blob: 81ba54ff96c79cf82293b3740647aeefb8f08b58 [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
Cyril Hrubise8420a82013-04-29 15:08:33 -07009#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070011#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/mm.h>
13#include <linux/shm.h>
14#include <linux/mman.h>
15#include <linux/pagemap.h>
16#include <linux/swap.h>
17#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/init.h>
20#include <linux/file.h>
21#include <linux/fs.h>
22#include <linux/personality.h>
23#include <linux/security.h>
24#include <linux/hugetlb.h>
25#include <linux/profile.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/mempolicy.h>
29#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070030#include <linux/mmu_notifier.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020031#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040032#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080033#include <linux/khugepaged.h>
Srikar Dronamraju2b144492012-02-09 14:56:42 +053034#include <linux/uprobes.h>
Michel Lespinassed3737182012-12-11 16:01:38 -080035#include <linux/rbtree_augmented.h>
Clark Williamscf4aebc22013-02-07 09:46:59 -060036#include <linux/sched/sysctl.h>
Andrew Shewmaker16408792013-04-29 15:08:12 -070037#include <linux/notifier.h>
38#include <linux/memory.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#include <asm/uaccess.h>
41#include <asm/cacheflush.h>
42#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020043#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Jan Beulich42b77722008-07-23 21:27:10 -070045#include "internal.h"
46
Kirill Korotaev3a459752006-09-07 14:17:04 +040047#ifndef arch_mmap_check
48#define arch_mmap_check(addr, len, flags) (0)
49#endif
50
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080051#ifndef arch_rebalance_pgtables
52#define arch_rebalance_pgtables(addr, len) (addr)
53#endif
54
Hugh Dickinse0da3822005-04-19 13:29:15 -070055static void unmap_region(struct mm_struct *mm,
56 struct vm_area_struct *vma, struct vm_area_struct *prev,
57 unsigned long start, unsigned long end);
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
62 *
63 * map_type prot
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
68 *
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 *
73 */
74pgprot_t protection_map[16] = {
75 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
76 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
77};
78
Hugh Dickins804af2c2006-07-26 21:39:49 +010079pgprot_t vm_get_page_prot(unsigned long vm_flags)
80{
Dave Kleikampb845f312008-07-08 00:28:51 +100081 return __pgprot(pgprot_val(protection_map[vm_flags &
82 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
83 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010084}
85EXPORT_SYMBOL(vm_get_page_prot);
86
Shaohua Li34679d72011-05-24 17:11:18 -070087int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
88int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -080089unsigned long sysctl_overcommit_kbytes __read_mostly;
Christoph Lameterc3d8c142005-09-06 15:16:33 -070090int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070091unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070092unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -070093/*
94 * Make sure vm_committed_as in one cacheline and not cacheline shared with
95 * other variables. It can be updated by several CPUs frequently.
96 */
97struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -0800100 * The global memory commitment made in the system can be a metric
101 * that can be used to drive ballooning decisions when Linux is hosted
102 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
103 * balancing memory across competing virtual machines that are hosted.
104 * Several metrics drive this policy engine including the guest reported
105 * memory commitment.
106 */
107unsigned long vm_memory_committed(void)
108{
109 return percpu_counter_read_positive(&vm_committed_as);
110}
111EXPORT_SYMBOL_GPL(vm_memory_committed);
112
113/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * Check that a process has enough memory to allocate a new virtual
115 * mapping. 0 means there is enough memory for the allocation to
116 * succeed and -ENOMEM implies there is not.
117 *
118 * We currently support three overcommit policies, which are set via the
119 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
120 *
121 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
122 * Additional code 2002 Jul 20 by Robert Love.
123 *
124 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
125 *
126 * Note this is a helper function intended to be used by LSMs which
127 * wish to use this logic.
128 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700129int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130{
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700131 unsigned long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133 vm_acct_memory(pages);
134
135 /*
136 * Sometimes we want to use more memory than we have
137 */
138 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
139 return 0;
140
141 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700142 free = global_page_state(NR_FREE_PAGES);
143 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Dmitry Finkc15bef32011-07-25 17:12:19 -0700145 /*
146 * shmem pages shouldn't be counted as free in this
147 * case, they can't be purged, only swapped out, and
148 * that won't affect the overall amount of available
149 * memory in the system.
150 */
151 free -= global_page_state(NR_SHMEM);
152
Shaohua Liec8acf22013-02-22 16:34:38 -0800153 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
155 /*
156 * Any slabs which are created with the
157 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
158 * which are reclaimable, under pressure. The dentry
159 * cache and most inode caches should fall into this
160 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700161 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700164 * Leave reserved pages. The pages are not for anonymous pages.
165 */
166 if (free <= totalreserve_pages)
167 goto error;
168 else
169 free -= totalreserve_pages;
170
171 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700172 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 */
174 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700175 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
177 if (free > pages)
178 return 0;
179
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700180 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 }
182
Jerome Marchand00619bc2013-11-12 15:08:31 -0800183 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700185 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 */
187 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700188 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700190 /*
191 * Don't let a single process grow so big a user can't recover
192 */
193 if (mm) {
194 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
195 allowed -= min(mm->total_vm / 32, reserve);
196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700198 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700200error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 vm_unacct_memory(pages);
202
203 return -ENOMEM;
204}
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700207 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 */
209static void __remove_shared_vm_struct(struct vm_area_struct *vma,
210 struct file *file, struct address_space *mapping)
211{
212 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500213 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 if (vma->vm_flags & VM_SHARED)
215 mapping->i_mmap_writable--;
216
217 flush_dcache_mmap_lock(mapping);
218 if (unlikely(vma->vm_flags & VM_NONLINEAR))
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700219 list_del_init(&vma->shared.nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700221 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 flush_dcache_mmap_unlock(mapping);
223}
224
225/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700226 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700227 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700229void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
231 struct file *file = vma->vm_file;
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (file) {
234 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700235 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700237 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700239}
240
241/*
242 * Close a vm structure and free it, returning the next.
243 */
244static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
245{
246 struct vm_area_struct *next = vma->vm_next;
247
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700248 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 if (vma->vm_ops && vma->vm_ops->close)
250 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700251 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700252 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700253 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700255 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700258static unsigned long do_brk(unsigned long addr, unsigned long len);
259
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100260SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
262 unsigned long rlim, retval;
263 unsigned long newbrk, oldbrk;
264 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700265 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800266 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268 down_write(&mm->mmap_sem);
269
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700270#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800271 /*
272 * CONFIG_COMPAT_BRK can still be overridden by setting
273 * randomize_va_space to 2, which will still cause mm->start_brk
274 * to be arbitrarily shifted
275 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700276 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800277 min_brk = mm->start_brk;
278 else
279 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700280#else
281 min_brk = mm->start_brk;
282#endif
283 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700285
286 /*
287 * Check against rlimit here. If this check is done later after the test
288 * of oldbrk with newbrk then it can escape the test and let the data
289 * segment grow beyond its set limit the in case where the limit is
290 * not page aligned -Ram Gupta
291 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800292 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100293 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
294 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700295 goto out;
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 newbrk = PAGE_ALIGN(brk);
298 oldbrk = PAGE_ALIGN(mm->brk);
299 if (oldbrk == newbrk)
300 goto set_brk;
301
302 /* Always allow shrinking brk. */
303 if (brk <= mm->brk) {
304 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
305 goto set_brk;
306 goto out;
307 }
308
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 /* Check against existing mmap mappings. */
310 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
311 goto out;
312
313 /* Ok, looks good - let it rip. */
314 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
315 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800316
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317set_brk:
318 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800319 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
320 up_write(&mm->mmap_sem);
321 if (populate)
322 mm_populate(oldbrk, newbrk - oldbrk);
323 return brk;
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325out:
326 retval = mm->brk;
327 up_write(&mm->mmap_sem);
328 return retval;
329}
330
Michel Lespinassed3737182012-12-11 16:01:38 -0800331static long vma_compute_subtree_gap(struct vm_area_struct *vma)
332{
333 unsigned long max, subtree_gap;
334 max = vma->vm_start;
335 if (vma->vm_prev)
336 max -= vma->vm_prev->vm_end;
337 if (vma->vm_rb.rb_left) {
338 subtree_gap = rb_entry(vma->vm_rb.rb_left,
339 struct vm_area_struct, vm_rb)->rb_subtree_gap;
340 if (subtree_gap > max)
341 max = subtree_gap;
342 }
343 if (vma->vm_rb.rb_right) {
344 subtree_gap = rb_entry(vma->vm_rb.rb_right,
345 struct vm_area_struct, vm_rb)->rb_subtree_gap;
346 if (subtree_gap > max)
347 max = subtree_gap;
348 }
349 return max;
350}
351
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700352#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353static int browse_rb(struct rb_root *root)
354{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800355 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 struct rb_node *nd, *pn = NULL;
357 unsigned long prev = 0, pend = 0;
358
359 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
360 struct vm_area_struct *vma;
361 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800362 if (vma->vm_start < prev) {
363 printk("vm_start %lx prev %lx\n", vma->vm_start, prev);
364 bug = 1;
365 }
366 if (vma->vm_start < pend) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800368 bug = 1;
369 }
370 if (vma->vm_start > vma->vm_end) {
371 printk("vm_end %lx < vm_start %lx\n",
372 vma->vm_end, vma->vm_start);
373 bug = 1;
374 }
375 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
376 printk("free gap %lx, correct %lx\n",
377 vma->rb_subtree_gap,
378 vma_compute_subtree_gap(vma));
379 bug = 1;
380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 i++;
382 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800383 prev = vma->vm_start;
384 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
386 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800387 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800389 if (i != j) {
390 printk("backwards %d, forwards %d\n", j, i);
391 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800393 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
Michel Lespinassed3737182012-12-11 16:01:38 -0800396static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
397{
398 struct rb_node *nd;
399
400 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
401 struct vm_area_struct *vma;
402 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
403 BUG_ON(vma != ignore &&
404 vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
408void validate_mm(struct mm_struct *mm)
409{
410 int bug = 0;
411 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800412 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700413 struct vm_area_struct *vma = mm->mmap;
414 while (vma) {
415 struct anon_vma_chain *avc;
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800416 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700417 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
418 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800419 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800420 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700421 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 i++;
423 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800424 if (i != mm->map_count) {
425 printk("map_count %d vm_next %d\n", mm->map_count, i);
426 bug = 1;
427 }
428 if (highest_address != mm->highest_vm_end) {
429 printk("mm->highest_vm_end %lx, found %lx\n",
430 mm->highest_vm_end, highest_address);
431 bug = 1;
432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800434 if (i != mm->map_count) {
435 printk("map_count %d rb %d\n", mm->map_count, i);
436 bug = 1;
437 }
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200438 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
440#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800441#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442#define validate_mm(mm) do { } while (0)
443#endif
444
Michel Lespinassed3737182012-12-11 16:01:38 -0800445RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
446 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
447
448/*
449 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
450 * vma->vm_prev->vm_end values changed, without modifying the vma's position
451 * in the rbtree.
452 */
453static void vma_gap_update(struct vm_area_struct *vma)
454{
455 /*
456 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
457 * function that does exacltly what we want.
458 */
459 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
460}
461
462static inline void vma_rb_insert(struct vm_area_struct *vma,
463 struct rb_root *root)
464{
465 /* All rb_subtree_gap values must be consistent prior to insertion */
466 validate_mm_rb(root, NULL);
467
468 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
469}
470
471static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
472{
473 /*
474 * All rb_subtree_gap values must be consistent prior to erase,
475 * with the possible exception of the vma being erased.
476 */
477 validate_mm_rb(root, vma);
478
479 /*
480 * Note rb_erase_augmented is a fairly large inline function,
481 * so make sure we instantiate it only once with our desired
482 * augmented rbtree callbacks.
483 */
484 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
485}
486
Michel Lespinassebf181b92012-10-08 16:31:39 -0700487/*
488 * vma has some anon_vma assigned, and is already inserted on that
489 * anon_vma's interval trees.
490 *
491 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
492 * vma must be removed from the anon_vma's interval trees using
493 * anon_vma_interval_tree_pre_update_vma().
494 *
495 * After the update, the vma will be reinserted using
496 * anon_vma_interval_tree_post_update_vma().
497 *
498 * The entire update must be protected by exclusive mmap_sem and by
499 * the root anon_vma's mutex.
500 */
501static inline void
502anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
503{
504 struct anon_vma_chain *avc;
505
506 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
507 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
508}
509
510static inline void
511anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
512{
513 struct anon_vma_chain *avc;
514
515 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
516 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
517}
518
Hugh Dickins6597d782012-10-08 16:29:07 -0700519static int find_vma_links(struct mm_struct *mm, unsigned long addr,
520 unsigned long end, struct vm_area_struct **pprev,
521 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522{
Hugh Dickins6597d782012-10-08 16:29:07 -0700523 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525 __rb_link = &mm->mm_rb.rb_node;
526 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 while (*__rb_link) {
529 struct vm_area_struct *vma_tmp;
530
531 __rb_parent = *__rb_link;
532 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
533
534 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700535 /* Fail if an existing vma overlaps the area */
536 if (vma_tmp->vm_start < end)
537 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 __rb_link = &__rb_parent->rb_left;
539 } else {
540 rb_prev = __rb_parent;
541 __rb_link = &__rb_parent->rb_right;
542 }
543 }
544
545 *pprev = NULL;
546 if (rb_prev)
547 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
548 *rb_link = __rb_link;
549 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700550 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551}
552
Cyril Hrubise8420a82013-04-29 15:08:33 -0700553static unsigned long count_vma_pages_range(struct mm_struct *mm,
554 unsigned long addr, unsigned long end)
555{
556 unsigned long nr_pages = 0;
557 struct vm_area_struct *vma;
558
559 /* Find first overlaping mapping */
560 vma = find_vma_intersection(mm, addr, end);
561 if (!vma)
562 return 0;
563
564 nr_pages = (min(end, vma->vm_end) -
565 max(addr, vma->vm_start)) >> PAGE_SHIFT;
566
567 /* Iterate over the rest of the overlaps */
568 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
569 unsigned long overlap_len;
570
571 if (vma->vm_start > end)
572 break;
573
574 overlap_len = min(end, vma->vm_end) - vma->vm_start;
575 nr_pages += overlap_len >> PAGE_SHIFT;
576 }
577
578 return nr_pages;
579}
580
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
582 struct rb_node **rb_link, struct rb_node *rb_parent)
583{
Michel Lespinassed3737182012-12-11 16:01:38 -0800584 /* Update tracking information for the gap following the new vma. */
585 if (vma->vm_next)
586 vma_gap_update(vma->vm_next);
587 else
588 mm->highest_vm_end = vma->vm_end;
589
590 /*
591 * vma->vm_prev wasn't known when we followed the rbtree to find the
592 * correct insertion point for that vma. As a result, we could not
593 * update the vma vm_rb parents rb_subtree_gap values on the way down.
594 * So, we first insert the vma with a zero rb_subtree_gap value
595 * (to be consistent with what we did on the way down), and then
596 * immediately update the gap to the correct value. Finally we
597 * rebalance the rbtree after all augmented values have been set.
598 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800600 vma->rb_subtree_gap = 0;
601 vma_gap_update(vma);
602 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603}
604
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700605static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
ZhenwenXu48aae422009-01-06 14:40:21 -0800607 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609 file = vma->vm_file;
610 if (file) {
611 struct address_space *mapping = file->f_mapping;
612
613 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500614 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (vma->vm_flags & VM_SHARED)
616 mapping->i_mmap_writable++;
617
618 flush_dcache_mmap_lock(mapping);
619 if (unlikely(vma->vm_flags & VM_NONLINEAR))
620 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
621 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700622 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 flush_dcache_mmap_unlock(mapping);
624 }
625}
626
627static void
628__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
629 struct vm_area_struct *prev, struct rb_node **rb_link,
630 struct rb_node *rb_parent)
631{
632 __vma_link_list(mm, vma, prev, rb_parent);
633 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}
635
636static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
637 struct vm_area_struct *prev, struct rb_node **rb_link,
638 struct rb_node *rb_parent)
639{
640 struct address_space *mapping = NULL;
641
642 if (vma->vm_file)
643 mapping = vma->vm_file->f_mapping;
644
Peter Zijlstra97a89412011-05-24 17:12:04 -0700645 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700646 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
648 __vma_link(mm, vma, prev, rb_link, rb_parent);
649 __vma_link_file(vma);
650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700652 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 mm->map_count++;
655 validate_mm(mm);
656}
657
658/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700659 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700660 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800662static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663{
Hugh Dickins6597d782012-10-08 16:29:07 -0700664 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800665 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Hugh Dickins6597d782012-10-08 16:29:07 -0700667 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
668 &prev, &rb_link, &rb_parent))
669 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 __vma_link(mm, vma, prev, rb_link, rb_parent);
671 mm->map_count++;
672}
673
674static inline void
675__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
676 struct vm_area_struct *prev)
677{
Michel Lespinassed3737182012-12-11 16:01:38 -0800678 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700679
Michel Lespinassed3737182012-12-11 16:01:38 -0800680 vma_rb_erase(vma, &mm->mm_rb);
681 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700682 if (next)
683 next->vm_prev = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (mm->mmap_cache == vma)
685 mm->mmap_cache = prev;
686}
687
688/*
689 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
690 * is already present in an i_mmap tree without adjusting the tree.
691 * The following helper function should be used when such adjustments
692 * are necessary. The "insert" vma (if any) is to be inserted
693 * before we drop the necessary locks.
694 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800695int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
697{
698 struct mm_struct *mm = vma->vm_mm;
699 struct vm_area_struct *next = vma->vm_next;
700 struct vm_area_struct *importer = NULL;
701 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700702 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700703 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800705 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 long adjust_next = 0;
707 int remove_next = 0;
708
709 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700710 struct vm_area_struct *exporter = NULL;
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if (end >= next->vm_end) {
713 /*
714 * vma expands, overlapping all the next, and
715 * perhaps the one after too (mprotect case 6).
716 */
717again: remove_next = 1 + (end > next->vm_end);
718 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700719 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 importer = vma;
721 } else if (end > next->vm_start) {
722 /*
723 * vma expands, overlapping part of the next:
724 * mprotect case 5 shifting the boundary up.
725 */
726 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700727 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 importer = vma;
729 } else if (end < vma->vm_end) {
730 /*
731 * vma shrinks, and !insert tells it's not
732 * split_vma inserting another: so it must be
733 * mprotect case 4 shifting the boundary down.
734 */
735 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700736 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 importer = next;
738 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Rik van Riel5beb4932010-03-05 13:42:07 -0800740 /*
741 * Easily overlooked: when mprotect shifts the boundary,
742 * make sure the expanding vma has anon_vma set if the
743 * shrinking vma had, to cover any anon pages imported.
744 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700745 if (exporter && exporter->anon_vma && !importer->anon_vma) {
746 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800747 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700748 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800749 }
750 }
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 if (file) {
753 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530754 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530756 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530757
758 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530759 uprobe_munmap(next, next->vm_start,
760 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530761 }
762
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700763 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700766 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 * are visible to arm/parisc __flush_dcache_page
768 * throughout; but we cannot insert into address
769 * space until vma start or end is updated.
770 */
771 __vma_link_file(insert);
772 }
773 }
774
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800775 vma_adjust_trans_huge(vma, start, end, adjust_next);
776
Michel Lespinassebf181b92012-10-08 16:31:39 -0700777 anon_vma = vma->anon_vma;
778 if (!anon_vma && adjust_next)
779 anon_vma = next->anon_vma;
780 if (anon_vma) {
Michel Lespinasseca42b262012-10-08 16:30:01 -0700781 VM_BUG_ON(adjust_next && next->anon_vma &&
782 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000783 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700784 anon_vma_interval_tree_pre_update_vma(vma);
785 if (adjust_next)
786 anon_vma_interval_tree_pre_update_vma(next);
787 }
Rik van Riel012f18002010-08-09 17:18:40 -0700788
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 if (root) {
790 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700791 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700793 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 }
795
Michel Lespinassed3737182012-12-11 16:01:38 -0800796 if (start != vma->vm_start) {
797 vma->vm_start = start;
798 start_changed = true;
799 }
800 if (end != vma->vm_end) {
801 vma->vm_end = end;
802 end_changed = true;
803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 vma->vm_pgoff = pgoff;
805 if (adjust_next) {
806 next->vm_start += adjust_next << PAGE_SHIFT;
807 next->vm_pgoff += adjust_next;
808 }
809
810 if (root) {
811 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700812 vma_interval_tree_insert(next, root);
813 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 flush_dcache_mmap_unlock(mapping);
815 }
816
817 if (remove_next) {
818 /*
819 * vma_merge has merged next into vma, and needs
820 * us to remove next before dropping the locks.
821 */
822 __vma_unlink(mm, next, vma);
823 if (file)
824 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 } else if (insert) {
826 /*
827 * split_vma has split insert from vma, and needs
828 * us to insert it before dropping the locks
829 * (it may either follow vma or precede it).
830 */
831 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800832 } else {
833 if (start_changed)
834 vma_gap_update(vma);
835 if (end_changed) {
836 if (!next)
837 mm->highest_vm_end = end;
838 else if (!adjust_next)
839 vma_gap_update(next);
840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 }
842
Michel Lespinassebf181b92012-10-08 16:31:39 -0700843 if (anon_vma) {
844 anon_vma_interval_tree_post_update_vma(vma);
845 if (adjust_next)
846 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800847 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700850 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530852 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100853 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530854
855 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100856 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530857 }
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700860 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530861 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700863 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800864 if (next->anon_vma)
865 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700867 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 kmem_cache_free(vm_area_cachep, next);
869 /*
870 * In mprotect's case 6 (see comments on vma_merge),
871 * we must remove another next too. It would clutter
872 * up the code too much to do both in one go.
873 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800874 next = vma->vm_next;
875 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800877 else if (next)
878 vma_gap_update(next);
879 else
880 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530882 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100883 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800886
887 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888}
889
890/*
891 * If the vma has a ->close operation then the driver probably needs to release
892 * per-vma resources, so we don't attempt to merge those.
893 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894static inline int is_mergeable_vma(struct vm_area_struct *vma,
895 struct file *file, unsigned long vm_flags)
896{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800897 /*
898 * VM_SOFTDIRTY should not prevent from VMA merging, if we
899 * match the flags but dirty bit -- the caller should mark
900 * merged VMA as dirty. If dirty bit won't be excluded from
901 * comparison, we increase pressue on the memory system forcing
902 * the kernel to generate new VMAs when old one could be
903 * extended instead.
904 */
905 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 return 0;
907 if (vma->vm_file != file)
908 return 0;
909 if (vma->vm_ops && vma->vm_ops->close)
910 return 0;
911 return 1;
912}
913
914static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700915 struct anon_vma *anon_vma2,
916 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917{
Shaohua Li965f55d2011-05-24 17:11:20 -0700918 /*
919 * The list_is_singular() test is to avoid merging VMA cloned from
920 * parents. This can improve scalability caused by anon_vma lock.
921 */
922 if ((!anon_vma1 || !anon_vma2) && (!vma ||
923 list_is_singular(&vma->anon_vma_chain)))
924 return 1;
925 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926}
927
928/*
929 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
930 * in front of (at a lower virtual address and file offset than) the vma.
931 *
932 * We cannot merge two vmas if they have differently assigned (non-NULL)
933 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
934 *
935 * We don't check here for the merged mmap wrapping around the end of pagecache
936 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
937 * wrap, nor mmaps which cover the final page at index -1UL.
938 */
939static int
940can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
941 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
942{
943 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700944 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (vma->vm_pgoff == vm_pgoff)
946 return 1;
947 }
948 return 0;
949}
950
951/*
952 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
953 * beyond (at a higher virtual address and file offset than) the vma.
954 *
955 * We cannot merge two vmas if they have differently assigned (non-NULL)
956 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
957 */
958static int
959can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
960 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
961{
962 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700963 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700965 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
967 return 1;
968 }
969 return 0;
970}
971
972/*
973 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
974 * whether that can be merged with its predecessor or its successor.
975 * Or both (it neatly fills a hole).
976 *
977 * In most cases - when called for mmap, brk or mremap - [addr,end) is
978 * certain not to be mapped by the time vma_merge is called; but when
979 * called for mprotect, it is certain to be already mapped (either at
980 * an offset within prev, or at the start of next), and the flags of
981 * this area are about to be changed to vm_flags - and the no-change
982 * case has already been eliminated.
983 *
984 * The following mprotect cases have to be considered, where AAAA is
985 * the area passed down from mprotect_fixup, never extending beyond one
986 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
987 *
988 * AAAA AAAA AAAA AAAA
989 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
990 * cannot merge might become might become might become
991 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
992 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
993 * mremap move: PPPPNNNNNNNN 8
994 * AAAA
995 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
996 * might become case 1 below case 2 below case 3 below
997 *
998 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
999 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1000 */
1001struct vm_area_struct *vma_merge(struct mm_struct *mm,
1002 struct vm_area_struct *prev, unsigned long addr,
1003 unsigned long end, unsigned long vm_flags,
1004 struct anon_vma *anon_vma, struct file *file,
1005 pgoff_t pgoff, struct mempolicy *policy)
1006{
1007 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1008 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001009 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011 /*
1012 * We later require that vma->vm_flags == vm_flags,
1013 * so this tests vma->vm_flags & VM_SPECIAL, too.
1014 */
1015 if (vm_flags & VM_SPECIAL)
1016 return NULL;
1017
1018 if (prev)
1019 next = prev->vm_next;
1020 else
1021 next = mm->mmap;
1022 area = next;
1023 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1024 next = next->vm_next;
1025
1026 /*
1027 * Can it merge with the predecessor?
1028 */
1029 if (prev && prev->vm_end == addr &&
1030 mpol_equal(vma_policy(prev), policy) &&
1031 can_vma_merge_after(prev, vm_flags,
1032 anon_vma, file, pgoff)) {
1033 /*
1034 * OK, it can. Can we now merge in the successor as well?
1035 */
1036 if (next && end == next->vm_start &&
1037 mpol_equal(policy, vma_policy(next)) &&
1038 can_vma_merge_before(next, vm_flags,
1039 anon_vma, file, pgoff+pglen) &&
1040 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001041 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001043 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 next->vm_end, prev->vm_pgoff, NULL);
1045 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001046 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001048 if (err)
1049 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001050 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 return prev;
1052 }
1053
1054 /*
1055 * Can this new request be merged in front of next?
1056 */
1057 if (next && end == next->vm_start &&
1058 mpol_equal(policy, vma_policy(next)) &&
1059 can_vma_merge_before(next, vm_flags,
1060 anon_vma, file, pgoff+pglen)) {
1061 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001062 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 addr, prev->vm_pgoff, NULL);
1064 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001065 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001067 if (err)
1068 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001069 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 return area;
1071 }
1072
1073 return NULL;
1074}
1075
1076/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001077 * Rough compatbility check to quickly see if it's even worth looking
1078 * at sharing an anon_vma.
1079 *
1080 * They need to have the same vm_file, and the flags can only differ
1081 * in things that mprotect may change.
1082 *
1083 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1084 * we can merge the two vma's. For example, we refuse to merge a vma if
1085 * there is a vm_ops->close() function, because that indicates that the
1086 * driver is doing some kind of reference counting. But that doesn't
1087 * really matter for the anon_vma sharing case.
1088 */
1089static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1090{
1091 return a->vm_end == b->vm_start &&
1092 mpol_equal(vma_policy(a), vma_policy(b)) &&
1093 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001094 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001095 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1096}
1097
1098/*
1099 * Do some basic sanity checking to see if we can re-use the anon_vma
1100 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1101 * the same as 'old', the other will be the new one that is trying
1102 * to share the anon_vma.
1103 *
1104 * NOTE! This runs with mm_sem held for reading, so it is possible that
1105 * the anon_vma of 'old' is concurrently in the process of being set up
1106 * by another page fault trying to merge _that_. But that's ok: if it
1107 * is being set up, that automatically means that it will be a singleton
1108 * acceptable for merging, so we can do all of this optimistically. But
1109 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1110 *
1111 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1112 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1113 * is to return an anon_vma that is "complex" due to having gone through
1114 * a fork).
1115 *
1116 * We also make sure that the two vma's are compatible (adjacent,
1117 * and with the same memory policies). That's all stable, even with just
1118 * a read lock on the mm_sem.
1119 */
1120static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1121{
1122 if (anon_vma_compatible(a, b)) {
1123 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1124
1125 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1126 return anon_vma;
1127 }
1128 return NULL;
1129}
1130
1131/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1133 * neighbouring vmas for a suitable anon_vma, before it goes off
1134 * to allocate a new anon_vma. It checks because a repetitive
1135 * sequence of mprotects and faults may otherwise lead to distinct
1136 * anon_vmas being allocated, preventing vma merge in subsequent
1137 * mprotect.
1138 */
1139struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1140{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001141 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
1144 near = vma->vm_next;
1145 if (!near)
1146 goto try_prev;
1147
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001148 anon_vma = reusable_anon_vma(near, vma, near);
1149 if (anon_vma)
1150 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001152 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 if (!near)
1154 goto none;
1155
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001156 anon_vma = reusable_anon_vma(near, near, vma);
1157 if (anon_vma)
1158 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159none:
1160 /*
1161 * There's no absolute need to look only at touching neighbours:
1162 * we could search further afield for "compatible" anon_vmas.
1163 * But it would probably just be a waste of time searching,
1164 * or lead to too many vmas hanging off the same anon_vma.
1165 * We're trying to allow mprotect remerging later on,
1166 * not trying to minimize memory used for anon_vmas.
1167 */
1168 return NULL;
1169}
1170
1171#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001172void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 struct file *file, long pages)
1174{
1175 const unsigned long stack_flags
1176 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1177
Huang Shijie44de9d02012-07-31 16:41:49 -07001178 mm->total_vm += pages;
1179
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 if (file) {
1181 mm->shared_vm += pages;
1182 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1183 mm->exec_vm += pages;
1184 } else if (flags & stack_flags)
1185 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187#endif /* CONFIG_PROC_FS */
1188
1189/*
Al Viro40401532012-02-13 03:58:52 +00001190 * If a hint addr is less than mmap_min_addr change hint to be as
1191 * low as possible but still greater than mmap_min_addr
1192 */
1193static inline unsigned long round_hint_to_min(unsigned long hint)
1194{
1195 hint &= PAGE_MASK;
1196 if (((void *)hint != NULL) &&
1197 (hint < mmap_min_addr))
1198 return PAGE_ALIGN(mmap_min_addr);
1199 return hint;
1200}
1201
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001202static inline int mlock_future_check(struct mm_struct *mm,
1203 unsigned long flags,
1204 unsigned long len)
1205{
1206 unsigned long locked, lock_limit;
1207
1208 /* mlock MCL_FUTURE? */
1209 if (flags & VM_LOCKED) {
1210 locked = len >> PAGE_SHIFT;
1211 locked += mm->locked_vm;
1212 lock_limit = rlimit(RLIMIT_MEMLOCK);
1213 lock_limit >>= PAGE_SHIFT;
1214 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1215 return -EAGAIN;
1216 }
1217 return 0;
1218}
1219
Al Viro40401532012-02-13 03:58:52 +00001220/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001221 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 */
1223
Al Viroe3fc6292012-05-30 20:08:42 -04001224unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001226 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001227 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228{
1229 struct mm_struct * mm = current->mm;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001230 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Michel Lespinasse41badc12013-02-22 16:32:47 -08001232 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001233
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 /*
1235 * Does the application expect PROT_READ to imply PROT_EXEC?
1236 *
1237 * (the exception is when the underlying filesystem is noexec
1238 * mounted, in which case we dont add PROT_EXEC.)
1239 */
1240 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001241 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 prot |= PROT_EXEC;
1243
1244 if (!len)
1245 return -EINVAL;
1246
Eric Paris7cd94142007-11-26 18:47:40 -05001247 if (!(flags & MAP_FIXED))
1248 addr = round_hint_to_min(addr);
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 /* Careful about overflows.. */
1251 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001252 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 return -ENOMEM;
1254
1255 /* offset overflow? */
1256 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1257 return -EOVERFLOW;
1258
1259 /* Too many mappings? */
1260 if (mm->map_count > sysctl_max_map_count)
1261 return -ENOMEM;
1262
1263 /* Obtain the address to map to. we verify (or select) it and ensure
1264 * that it represents a valid section of the address space.
1265 */
1266 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1267 if (addr & ~PAGE_MASK)
1268 return addr;
1269
1270 /* Do simple checking here so the lower-level routines won't have
1271 * to. we assume access permissions have been handled by the open
1272 * of the memory object, so we don't do any here.
1273 */
1274 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1275 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1276
Huang Shijiecdf7b342009-09-21 17:03:36 -07001277 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (!can_do_mlock())
1279 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001280
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001281 if (mlock_future_check(mm, vm_flags, len))
1282 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001285 struct inode *inode = file_inode(file);
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 switch (flags & MAP_TYPE) {
1288 case MAP_SHARED:
1289 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1290 return -EACCES;
1291
1292 /*
1293 * Make sure we don't allow writing to an append-only
1294 * file..
1295 */
1296 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1297 return -EACCES;
1298
1299 /*
1300 * Make sure there are no mandatory locks on the file.
1301 */
1302 if (locks_verify_locked(inode))
1303 return -EAGAIN;
1304
1305 vm_flags |= VM_SHARED | VM_MAYSHARE;
1306 if (!(file->f_mode & FMODE_WRITE))
1307 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1308
1309 /* fall through */
1310 case MAP_PRIVATE:
1311 if (!(file->f_mode & FMODE_READ))
1312 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001313 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001314 if (vm_flags & VM_EXEC)
1315 return -EPERM;
1316 vm_flags &= ~VM_MAYEXEC;
1317 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001318
Al Viro72c2d532013-09-22 16:27:52 -04001319 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001320 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001321 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1322 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 break;
1324
1325 default:
1326 return -EINVAL;
1327 }
1328 } else {
1329 switch (flags & MAP_TYPE) {
1330 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001331 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1332 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001333 /*
1334 * Ignore pgoff.
1335 */
1336 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 vm_flags |= VM_SHARED | VM_MAYSHARE;
1338 break;
1339 case MAP_PRIVATE:
1340 /*
1341 * Set pgoff according to addr for anon_vma.
1342 */
1343 pgoff = addr >> PAGE_SHIFT;
1344 break;
1345 default:
1346 return -EINVAL;
1347 }
1348 }
1349
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001350 /*
1351 * Set 'VM_NORESERVE' if we should not account for the
1352 * memory use of this mapping.
1353 */
1354 if (flags & MAP_NORESERVE) {
1355 /* We honor MAP_NORESERVE if allowed to overcommit */
1356 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1357 vm_flags |= VM_NORESERVE;
1358
1359 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1360 if (file && is_file_hugepages(file))
1361 vm_flags |= VM_NORESERVE;
1362 }
1363
1364 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001365 if (!IS_ERR_VALUE(addr) &&
1366 ((vm_flags & VM_LOCKED) ||
1367 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001368 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001369 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001370}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001371
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001372SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1373 unsigned long, prot, unsigned long, flags,
1374 unsigned long, fd, unsigned long, pgoff)
1375{
1376 struct file *file = NULL;
1377 unsigned long retval = -EBADF;
1378
1379 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001380 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001381 file = fget(fd);
1382 if (!file)
1383 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001384 if (is_file_hugepages(file))
1385 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001386 retval = -EINVAL;
1387 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1388 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001389 } else if (flags & MAP_HUGETLB) {
1390 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001391 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001392
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001393 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001394 if (!hs)
1395 return -EINVAL;
1396
1397 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001398 /*
1399 * VM_NORESERVE is used because the reservations will be
1400 * taken when vm_ops->mmap() is called
1401 * A dummy user value is used because we are not locking
1402 * memory so no accounting is necessary
1403 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001404 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001405 VM_NORESERVE,
1406 &user, HUGETLB_ANONHUGE_INODE,
1407 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001408 if (IS_ERR(file))
1409 return PTR_ERR(file);
1410 }
1411
1412 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1413
Al Viroeb36c582012-05-30 20:17:35 -04001414 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001415out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001416 if (file)
1417 fput(file);
1418out:
1419 return retval;
1420}
1421
Christoph Hellwiga4679372010-03-10 15:21:15 -08001422#ifdef __ARCH_WANT_SYS_OLD_MMAP
1423struct mmap_arg_struct {
1424 unsigned long addr;
1425 unsigned long len;
1426 unsigned long prot;
1427 unsigned long flags;
1428 unsigned long fd;
1429 unsigned long offset;
1430};
1431
1432SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1433{
1434 struct mmap_arg_struct a;
1435
1436 if (copy_from_user(&a, arg, sizeof(a)))
1437 return -EFAULT;
1438 if (a.offset & ~PAGE_MASK)
1439 return -EINVAL;
1440
1441 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1442 a.offset >> PAGE_SHIFT);
1443}
1444#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1445
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001446/*
1447 * Some shared mappigns will want the pages marked read-only
1448 * to track write events. If so, we'll downgrade vm_page_prot
1449 * to the private version (using protection_map[] without the
1450 * VM_SHARED bit).
1451 */
1452int vma_wants_writenotify(struct vm_area_struct *vma)
1453{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001454 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001455
1456 /* If it was private or non-writable, the write bit is already clear */
1457 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1458 return 0;
1459
1460 /* The backer wishes to know when pages are first written to? */
1461 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1462 return 1;
1463
1464 /* The open routine did something to the protections already? */
1465 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001466 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001467 return 0;
1468
1469 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001470 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001471 return 0;
1472
1473 /* Can the mapping track the dirty pages? */
1474 return vma->vm_file && vma->vm_file->f_mapping &&
1475 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1476}
1477
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001478/*
1479 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001480 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001481 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001482static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001483{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001484 /*
1485 * hugetlb has its own accounting separate from the core VM
1486 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1487 */
1488 if (file && is_file_hugepages(file))
1489 return 0;
1490
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001491 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1492}
1493
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001494unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001495 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001496{
1497 struct mm_struct *mm = current->mm;
1498 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001499 int error;
1500 struct rb_node **rb_link, *rb_parent;
1501 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001502
Cyril Hrubise8420a82013-04-29 15:08:33 -07001503 /* Check against address space limit. */
1504 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1505 unsigned long nr_pages;
1506
1507 /*
1508 * MAP_FIXED may remove pages of mappings that intersects with
1509 * requested mapping. Account for the pages it would unmap.
1510 */
1511 if (!(vm_flags & MAP_FIXED))
1512 return -ENOMEM;
1513
1514 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1515
1516 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1517 return -ENOMEM;
1518 }
1519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 /* Clear old maps */
1521 error = -ENOMEM;
1522munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001523 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 if (do_munmap(mm, addr, len))
1525 return -ENOMEM;
1526 goto munmap_back;
1527 }
1528
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001529 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001530 * Private writable mapping: check memory availability
1531 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001532 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001533 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001534 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001535 return -ENOMEM;
1536 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 }
1538
1539 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001540 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001542 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1543 if (vma)
1544 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
1546 /*
1547 * Determine the object being mapped and call the appropriate
1548 * specific mapper. the address has already been validated, but
1549 * not unmapped, but the maps are removed from the list.
1550 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001551 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (!vma) {
1553 error = -ENOMEM;
1554 goto unacct_error;
1555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
1557 vma->vm_mm = mm;
1558 vma->vm_start = addr;
1559 vma->vm_end = addr + len;
1560 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001561 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001563 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
1565 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 if (vm_flags & VM_DENYWRITE) {
1567 error = deny_write_access(file);
1568 if (error)
1569 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
Al Virocb0942b2012-08-27 14:48:26 -04001571 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 error = file->f_op->mmap(file, vma);
1573 if (error)
1574 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001575
1576 /* Can addr have changed??
1577 *
1578 * Answer: Yes, several device drivers can do it in their
1579 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001580 * Bug: If addr is changed, prev, rb_link, rb_parent should
1581 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001582 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001583 WARN_ON_ONCE(addr != vma->vm_start);
1584
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001585 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001586 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 } else if (vm_flags & VM_SHARED) {
1588 error = shmem_zero_setup(vma);
1589 if (error)
1590 goto free_vma;
1591 }
1592
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001593 if (vma_wants_writenotify(vma)) {
1594 pgprot_t pprot = vma->vm_page_prot;
1595
1596 /* Can vma->vm_page_prot have changed??
1597 *
1598 * Answer: Yes, drivers may have changed it in their
1599 * f_op->mmap method.
1600 *
1601 * Ensures that vmas marked as uncached stay that way.
1602 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001603 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001604 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1605 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1606 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001607
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001608 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001609 /* Once vma denies write, undo our temporary denial count */
Oleg Nesterove8686772013-09-11 14:20:20 -07001610 if (vm_flags & VM_DENYWRITE)
1611 allow_write_access(file);
1612 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001613out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001614 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001615
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001616 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001618 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1619 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001620 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001621 else
1622 vma->vm_flags &= ~VM_LOCKED;
1623 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301624
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001625 if (file)
1626 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301627
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001628 /*
1629 * New (or expanded) vma always get soft dirty status.
1630 * Otherwise user-space soft-dirty page tracker won't
1631 * be able to distinguish situation when vma area unmapped,
1632 * then new mapped in-place (which must be aimed as
1633 * a completely new data area).
1634 */
1635 vma->vm_flags |= VM_SOFTDIRTY;
1636
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 return addr;
1638
1639unmap_and_free_vma:
Oleg Nesterove8686772013-09-11 14:20:20 -07001640 if (vm_flags & VM_DENYWRITE)
1641 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 vma->vm_file = NULL;
1643 fput(file);
1644
1645 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001646 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1647 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648free_vma:
1649 kmem_cache_free(vm_area_cachep, vma);
1650unacct_error:
1651 if (charged)
1652 vm_unacct_memory(charged);
1653 return error;
1654}
1655
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001656unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1657{
1658 /*
1659 * We implement the search by looking for an rbtree node that
1660 * immediately follows a suitable gap. That is,
1661 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1662 * - gap_end = vma->vm_start >= info->low_limit + length;
1663 * - gap_end - gap_start >= length
1664 */
1665
1666 struct mm_struct *mm = current->mm;
1667 struct vm_area_struct *vma;
1668 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1669
1670 /* Adjust search length to account for worst case alignment overhead */
1671 length = info->length + info->align_mask;
1672 if (length < info->length)
1673 return -ENOMEM;
1674
1675 /* Adjust search limits by the desired length */
1676 if (info->high_limit < length)
1677 return -ENOMEM;
1678 high_limit = info->high_limit - length;
1679
1680 if (info->low_limit > high_limit)
1681 return -ENOMEM;
1682 low_limit = info->low_limit + length;
1683
1684 /* Check if rbtree root looks promising */
1685 if (RB_EMPTY_ROOT(&mm->mm_rb))
1686 goto check_highest;
1687 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1688 if (vma->rb_subtree_gap < length)
1689 goto check_highest;
1690
1691 while (true) {
1692 /* Visit left subtree if it looks promising */
1693 gap_end = vma->vm_start;
1694 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1695 struct vm_area_struct *left =
1696 rb_entry(vma->vm_rb.rb_left,
1697 struct vm_area_struct, vm_rb);
1698 if (left->rb_subtree_gap >= length) {
1699 vma = left;
1700 continue;
1701 }
1702 }
1703
1704 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1705check_current:
1706 /* Check if current node has a suitable gap */
1707 if (gap_start > high_limit)
1708 return -ENOMEM;
1709 if (gap_end >= low_limit && gap_end - gap_start >= length)
1710 goto found;
1711
1712 /* Visit right subtree if it looks promising */
1713 if (vma->vm_rb.rb_right) {
1714 struct vm_area_struct *right =
1715 rb_entry(vma->vm_rb.rb_right,
1716 struct vm_area_struct, vm_rb);
1717 if (right->rb_subtree_gap >= length) {
1718 vma = right;
1719 continue;
1720 }
1721 }
1722
1723 /* Go back up the rbtree to find next candidate node */
1724 while (true) {
1725 struct rb_node *prev = &vma->vm_rb;
1726 if (!rb_parent(prev))
1727 goto check_highest;
1728 vma = rb_entry(rb_parent(prev),
1729 struct vm_area_struct, vm_rb);
1730 if (prev == vma->vm_rb.rb_left) {
1731 gap_start = vma->vm_prev->vm_end;
1732 gap_end = vma->vm_start;
1733 goto check_current;
1734 }
1735 }
1736 }
1737
1738check_highest:
1739 /* Check highest gap, which does not precede any rbtree node */
1740 gap_start = mm->highest_vm_end;
1741 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1742 if (gap_start > high_limit)
1743 return -ENOMEM;
1744
1745found:
1746 /* We found a suitable gap. Clip it with the original low_limit. */
1747 if (gap_start < info->low_limit)
1748 gap_start = info->low_limit;
1749
1750 /* Adjust gap address to the desired alignment */
1751 gap_start += (info->align_offset - gap_start) & info->align_mask;
1752
1753 VM_BUG_ON(gap_start + info->length > info->high_limit);
1754 VM_BUG_ON(gap_start + info->length > gap_end);
1755 return gap_start;
1756}
1757
1758unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1759{
1760 struct mm_struct *mm = current->mm;
1761 struct vm_area_struct *vma;
1762 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1763
1764 /* Adjust search length to account for worst case alignment overhead */
1765 length = info->length + info->align_mask;
1766 if (length < info->length)
1767 return -ENOMEM;
1768
1769 /*
1770 * Adjust search limits by the desired length.
1771 * See implementation comment at top of unmapped_area().
1772 */
1773 gap_end = info->high_limit;
1774 if (gap_end < length)
1775 return -ENOMEM;
1776 high_limit = gap_end - length;
1777
1778 if (info->low_limit > high_limit)
1779 return -ENOMEM;
1780 low_limit = info->low_limit + length;
1781
1782 /* Check highest gap, which does not precede any rbtree node */
1783 gap_start = mm->highest_vm_end;
1784 if (gap_start <= high_limit)
1785 goto found_highest;
1786
1787 /* Check if rbtree root looks promising */
1788 if (RB_EMPTY_ROOT(&mm->mm_rb))
1789 return -ENOMEM;
1790 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1791 if (vma->rb_subtree_gap < length)
1792 return -ENOMEM;
1793
1794 while (true) {
1795 /* Visit right subtree if it looks promising */
1796 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1797 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1798 struct vm_area_struct *right =
1799 rb_entry(vma->vm_rb.rb_right,
1800 struct vm_area_struct, vm_rb);
1801 if (right->rb_subtree_gap >= length) {
1802 vma = right;
1803 continue;
1804 }
1805 }
1806
1807check_current:
1808 /* Check if current node has a suitable gap */
1809 gap_end = vma->vm_start;
1810 if (gap_end < low_limit)
1811 return -ENOMEM;
1812 if (gap_start <= high_limit && gap_end - gap_start >= length)
1813 goto found;
1814
1815 /* Visit left subtree if it looks promising */
1816 if (vma->vm_rb.rb_left) {
1817 struct vm_area_struct *left =
1818 rb_entry(vma->vm_rb.rb_left,
1819 struct vm_area_struct, vm_rb);
1820 if (left->rb_subtree_gap >= length) {
1821 vma = left;
1822 continue;
1823 }
1824 }
1825
1826 /* Go back up the rbtree to find next candidate node */
1827 while (true) {
1828 struct rb_node *prev = &vma->vm_rb;
1829 if (!rb_parent(prev))
1830 return -ENOMEM;
1831 vma = rb_entry(rb_parent(prev),
1832 struct vm_area_struct, vm_rb);
1833 if (prev == vma->vm_rb.rb_right) {
1834 gap_start = vma->vm_prev ?
1835 vma->vm_prev->vm_end : 0;
1836 goto check_current;
1837 }
1838 }
1839 }
1840
1841found:
1842 /* We found a suitable gap. Clip it with the original high_limit. */
1843 if (gap_end > info->high_limit)
1844 gap_end = info->high_limit;
1845
1846found_highest:
1847 /* Compute highest gap address at the desired alignment */
1848 gap_end -= info->length;
1849 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1850
1851 VM_BUG_ON(gap_end < info->low_limit);
1852 VM_BUG_ON(gap_end < gap_start);
1853 return gap_end;
1854}
1855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856/* Get an address range which is currently unmapped.
1857 * For shmat() with addr=0.
1858 *
1859 * Ugly calling convention alert:
1860 * Return value with the low bits set means error value,
1861 * ie
1862 * if (ret & ~PAGE_MASK)
1863 * error = ret;
1864 *
1865 * This function "knows" that -ENOMEM has the bits set.
1866 */
1867#ifndef HAVE_ARCH_UNMAPPED_AREA
1868unsigned long
1869arch_get_unmapped_area(struct file *filp, unsigned long addr,
1870 unsigned long len, unsigned long pgoff, unsigned long flags)
1871{
1872 struct mm_struct *mm = current->mm;
1873 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001874 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001876 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 return -ENOMEM;
1878
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001879 if (flags & MAP_FIXED)
1880 return addr;
1881
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 if (addr) {
1883 addr = PAGE_ALIGN(addr);
1884 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001885 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 (!vma || addr + len <= vma->vm_start))
1887 return addr;
1888 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001890 info.flags = 0;
1891 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001892 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001893 info.high_limit = TASK_SIZE;
1894 info.align_mask = 0;
1895 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896}
1897#endif
1898
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899/*
1900 * This mmap-allocator allocates new areas top-down from below the
1901 * stack's low limit (the base):
1902 */
1903#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1904unsigned long
1905arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1906 const unsigned long len, const unsigned long pgoff,
1907 const unsigned long flags)
1908{
1909 struct vm_area_struct *vma;
1910 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001911 unsigned long addr = addr0;
1912 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
1914 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001915 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 return -ENOMEM;
1917
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001918 if (flags & MAP_FIXED)
1919 return addr;
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 /* requesting a specific address */
1922 if (addr) {
1923 addr = PAGE_ALIGN(addr);
1924 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001925 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 (!vma || addr + len <= vma->vm_start))
1927 return addr;
1928 }
1929
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001930 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1931 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001932 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001933 info.high_limit = mm->mmap_base;
1934 info.align_mask = 0;
1935 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 /*
1938 * A failed mmap() very likely causes application failure,
1939 * so fall back to the bottom-up function here. This scenario
1940 * can happen with large stack limits and large mmap()
1941 * allocations.
1942 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001943 if (addr & ~PAGE_MASK) {
1944 VM_BUG_ON(addr != -ENOMEM);
1945 info.flags = 0;
1946 info.low_limit = TASK_UNMAPPED_BASE;
1947 info.high_limit = TASK_SIZE;
1948 addr = vm_unmapped_area(&info);
1949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951 return addr;
1952}
1953#endif
1954
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955unsigned long
1956get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1957 unsigned long pgoff, unsigned long flags)
1958{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001959 unsigned long (*get_area)(struct file *, unsigned long,
1960 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
Al Viro9206de92009-12-03 15:23:11 -05001962 unsigned long error = arch_mmap_check(addr, len, flags);
1963 if (error)
1964 return error;
1965
1966 /* Careful about overflows.. */
1967 if (len > TASK_SIZE)
1968 return -ENOMEM;
1969
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001970 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04001971 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001972 get_area = file->f_op->get_unmapped_area;
1973 addr = get_area(file, addr, len, pgoff, flags);
1974 if (IS_ERR_VALUE(addr))
1975 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001977 if (addr > TASK_SIZE - len)
1978 return -ENOMEM;
1979 if (addr & ~PAGE_MASK)
1980 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001981
Al Viro9ac4ed42012-05-30 17:13:15 -04001982 addr = arch_rebalance_pgtables(addr, len);
1983 error = security_mmap_addr(addr);
1984 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985}
1986
1987EXPORT_SYMBOL(get_unmapped_area);
1988
1989/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001990struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
1992 struct vm_area_struct *vma = NULL;
1993
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001994 /* Check the cache first. */
1995 /* (Cache hit rate is typically around 35%.) */
Jan Stancekb6a9b7f2013-04-04 11:35:10 -07001996 vma = ACCESS_ONCE(mm->mmap_cache);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001997 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1998 struct rb_node *rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002000 rb_node = mm->mm_rb.rb_node;
2001 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002003 while (rb_node) {
2004 struct vm_area_struct *vma_tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002006 vma_tmp = rb_entry(rb_node,
2007 struct vm_area_struct, vm_rb);
2008
2009 if (vma_tmp->vm_end > addr) {
2010 vma = vma_tmp;
2011 if (vma_tmp->vm_start <= addr)
2012 break;
2013 rb_node = rb_node->rb_left;
2014 } else
2015 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 }
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002017 if (vma)
2018 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
2020 return vma;
2021}
2022
2023EXPORT_SYMBOL(find_vma);
2024
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002025/*
2026 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002027 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028struct vm_area_struct *
2029find_vma_prev(struct mm_struct *mm, unsigned long addr,
2030 struct vm_area_struct **pprev)
2031{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002032 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002034 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002035 if (vma) {
2036 *pprev = vma->vm_prev;
2037 } else {
2038 struct rb_node *rb_node = mm->mm_rb.rb_node;
2039 *pprev = NULL;
2040 while (rb_node) {
2041 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2042 rb_node = rb_node->rb_right;
2043 }
2044 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002045 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046}
2047
2048/*
2049 * Verify that the stack growth is acceptable and
2050 * update accounting. This is shared with both the
2051 * grow-up and grow-down cases.
2052 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002053static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054{
2055 struct mm_struct *mm = vma->vm_mm;
2056 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08002057 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
2059 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002060 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 return -ENOMEM;
2062
2063 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08002064 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 return -ENOMEM;
2066
2067 /* mlock limit tests */
2068 if (vma->vm_flags & VM_LOCKED) {
2069 unsigned long locked;
2070 unsigned long limit;
2071 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002072 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2073 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 if (locked > limit && !capable(CAP_IPC_LOCK))
2075 return -ENOMEM;
2076 }
2077
Adam Litke0d59a012007-01-30 14:35:39 -08002078 /* Check to ensure the stack will not grow into a hugetlb-only region */
2079 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2080 vma->vm_end - size;
2081 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2082 return -EFAULT;
2083
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 /*
2085 * Overcommit.. This must be the final test, as it will
2086 * update security statistics.
2087 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002088 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 return -ENOMEM;
2090
2091 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 if (vma->vm_flags & VM_LOCKED)
2093 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002094 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 return 0;
2096}
2097
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002098#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002100 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2101 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002103int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
2105 int error;
2106
2107 if (!(vma->vm_flags & VM_GROWSUP))
2108 return -EFAULT;
2109
2110 /*
2111 * We must make sure the anon_vma is allocated
2112 * so that the anon_vma locking is not a noop.
2113 */
2114 if (unlikely(anon_vma_prepare(vma)))
2115 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002116 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117
2118 /*
2119 * vma->vm_start/vm_end cannot change under us because the caller
2120 * is required to hold the mmap_sem in read mode. We need the
2121 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002122 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 */
Helge Deller06b32f32006-12-19 19:28:33 +01002124 if (address < PAGE_ALIGN(address+4))
2125 address = PAGE_ALIGN(address+4);
2126 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002127 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002128 return -ENOMEM;
2129 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 error = 0;
2131
2132 /* Somebody else might have raced and expanded it already */
2133 if (address > vma->vm_end) {
2134 unsigned long size, grow;
2135
2136 size = address - vma->vm_start;
2137 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2138
Hugh Dickins42c36f62011-05-09 17:44:42 -07002139 error = -ENOMEM;
2140 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2141 error = acct_stack_growth(vma, size, grow);
2142 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002143 /*
2144 * vma_gap_update() doesn't support concurrent
2145 * updates, but we only hold a shared mmap_sem
2146 * lock here, so we need to protect against
2147 * concurrent vma expansions.
2148 * vma_lock_anon_vma() doesn't help here, as
2149 * we don't guarantee that all growable vmas
2150 * in a mm share the same root anon vma.
2151 * So, we reuse mm->page_table_lock to guard
2152 * against concurrent vma expansions.
2153 */
2154 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002155 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002156 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002157 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002158 if (vma->vm_next)
2159 vma_gap_update(vma->vm_next);
2160 else
2161 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002162 spin_unlock(&vma->vm_mm->page_table_lock);
2163
Hugh Dickins42c36f62011-05-09 17:44:42 -07002164 perf_event_mmap(vma);
2165 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002168 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002169 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002170 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 return error;
2172}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002173#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2174
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175/*
2176 * vma is the first one with address < vma->vm_start. Have to extend vma.
2177 */
Michal Hockod05f3162011-05-24 17:11:44 -07002178int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002179 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180{
2181 int error;
2182
2183 /*
2184 * We must make sure the anon_vma is allocated
2185 * so that the anon_vma locking is not a noop.
2186 */
2187 if (unlikely(anon_vma_prepare(vma)))
2188 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002189
2190 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002191 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002192 if (error)
2193 return error;
2194
Rik van Rielbb4a3402010-08-09 17:18:37 -07002195 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
2197 /*
2198 * vma->vm_start/vm_end cannot change under us because the caller
2199 * is required to hold the mmap_sem in read mode. We need the
2200 * anon_vma lock to serialize against concurrent expand_stacks.
2201 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202
2203 /* Somebody else might have raced and expanded it already */
2204 if (address < vma->vm_start) {
2205 unsigned long size, grow;
2206
2207 size = vma->vm_end - address;
2208 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2209
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002210 error = -ENOMEM;
2211 if (grow <= vma->vm_pgoff) {
2212 error = acct_stack_growth(vma, size, grow);
2213 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002214 /*
2215 * vma_gap_update() doesn't support concurrent
2216 * updates, but we only hold a shared mmap_sem
2217 * lock here, so we need to protect against
2218 * concurrent vma expansions.
2219 * vma_lock_anon_vma() doesn't help here, as
2220 * we don't guarantee that all growable vmas
2221 * in a mm share the same root anon vma.
2222 * So, we reuse mm->page_table_lock to guard
2223 * against concurrent vma expansions.
2224 */
2225 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002226 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002227 vma->vm_start = address;
2228 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002229 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002230 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002231 spin_unlock(&vma->vm_mm->page_table_lock);
2232
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002233 perf_event_mmap(vma);
2234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 }
2236 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002237 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002238 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002239 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 return error;
2241}
2242
Linus Torvalds09884962013-02-27 08:36:04 -08002243/*
2244 * Note how expand_stack() refuses to expand the stack all the way to
2245 * abut the next virtual mapping, *unless* that mapping itself is also
2246 * a stack mapping. We want to leave room for a guard page, after all
2247 * (the guard page itself is not added here, that is done by the
2248 * actual page faulting logic)
2249 *
2250 * This matches the behavior of the guard page logic (see mm/memory.c:
2251 * check_stack_guard_page()), which only allows the guard page to be
2252 * removed under these circumstances.
2253 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002254#ifdef CONFIG_STACK_GROWSUP
2255int expand_stack(struct vm_area_struct *vma, unsigned long address)
2256{
Linus Torvalds09884962013-02-27 08:36:04 -08002257 struct vm_area_struct *next;
2258
2259 address &= PAGE_MASK;
2260 next = vma->vm_next;
2261 if (next && next->vm_start == address + PAGE_SIZE) {
2262 if (!(next->vm_flags & VM_GROWSUP))
2263 return -ENOMEM;
2264 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002265 return expand_upwards(vma, address);
2266}
2267
2268struct vm_area_struct *
2269find_extend_vma(struct mm_struct *mm, unsigned long addr)
2270{
2271 struct vm_area_struct *vma, *prev;
2272
2273 addr &= PAGE_MASK;
2274 vma = find_vma_prev(mm, addr, &prev);
2275 if (vma && (vma->vm_start <= addr))
2276 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002277 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002278 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002279 if (prev->vm_flags & VM_LOCKED)
2280 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002281 return prev;
2282}
2283#else
2284int expand_stack(struct vm_area_struct *vma, unsigned long address)
2285{
Linus Torvalds09884962013-02-27 08:36:04 -08002286 struct vm_area_struct *prev;
2287
2288 address &= PAGE_MASK;
2289 prev = vma->vm_prev;
2290 if (prev && prev->vm_end == address) {
2291 if (!(prev->vm_flags & VM_GROWSDOWN))
2292 return -ENOMEM;
2293 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002294 return expand_downwards(vma, address);
2295}
2296
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297struct vm_area_struct *
2298find_extend_vma(struct mm_struct * mm, unsigned long addr)
2299{
2300 struct vm_area_struct * vma;
2301 unsigned long start;
2302
2303 addr &= PAGE_MASK;
2304 vma = find_vma(mm,addr);
2305 if (!vma)
2306 return NULL;
2307 if (vma->vm_start <= addr)
2308 return vma;
2309 if (!(vma->vm_flags & VM_GROWSDOWN))
2310 return NULL;
2311 start = vma->vm_start;
2312 if (expand_stack(vma, addr))
2313 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002314 if (vma->vm_flags & VM_LOCKED)
2315 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 return vma;
2317}
2318#endif
2319
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002321 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002323 *
2324 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002326static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002328 unsigned long nr_accounted = 0;
2329
Hugh Dickins365e9c872005-10-29 18:16:18 -07002330 /* Update high watermark before we lower total_vm */
2331 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002333 long nrpages = vma_pages(vma);
2334
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002335 if (vma->vm_flags & VM_ACCOUNT)
2336 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002337 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002338 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002339 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002340 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 validate_mm(mm);
2342}
2343
2344/*
2345 * Get rid of page table information in the indicated region.
2346 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002347 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 */
2349static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002350 struct vm_area_struct *vma, struct vm_area_struct *prev,
2351 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352{
Hugh Dickinse0da3822005-04-19 13:29:15 -07002353 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002354 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355
2356 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002357 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002358 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002359 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002360 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002361 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002362 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363}
2364
2365/*
2366 * Create a list of vma's touched by the unmap, removing them from the mm's
2367 * vma list as we go..
2368 */
2369static void
2370detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2371 struct vm_area_struct *prev, unsigned long end)
2372{
2373 struct vm_area_struct **insertion_point;
2374 struct vm_area_struct *tail_vma = NULL;
2375
2376 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002377 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002379 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 mm->map_count--;
2381 tail_vma = vma;
2382 vma = vma->vm_next;
2383 } while (vma && vma->vm_start < end);
2384 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002385 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002386 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002387 vma_gap_update(vma);
2388 } else
2389 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 tail_vma->vm_next = NULL;
2391 mm->mmap_cache = NULL; /* Kill the cache. */
2392}
2393
2394/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002395 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2396 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002398static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 unsigned long addr, int new_below)
2400{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08002402 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403
Andi Kleena5516432008-07-23 21:27:41 -07002404 if (is_vm_hugetlb_page(vma) && (addr &
2405 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 return -EINVAL;
2407
Christoph Lametere94b1762006-12-06 20:33:17 -08002408 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002410 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
2412 /* most fields are the same, copy all, and then fixup */
2413 *new = *vma;
2414
Rik van Riel5beb4932010-03-05 13:42:07 -08002415 INIT_LIST_HEAD(&new->anon_vma_chain);
2416
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 if (new_below)
2418 new->vm_end = addr;
2419 else {
2420 new->vm_start = addr;
2421 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2422 }
2423
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002424 err = vma_dup_policy(vma, new);
2425 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002426 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427
Rik van Riel5beb4932010-03-05 13:42:07 -08002428 if (anon_vma_clone(new, vma))
2429 goto out_free_mpol;
2430
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002431 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 get_file(new->vm_file);
2433
2434 if (new->vm_ops && new->vm_ops->open)
2435 new->vm_ops->open(new);
2436
2437 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002438 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2440 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002441 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
Rik van Riel5beb4932010-03-05 13:42:07 -08002443 /* Success. */
2444 if (!err)
2445 return 0;
2446
2447 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002448 if (new->vm_ops && new->vm_ops->close)
2449 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002450 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002451 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002452 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002453 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002454 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002455 out_free_vma:
2456 kmem_cache_free(vm_area_cachep, new);
2457 out_err:
2458 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459}
2460
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002461/*
2462 * Split a vma into two pieces at address 'addr', a new vma is allocated
2463 * either for the first part or the tail.
2464 */
2465int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2466 unsigned long addr, int new_below)
2467{
2468 if (mm->map_count >= sysctl_max_map_count)
2469 return -ENOMEM;
2470
2471 return __split_vma(mm, vma, addr, new_below);
2472}
2473
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474/* Munmap is split into 2 main parts -- this part which finds
2475 * what needs doing, and the areas themselves, which do the
2476 * work. This now handles partial unmappings.
2477 * Jeremy Fitzhardinge <jeremy@goop.org>
2478 */
2479int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2480{
2481 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002482 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
2484 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2485 return -EINVAL;
2486
2487 if ((len = PAGE_ALIGN(len)) == 0)
2488 return -EINVAL;
2489
2490 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002491 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002492 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002494 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002495 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496
2497 /* if it doesn't overlap, we have nothing.. */
2498 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002499 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 return 0;
2501
2502 /*
2503 * If we need to split any vma, do it now to save pain later.
2504 *
2505 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2506 * unmapped vm_area_struct will remain in use: so lower split_vma
2507 * places tmp vma above, and higher split_vma places tmp vma below.
2508 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002509 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002510 int error;
2511
2512 /*
2513 * Make sure that map_count on return from munmap() will
2514 * not exceed its limit; but let map_count go just above
2515 * its limit temporarily, to help free resources as expected.
2516 */
2517 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2518 return -ENOMEM;
2519
2520 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 if (error)
2522 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002523 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 }
2525
2526 /* Does it split the last one? */
2527 last = find_vma(mm, end);
2528 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002529 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (error)
2531 return error;
2532 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002533 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
2535 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002536 * unlock any mlock()ed ranges before detaching vmas
2537 */
2538 if (mm->locked_vm) {
2539 struct vm_area_struct *tmp = vma;
2540 while (tmp && tmp->vm_start < end) {
2541 if (tmp->vm_flags & VM_LOCKED) {
2542 mm->locked_vm -= vma_pages(tmp);
2543 munlock_vma_pages_all(tmp);
2544 }
2545 tmp = tmp->vm_next;
2546 }
2547 }
2548
2549 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 * Remove the vma's, and unmap the actual pages
2551 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002552 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2553 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554
2555 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002556 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
2558 return 0;
2559}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Al Virobfce2812012-04-20 21:57:04 -04002561int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002562{
2563 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002564 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002565
2566 down_write(&mm->mmap_sem);
2567 ret = do_munmap(mm, start, len);
2568 up_write(&mm->mmap_sem);
2569 return ret;
2570}
2571EXPORT_SYMBOL(vm_munmap);
2572
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002573SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002576 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577}
2578
2579static inline void verify_mm_writelocked(struct mm_struct *mm)
2580{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002581#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2583 WARN_ON(1);
2584 up_read(&mm->mmap_sem);
2585 }
2586#endif
2587}
2588
2589/*
2590 * this is really a simplified "do_mmap". it only handles
2591 * anonymous maps. eventually we may be able to do some
2592 * brk-specific accounting here.
2593 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002594static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595{
2596 struct mm_struct * mm = current->mm;
2597 struct vm_area_struct * vma, * prev;
2598 unsigned long flags;
2599 struct rb_node ** rb_link, * rb_parent;
2600 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002601 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
2603 len = PAGE_ALIGN(len);
2604 if (!len)
2605 return addr;
2606
Kirill Korotaev3a459752006-09-07 14:17:04 +04002607 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2608
Al Viro2c6a1012009-12-03 19:40:46 -05002609 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2610 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002611 return error;
2612
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002613 error = mlock_future_check(mm, mm->def_flags, len);
2614 if (error)
2615 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
2617 /*
2618 * mm->mmap_sem is required to protect against another thread
2619 * changing the mappings in case we sleep.
2620 */
2621 verify_mm_writelocked(mm);
2622
2623 /*
2624 * Clear old maps. this also does some error checking for us
2625 */
2626 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002627 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 if (do_munmap(mm, addr, len))
2629 return -ENOMEM;
2630 goto munmap_back;
2631 }
2632
2633 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002634 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 return -ENOMEM;
2636
2637 if (mm->map_count > sysctl_max_map_count)
2638 return -ENOMEM;
2639
Al Viro191c5422012-02-13 03:58:52 +00002640 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 return -ENOMEM;
2642
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002644 vma = vma_merge(mm, prev, addr, addr + len, flags,
2645 NULL, NULL, pgoff, NULL);
2646 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 goto out;
2648
2649 /*
2650 * create a vma struct for an anonymous mapping
2651 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002652 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 if (!vma) {
2654 vm_unacct_memory(len >> PAGE_SHIFT);
2655 return -ENOMEM;
2656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
Rik van Riel5beb4932010-03-05 13:42:07 -08002658 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 vma->vm_mm = mm;
2660 vma->vm_start = addr;
2661 vma->vm_end = addr + len;
2662 vma->vm_pgoff = pgoff;
2663 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002664 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 vma_link(mm, vma, prev, rb_link, rb_parent);
2666out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002667 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002669 if (flags & VM_LOCKED)
2670 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002671 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 return addr;
2673}
2674
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002675unsigned long vm_brk(unsigned long addr, unsigned long len)
2676{
2677 struct mm_struct *mm = current->mm;
2678 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002679 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002680
2681 down_write(&mm->mmap_sem);
2682 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002683 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002684 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002685 if (populate)
2686 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002687 return ret;
2688}
2689EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
2691/* Release all mmaps. */
2692void exit_mmap(struct mm_struct *mm)
2693{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002694 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002695 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 unsigned long nr_accounted = 0;
2697
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002698 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002699 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002700
Rik van Rielba470de2008-10-18 20:26:50 -07002701 if (mm->locked_vm) {
2702 vma = mm->mmap;
2703 while (vma) {
2704 if (vma->vm_flags & VM_LOCKED)
2705 munlock_vma_pages_all(vma);
2706 vma = vma->vm_next;
2707 }
2708 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002709
2710 arch_exit_mmap(mm);
2711
Rik van Rielba470de2008-10-18 20:26:50 -07002712 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002713 if (!vma) /* Can happen if dup_mmap() received an OOM */
2714 return;
2715
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002718 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002719 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002720 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002721 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002722
Hugh Dickins6ee86302013-04-29 15:07:44 -07002723 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002724 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002727 * Walk the list again, actually closing and freeing it,
2728 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002730 while (vma) {
2731 if (vma->vm_flags & VM_ACCOUNT)
2732 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002733 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002734 }
2735 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002736
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -08002737 WARN_ON(atomic_long_read(&mm->nr_ptes) >
2738 (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739}
2740
2741/* Insert vm structure into process list sorted by address
2742 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002743 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002745int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746{
Hugh Dickins6597d782012-10-08 16:29:07 -07002747 struct vm_area_struct *prev;
2748 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749
2750 /*
2751 * The vm_pgoff of a purely anonymous vma should be irrelevant
2752 * until its first write fault, when page's anon_vma and index
2753 * are set. But now set the vm_pgoff it will almost certainly
2754 * end up with (unless mremap moves it elsewhere before that
2755 * first wfault), so /proc/pid/maps tells a consistent story.
2756 *
2757 * By setting it to reflect the virtual start address of the
2758 * vma, merges and splits can happen in a seamless way, just
2759 * using the existing file pgoff checks and manipulations.
2760 * Similarly in do_mmap_pgoff and in do_brk.
2761 */
2762 if (!vma->vm_file) {
2763 BUG_ON(vma->anon_vma);
2764 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2765 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002766 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2767 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002769 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002770 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002771 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302772
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 vma_link(mm, vma, prev, rb_link, rb_parent);
2774 return 0;
2775}
2776
2777/*
2778 * Copy the vma structure to a new location in the same mm,
2779 * prior to moving page table entries, to effect an mremap move.
2780 */
2781struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002782 unsigned long addr, unsigned long len, pgoff_t pgoff,
2783 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784{
2785 struct vm_area_struct *vma = *vmap;
2786 unsigned long vma_start = vma->vm_start;
2787 struct mm_struct *mm = vma->vm_mm;
2788 struct vm_area_struct *new_vma, *prev;
2789 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002790 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
2792 /*
2793 * If anonymous vma has not yet been faulted, update new pgoff
2794 * to match new location, to increase its chance of merging.
2795 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002796 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002798 faulted_in_anon_vma = false;
2799 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800
Hugh Dickins6597d782012-10-08 16:29:07 -07002801 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2802 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2804 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2805 if (new_vma) {
2806 /*
2807 * Source vma may have been merged into new_vma
2808 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002809 if (unlikely(vma_start >= new_vma->vm_start &&
2810 vma_start < new_vma->vm_end)) {
2811 /*
2812 * The only way we can get a vma_merge with
2813 * self during an mremap is if the vma hasn't
2814 * been faulted in yet and we were allowed to
2815 * reset the dst vma->vm_pgoff to the
2816 * destination address of the mremap to allow
2817 * the merge to happen. mremap must change the
2818 * vm_pgoff linearity between src and dst vmas
2819 * (in turn preventing a vma_merge) to be
2820 * safe. It is only safe to keep the vm_pgoff
2821 * linear if there are no pages mapped yet.
2822 */
2823 VM_BUG_ON(faulted_in_anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002824 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002825 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002826 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002828 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 if (new_vma) {
2830 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 new_vma->vm_start = addr;
2832 new_vma->vm_end = addr + len;
2833 new_vma->vm_pgoff = pgoff;
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002834 if (vma_dup_policy(vma, new_vma))
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002835 goto out_free_vma;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002836 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2837 if (anon_vma_clone(new_vma, vma))
2838 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002839 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 get_file(new_vma->vm_file);
2841 if (new_vma->vm_ops && new_vma->vm_ops->open)
2842 new_vma->vm_ops->open(new_vma);
2843 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002844 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 }
2846 }
2847 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002848
2849 out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002850 mpol_put(vma_policy(new_vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002851 out_free_vma:
2852 kmem_cache_free(vm_area_cachep, new_vma);
2853 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002855
2856/*
2857 * Return true if the calling process may expand its vm space by the passed
2858 * number of pages
2859 */
2860int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2861{
2862 unsigned long cur = mm->total_vm; /* pages */
2863 unsigned long lim;
2864
Jiri Slaby59e99e52010-03-05 13:41:44 -08002865 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002866
2867 if (cur + npages > lim)
2868 return 0;
2869 return 1;
2870}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002871
2872
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002873static int special_mapping_fault(struct vm_area_struct *vma,
2874 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002875{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002876 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002877 struct page **pages;
2878
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002879 /*
2880 * special mappings have no vm_file, and in that case, the mm
2881 * uses vm_pgoff internally. So we have to subtract it from here.
2882 * We are allowed to do this because we are the mm; do not copy
2883 * this code into drivers!
2884 */
2885 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002886
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002887 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2888 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002889
2890 if (*pages) {
2891 struct page *page = *pages;
2892 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002893 vmf->page = page;
2894 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002895 }
2896
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002897 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002898}
2899
2900/*
2901 * Having a close hook prevents vma merging regardless of flags.
2902 */
2903static void special_mapping_close(struct vm_area_struct *vma)
2904{
2905}
2906
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002907static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002908 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002909 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002910};
2911
2912/*
2913 * Called with mm->mmap_sem held for writing.
2914 * Insert a new vma covering the given region, with the given flags.
2915 * Its pages are supplied by the given array of struct page *.
2916 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2917 * The region past the last page supplied will always produce SIGBUS.
2918 * The array pointer and the pages it points to are assumed to stay alive
2919 * for as long as this mapping might exist.
2920 */
Stefani Seibold3935ed62014-03-17 23:22:02 +01002921struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002922 unsigned long addr, unsigned long len,
2923 unsigned long vm_flags, struct page **pages)
2924{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002925 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002926 struct vm_area_struct *vma;
2927
2928 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2929 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01002930 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002931
Rik van Riel5beb4932010-03-05 13:42:07 -08002932 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002933 vma->vm_mm = mm;
2934 vma->vm_start = addr;
2935 vma->vm_end = addr + len;
2936
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002937 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07002938 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002939
2940 vma->vm_ops = &special_mapping_vmops;
2941 vma->vm_private_data = pages;
2942
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002943 ret = insert_vm_struct(mm, vma);
2944 if (ret)
2945 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002946
2947 mm->total_vm += len >> PAGE_SHIFT;
2948
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002949 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002950
Stefani Seibold3935ed62014-03-17 23:22:02 +01002951 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002952
2953out:
2954 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01002955 return ERR_PTR(ret);
2956}
2957
2958int install_special_mapping(struct mm_struct *mm,
2959 unsigned long addr, unsigned long len,
2960 unsigned long vm_flags, struct page **pages)
2961{
2962 struct vm_area_struct *vma = _install_special_mapping(mm,
2963 addr, len, vm_flags, pages);
2964
2965 if (IS_ERR(vma))
2966 return PTR_ERR(vma);
2967 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002968}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002969
2970static DEFINE_MUTEX(mm_all_locks_mutex);
2971
Peter Zijlstra454ed842008-08-11 09:30:25 +02002972static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002973{
Michel Lespinassebf181b92012-10-08 16:31:39 -07002974 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002975 /*
2976 * The LSB of head.next can't change from under us
2977 * because we hold the mm_all_locks_mutex.
2978 */
Jiri Kosina572043c2013-01-11 14:31:59 -08002979 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002980 /*
2981 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00002982 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002983 * the same anon_vma we won't take it again.
2984 *
2985 * No need of atomic instructions here, head.next
2986 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00002987 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002988 */
2989 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07002990 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002991 BUG();
2992 }
2993}
2994
Peter Zijlstra454ed842008-08-11 09:30:25 +02002995static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002996{
2997 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2998 /*
2999 * AS_MM_ALL_LOCKS can't change from under us because
3000 * we hold the mm_all_locks_mutex.
3001 *
3002 * Operations on ->flags have to be atomic because
3003 * even if AS_MM_ALL_LOCKS is stable thanks to the
3004 * mm_all_locks_mutex, there may be other cpus
3005 * changing other bitflags in parallel to us.
3006 */
3007 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3008 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003009 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003010 }
3011}
3012
3013/*
3014 * This operation locks against the VM for all pte/vma/mm related
3015 * operations that could ever happen on a certain mm. This includes
3016 * vmtruncate, try_to_unmap, and all page faults.
3017 *
3018 * The caller must take the mmap_sem in write mode before calling
3019 * mm_take_all_locks(). The caller isn't allowed to release the
3020 * mmap_sem until mm_drop_all_locks() returns.
3021 *
3022 * mmap_sem in write mode is required in order to block all operations
3023 * that could modify pagetables and free pages without need of
3024 * altering the vma layout (for example populate_range() with
3025 * nonlinear vmas). It's also needed in write mode to avoid new
3026 * anon_vmas to be associated with existing vmas.
3027 *
3028 * A single task can't take more than one mm_take_all_locks() in a row
3029 * or it would deadlock.
3030 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003031 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003032 * mapping->flags avoid to take the same lock twice, if more than one
3033 * vma in this mm is backed by the same anon_vma or address_space.
3034 *
3035 * We can take all the locks in random order because the VM code
Yuanhan Liu631b0cf2013-02-04 14:28:48 -08003036 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003037 * takes more than one of them in a row. Secondly we're protected
3038 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3039 *
3040 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3041 * that may have to take thousand of locks.
3042 *
3043 * mm_take_all_locks() can fail if it's interrupted by signals.
3044 */
3045int mm_take_all_locks(struct mm_struct *mm)
3046{
3047 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003048 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003049
3050 BUG_ON(down_read_trylock(&mm->mmap_sem));
3051
3052 mutex_lock(&mm_all_locks_mutex);
3053
3054 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3055 if (signal_pending(current))
3056 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003057 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003058 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003059 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003060
3061 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3062 if (signal_pending(current))
3063 goto out_unlock;
3064 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003065 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3066 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003067 }
3068
Kautuk Consul584cff52011-10-31 17:08:59 -07003069 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003070
3071out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003072 mm_drop_all_locks(mm);
3073 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003074}
3075
3076static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3077{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003078 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003079 /*
3080 * The LSB of head.next can't change to 0 from under
3081 * us because we hold the mm_all_locks_mutex.
3082 *
3083 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003084 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003085 * never see our bitflag.
3086 *
3087 * No need of atomic instructions here, head.next
3088 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003089 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003090 */
3091 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003092 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003093 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003094 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003095 }
3096}
3097
3098static void vm_unlock_mapping(struct address_space *mapping)
3099{
3100 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3101 /*
3102 * AS_MM_ALL_LOCKS can't change to 0 from under us
3103 * because we hold the mm_all_locks_mutex.
3104 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003105 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003106 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3107 &mapping->flags))
3108 BUG();
3109 }
3110}
3111
3112/*
3113 * The mmap_sem cannot be released by the caller until
3114 * mm_drop_all_locks() returns.
3115 */
3116void mm_drop_all_locks(struct mm_struct *mm)
3117{
3118 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003119 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003120
3121 BUG_ON(down_read_trylock(&mm->mmap_sem));
3122 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3123
3124 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3125 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003126 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3127 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003128 if (vma->vm_file && vma->vm_file->f_mapping)
3129 vm_unlock_mapping(vma->vm_file->f_mapping);
3130 }
3131
3132 mutex_unlock(&mm_all_locks_mutex);
3133}
David Howells8feae132009-01-08 12:04:47 +00003134
3135/*
3136 * initialise the VMA slab
3137 */
3138void __init mmap_init(void)
3139{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003140 int ret;
3141
3142 ret = percpu_counter_init(&vm_committed_as, 0);
3143 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003144}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003145
3146/*
3147 * Initialise sysctl_user_reserve_kbytes.
3148 *
3149 * This is intended to prevent a user from starting a single memory hogging
3150 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3151 * mode.
3152 *
3153 * The default value is min(3% of free memory, 128MB)
3154 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3155 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003156static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003157{
3158 unsigned long free_kbytes;
3159
3160 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3161
3162 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3163 return 0;
3164}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003165subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003166
3167/*
3168 * Initialise sysctl_admin_reserve_kbytes.
3169 *
3170 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3171 * to log in and kill a memory hogging process.
3172 *
3173 * Systems with more than 256MB will reserve 8MB, enough to recover
3174 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3175 * only reserve 3% of free pages by default.
3176 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003177static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003178{
3179 unsigned long free_kbytes;
3180
3181 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3182
3183 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3184 return 0;
3185}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003186subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003187
3188/*
3189 * Reinititalise user and admin reserves if memory is added or removed.
3190 *
3191 * The default user reserve max is 128MB, and the default max for the
3192 * admin reserve is 8MB. These are usually, but not always, enough to
3193 * enable recovery from a memory hogging process using login/sshd, a shell,
3194 * and tools like top. It may make sense to increase or even disable the
3195 * reserve depending on the existence of swap or variations in the recovery
3196 * tools. So, the admin may have changed them.
3197 *
3198 * If memory is added and the reserves have been eliminated or increased above
3199 * the default max, then we'll trust the admin.
3200 *
3201 * If memory is removed and there isn't enough free memory, then we
3202 * need to reset the reserves.
3203 *
3204 * Otherwise keep the reserve set by the admin.
3205 */
3206static int reserve_mem_notifier(struct notifier_block *nb,
3207 unsigned long action, void *data)
3208{
3209 unsigned long tmp, free_kbytes;
3210
3211 switch (action) {
3212 case MEM_ONLINE:
3213 /* Default max is 128MB. Leave alone if modified by operator. */
3214 tmp = sysctl_user_reserve_kbytes;
3215 if (0 < tmp && tmp < (1UL << 17))
3216 init_user_reserve();
3217
3218 /* Default max is 8MB. Leave alone if modified by operator. */
3219 tmp = sysctl_admin_reserve_kbytes;
3220 if (0 < tmp && tmp < (1UL << 13))
3221 init_admin_reserve();
3222
3223 break;
3224 case MEM_OFFLINE:
3225 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3226
3227 if (sysctl_user_reserve_kbytes > free_kbytes) {
3228 init_user_reserve();
3229 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3230 sysctl_user_reserve_kbytes);
3231 }
3232
3233 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3234 init_admin_reserve();
3235 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3236 sysctl_admin_reserve_kbytes);
3237 }
3238 break;
3239 default:
3240 break;
3241 }
3242 return NOTIFY_OK;
3243}
3244
3245static struct notifier_block reserve_mem_nb = {
3246 .notifier_call = reserve_mem_notifier,
3247};
3248
3249static int __meminit init_reserve_notifier(void)
3250{
3251 if (register_hotmemory_notifier(&reserve_mem_nb))
3252 printk("Failed registering memory add/remove notifier for admin reserve");
3253
3254 return 0;
3255}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003256subsys_initcall(init_reserve_notifier);