mutex.
3229 *
3230 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003231 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003232 * never see our bitflag.
3233 *
3234 * No need of atomic instructions here, head.next
3235 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003236 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003237 */
3238 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003239 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003240 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003241 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003242 }
3243}
3244
3245static void vm_unlock_mapping(struct address_space *mapping)
3246{
3247 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3248 /*
3249 * AS_MM_ALL_LOCKS can't change to 0 from under us
3250 * because we hold the mm_all_locks_mutex.
3251 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003252 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003253 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3254 &mapping->flags))
3255 BUG();
3256 }
3257}
3258
3259/*
3260 * The mmap_sem cannot be released by the caller until
3261 * mm_drop_all_locks() returns.
3262 */
3263void mm_drop_all_locks(struct mm_struct *mm)
3264{
3265 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003266 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003267
3268 BUG_ON(down_read_trylock(&mm->mmap_sem));
3269 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3270
3271 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3272 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003273 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3274 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003275 if (vma->vm_file && vma->vm_file->f_mapping)
3276 vm_unlock_mapping(vma->vm_file->f_mapping);
3277 }
3278
3279 mutex_unlock(&mm_all_locks_mutex);
3280}
David Howells8feae132009-01-08 12:04:47 +00003281
3282/*
3283 * initialise the VMA slab
3284 */
3285void __init mmap_init(void)
3286{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003287 int ret;
3288
Tejun Heo908c7f12014-09-08 09:51:29 +09003289 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003290 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003291}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003292
3293/*
3294 * Initialise sysctl_user_reserve_kbytes.
3295 *
3296 * This is intended to prevent a user from starting a single memory hogging
3297 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3298 * mode.
3299 *
3300 * The default value is min(3% of free memory, 128MB)
3301 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3302 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003303static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003304{
3305 unsigned long free_kbytes;
3306
3307 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3308
3309 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3310 return 0;
3311}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003312subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003313
3314/*
3315 * Initialise sysctl_admin_reserve_kbytes.
3316 *
3317 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3318 * to log in and kill a memory hogging process.
3319 *
3320 * Systems with more than 256MB will reserve 8MB, enough to recover
3321 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3322 * only reserve 3% of free pages by default.
3323 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003324static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003325{
3326 unsigned long free_kbytes;
3327
3328 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3329
3330 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3331 return 0;
3332}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003333subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003334
3335/*
3336 * Reinititalise user and admin reserves if memory is added or removed.
3337 *
3338 * The default user reserve max is 128MB, and the default max for the
3339 * admin reserve is 8MB. These are usually, but not always, enough to
3340 * enable recovery from a memory hogging process using login/sshd, a shell,
3341 * and tools like top. It may make sense to increase or even disable the
3342 * reserve depending on the existence of swap or variations in the recovery
3343 * tools. So, the admin may have changed them.
3344 *
3345 * If memory is added and the reserves have been eliminated or increased above
3346 * the default max, then we'll trust the admin.
3347 *
3348 * If memory is removed and there isn't enough free memory, then we
3349 * need to reset the reserves.
3350 *
3351 * Otherwise keep the reserve set by the admin.
3352 */
3353static int reserve_mem_notifier(struct notifier_block *nb,
3354 unsigned long action, void *data)
3355{
3356 unsigned long tmp, free_kbytes;
3357
3358 switch (action) {
3359 case MEM_ONLINE:
3360 /* Default max is 128MB. Leave alone if modified by operator. */
3361 tmp = sysctl_user_reserve_kbytes;
3362 if (0 < tmp && tmp < (1UL << 17))
3363 init_user_reserve();
3364
3365 /* Default max is 8MB. Leave alone if modified by operator. */
3366 tmp = sysctl_admin_reserve_kbytes;
3367 if (0 < tmp && tmp < (1UL << 13))
3368 init_admin_reserve();
3369
3370 break;
3371 case MEM_OFFLINE:
3372 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3373
3374 if (sysctl_user_reserve_kbytes > free_kbytes) {
3375 init_user_reserve();
3376 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3377 sysctl_user_reserve_kbytes);
3378 }
3379
3380 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3381 init_admin_reserve();
3382 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3383 sysctl_admin_reserve_kbytes);
3384 }
3385 break;
3386 default:
3387 break;
3388 }
3389 return NOTIFY_OK;
3390}
3391
3392static struct notifier_block reserve_mem_nb = {
3393 .notifier_call = reserve_mem_notifier,
3394};
3395
3396static int __meminit init_reserve_notifier(void)
3397{
3398 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003399 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003400
3401 return 0;
3402}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003403subsys_initcall(init_reserve_notifier);