blob: b1202cf81f4bddf1dcc8393f1231ed85b1e64e6e [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>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070013#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/shm.h>
15#include <linux/mman.h>
16#include <linux/pagemap.h>
17#include <linux/swap.h>
18#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/init.h>
21#include <linux/file.h>
22#include <linux/fs.h>
23#include <linux/personality.h>
24#include <linux/security.h>
25#include <linux/hugetlb.h>
26#include <linux/profile.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040027#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/mount.h>
29#include <linux/mempolicy.h>
30#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070031#include <linux/mmu_notifier.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020032#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040033#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080034#include <linux/khugepaged.h>
Srikar Dronamraju2b144492012-02-09 14:56:42 +053035#include <linux/uprobes.h>
Michel Lespinassed3737182012-12-11 16:01:38 -080036#include <linux/rbtree_augmented.h>
Clark Williamscf4aebc22013-02-07 09:46:59 -060037#include <linux/sched/sysctl.h>
Andrew Shewmaker16408792013-04-29 15:08:12 -070038#include <linux/notifier.h>
39#include <linux/memory.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/uaccess.h>
42#include <asm/cacheflush.h>
43#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020044#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Jan Beulich42b77722008-07-23 21:27:10 -070046#include "internal.h"
47
Kirill Korotaev3a459752006-09-07 14:17:04 +040048#ifndef arch_mmap_check
49#define arch_mmap_check(addr, len, flags) (0)
50#endif
51
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080052#ifndef arch_rebalance_pgtables
53#define arch_rebalance_pgtables(addr, len) (addr)
54#endif
55
Hugh Dickinse0da3822005-04-19 13:29:15 -070056static void unmap_region(struct mm_struct *mm,
57 struct vm_area_struct *vma, struct vm_area_struct *prev,
58 unsigned long start, unsigned long end);
59
Linus Torvalds1da177e2005-04-16 15:20:36 -070060/* description of effects of mapping type and prot in current implementation.
61 * this is due to the limited x86 page protection hardware. The expected
62 * behavior is in parens:
63 *
64 * map_type prot
65 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
66 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
67 * w: (no) no w: (no) no w: (yes) yes w: (no) no
68 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
69 *
70 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (copy) copy w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
73 *
74 */
75pgprot_t protection_map[16] = {
76 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
77 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
78};
79
Hugh Dickins804af2c2006-07-26 21:39:49 +010080pgprot_t vm_get_page_prot(unsigned long vm_flags)
81{
Dave Kleikampb845f312008-07-08 00:28:51 +100082 return __pgprot(pgprot_val(protection_map[vm_flags &
83 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
84 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010085}
86EXPORT_SYMBOL(vm_get_page_prot);
87
Shaohua Li34679d72011-05-24 17:11:18 -070088int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
89int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -080090unsigned long sysctl_overcommit_kbytes __read_mostly;
Christoph Lameterc3d8c142005-09-06 15:16:33 -070091int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070092unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070093unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -070094/*
95 * Make sure vm_committed_as in one cacheline and not cacheline shared with
96 * other variables. It can be updated by several CPUs frequently.
97 */
98struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -0800101 * The global memory commitment made in the system can be a metric
102 * that can be used to drive ballooning decisions when Linux is hosted
103 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
104 * balancing memory across competing virtual machines that are hosted.
105 * Several metrics drive this policy engine including the guest reported
106 * memory commitment.
107 */
108unsigned long vm_memory_committed(void)
109{
110 return percpu_counter_read_positive(&vm_committed_as);
111}
112EXPORT_SYMBOL_GPL(vm_memory_committed);
113
114/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 * Check that a process has enough memory to allocate a new virtual
116 * mapping. 0 means there is enough memory for the allocation to
117 * succeed and -ENOMEM implies there is not.
118 *
119 * We currently support three overcommit policies, which are set via the
120 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
121 *
122 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
123 * Additional code 2002 Jul 20 by Robert Love.
124 *
125 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
126 *
127 * Note this is a helper function intended to be used by LSMs which
128 * wish to use this logic.
129 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700130int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700132 unsigned long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
134 vm_acct_memory(pages);
135
136 /*
137 * Sometimes we want to use more memory than we have
138 */
139 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
140 return 0;
141
142 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700143 free = global_page_state(NR_FREE_PAGES);
144 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Dmitry Finkc15bef32011-07-25 17:12:19 -0700146 /*
147 * shmem pages shouldn't be counted as free in this
148 * case, they can't be purged, only swapped out, and
149 * that won't affect the overall amount of available
150 * memory in the system.
151 */
152 free -= global_page_state(NR_SHMEM);
153
Shaohua Liec8acf22013-02-22 16:34:38 -0800154 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
156 /*
157 * Any slabs which are created with the
158 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
159 * which are reclaimable, under pressure. The dentry
160 * cache and most inode caches should fall into this
161 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700162 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700165 * Leave reserved pages. The pages are not for anonymous pages.
166 */
167 if (free <= totalreserve_pages)
168 goto error;
169 else
170 free -= totalreserve_pages;
171
172 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700173 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 */
175 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700176 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178 if (free > pages)
179 return 0;
180
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700181 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 }
183
Jerome Marchand00619bc2013-11-12 15:08:31 -0800184 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700186 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 */
188 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700189 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700191 /*
192 * Don't let a single process grow so big a user can't recover
193 */
194 if (mm) {
195 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
196 allowed -= min(mm->total_vm / 32, reserve);
197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700199 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700201error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 vm_unacct_memory(pages);
203
204 return -ENOMEM;
205}
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700208 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 */
210static void __remove_shared_vm_struct(struct vm_area_struct *vma,
211 struct file *file, struct address_space *mapping)
212{
213 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500214 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (vma->vm_flags & VM_SHARED)
216 mapping->i_mmap_writable--;
217
218 flush_dcache_mmap_lock(mapping);
219 if (unlikely(vma->vm_flags & VM_NONLINEAR))
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700220 list_del_init(&vma->shared.nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700222 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 flush_dcache_mmap_unlock(mapping);
224}
225
226/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700227 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700228 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700230void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 struct file *file = vma->vm_file;
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 if (file) {
235 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700236 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700238 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700240}
241
242/*
243 * Close a vm structure and free it, returning the next.
244 */
245static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
246{
247 struct vm_area_struct *next = vma->vm_next;
248
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700249 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 if (vma->vm_ops && vma->vm_ops->close)
251 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700252 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700253 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700254 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700256 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700259static unsigned long do_brk(unsigned long addr, unsigned long len);
260
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100261SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
263 unsigned long rlim, retval;
264 unsigned long newbrk, oldbrk;
265 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700266 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800267 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269 down_write(&mm->mmap_sem);
270
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700271#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800272 /*
273 * CONFIG_COMPAT_BRK can still be overridden by setting
274 * randomize_va_space to 2, which will still cause mm->start_brk
275 * to be arbitrarily shifted
276 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700277 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800278 min_brk = mm->start_brk;
279 else
280 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700281#else
282 min_brk = mm->start_brk;
283#endif
284 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700286
287 /*
288 * Check against rlimit here. If this check is done later after the test
289 * of oldbrk with newbrk then it can escape the test and let the data
290 * segment grow beyond its set limit the in case where the limit is
291 * not page aligned -Ram Gupta
292 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800293 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100294 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
295 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700296 goto out;
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 newbrk = PAGE_ALIGN(brk);
299 oldbrk = PAGE_ALIGN(mm->brk);
300 if (oldbrk == newbrk)
301 goto set_brk;
302
303 /* Always allow shrinking brk. */
304 if (brk <= mm->brk) {
305 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
306 goto set_brk;
307 goto out;
308 }
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 /* Check against existing mmap mappings. */
311 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
312 goto out;
313
314 /* Ok, looks good - let it rip. */
315 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
316 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318set_brk:
319 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800320 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
321 up_write(&mm->mmap_sem);
322 if (populate)
323 mm_populate(oldbrk, newbrk - oldbrk);
324 return brk;
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326out:
327 retval = mm->brk;
328 up_write(&mm->mmap_sem);
329 return retval;
330}
331
Michel Lespinassed3737182012-12-11 16:01:38 -0800332static long vma_compute_subtree_gap(struct vm_area_struct *vma)
333{
334 unsigned long max, subtree_gap;
335 max = vma->vm_start;
336 if (vma->vm_prev)
337 max -= vma->vm_prev->vm_end;
338 if (vma->vm_rb.rb_left) {
339 subtree_gap = rb_entry(vma->vm_rb.rb_left,
340 struct vm_area_struct, vm_rb)->rb_subtree_gap;
341 if (subtree_gap > max)
342 max = subtree_gap;
343 }
344 if (vma->vm_rb.rb_right) {
345 subtree_gap = rb_entry(vma->vm_rb.rb_right,
346 struct vm_area_struct, vm_rb)->rb_subtree_gap;
347 if (subtree_gap > max)
348 max = subtree_gap;
349 }
350 return max;
351}
352
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700353#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354static int browse_rb(struct rb_root *root)
355{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800356 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 struct rb_node *nd, *pn = NULL;
358 unsigned long prev = 0, pend = 0;
359
360 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
361 struct vm_area_struct *vma;
362 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800363 if (vma->vm_start < prev) {
364 printk("vm_start %lx prev %lx\n", vma->vm_start, prev);
365 bug = 1;
366 }
367 if (vma->vm_start < pend) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800369 bug = 1;
370 }
371 if (vma->vm_start > vma->vm_end) {
372 printk("vm_end %lx < vm_start %lx\n",
373 vma->vm_end, vma->vm_start);
374 bug = 1;
375 }
376 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
377 printk("free gap %lx, correct %lx\n",
378 vma->rb_subtree_gap,
379 vma_compute_subtree_gap(vma));
380 bug = 1;
381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 i++;
383 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800384 prev = vma->vm_start;
385 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 }
387 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800388 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800390 if (i != j) {
391 printk("backwards %d, forwards %d\n", j, i);
392 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800394 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Michel Lespinassed3737182012-12-11 16:01:38 -0800397static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
398{
399 struct rb_node *nd;
400
401 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
402 struct vm_area_struct *vma;
403 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
404 BUG_ON(vma != ignore &&
405 vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407}
408
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700409static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
411 int bug = 0;
412 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800413 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700414 struct vm_area_struct *vma = mm->mmap;
415 while (vma) {
416 struct anon_vma_chain *avc;
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800417 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700418 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
419 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800420 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800421 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700422 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 i++;
424 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800425 if (i != mm->map_count) {
426 printk("map_count %d vm_next %d\n", mm->map_count, i);
427 bug = 1;
428 }
429 if (highest_address != mm->highest_vm_end) {
430 printk("mm->highest_vm_end %lx, found %lx\n",
431 mm->highest_vm_end, highest_address);
432 bug = 1;
433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800435 if (i != mm->map_count) {
436 printk("map_count %d rb %d\n", mm->map_count, i);
437 bug = 1;
438 }
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200439 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
441#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800442#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443#define validate_mm(mm) do { } while (0)
444#endif
445
Michel Lespinassed3737182012-12-11 16:01:38 -0800446RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
447 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
448
449/*
450 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
451 * vma->vm_prev->vm_end values changed, without modifying the vma's position
452 * in the rbtree.
453 */
454static void vma_gap_update(struct vm_area_struct *vma)
455{
456 /*
457 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
458 * function that does exacltly what we want.
459 */
460 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
461}
462
463static inline void vma_rb_insert(struct vm_area_struct *vma,
464 struct rb_root *root)
465{
466 /* All rb_subtree_gap values must be consistent prior to insertion */
467 validate_mm_rb(root, NULL);
468
469 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
470}
471
472static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
473{
474 /*
475 * All rb_subtree_gap values must be consistent prior to erase,
476 * with the possible exception of the vma being erased.
477 */
478 validate_mm_rb(root, vma);
479
480 /*
481 * Note rb_erase_augmented is a fairly large inline function,
482 * so make sure we instantiate it only once with our desired
483 * augmented rbtree callbacks.
484 */
485 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
486}
487
Michel Lespinassebf181b92012-10-08 16:31:39 -0700488/*
489 * vma has some anon_vma assigned, and is already inserted on that
490 * anon_vma's interval trees.
491 *
492 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
493 * vma must be removed from the anon_vma's interval trees using
494 * anon_vma_interval_tree_pre_update_vma().
495 *
496 * After the update, the vma will be reinserted using
497 * anon_vma_interval_tree_post_update_vma().
498 *
499 * The entire update must be protected by exclusive mmap_sem and by
500 * the root anon_vma's mutex.
501 */
502static inline void
503anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
504{
505 struct anon_vma_chain *avc;
506
507 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
508 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
509}
510
511static inline void
512anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
513{
514 struct anon_vma_chain *avc;
515
516 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
517 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
518}
519
Hugh Dickins6597d782012-10-08 16:29:07 -0700520static int find_vma_links(struct mm_struct *mm, unsigned long addr,
521 unsigned long end, struct vm_area_struct **pprev,
522 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
Hugh Dickins6597d782012-10-08 16:29:07 -0700524 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
526 __rb_link = &mm->mm_rb.rb_node;
527 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
529 while (*__rb_link) {
530 struct vm_area_struct *vma_tmp;
531
532 __rb_parent = *__rb_link;
533 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
534
535 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700536 /* Fail if an existing vma overlaps the area */
537 if (vma_tmp->vm_start < end)
538 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 __rb_link = &__rb_parent->rb_left;
540 } else {
541 rb_prev = __rb_parent;
542 __rb_link = &__rb_parent->rb_right;
543 }
544 }
545
546 *pprev = NULL;
547 if (rb_prev)
548 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
549 *rb_link = __rb_link;
550 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700551 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
Cyril Hrubise8420a82013-04-29 15:08:33 -0700554static unsigned long count_vma_pages_range(struct mm_struct *mm,
555 unsigned long addr, unsigned long end)
556{
557 unsigned long nr_pages = 0;
558 struct vm_area_struct *vma;
559
560 /* Find first overlaping mapping */
561 vma = find_vma_intersection(mm, addr, end);
562 if (!vma)
563 return 0;
564
565 nr_pages = (min(end, vma->vm_end) -
566 max(addr, vma->vm_start)) >> PAGE_SHIFT;
567
568 /* Iterate over the rest of the overlaps */
569 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
570 unsigned long overlap_len;
571
572 if (vma->vm_start > end)
573 break;
574
575 overlap_len = min(end, vma->vm_end) - vma->vm_start;
576 nr_pages += overlap_len >> PAGE_SHIFT;
577 }
578
579 return nr_pages;
580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
583 struct rb_node **rb_link, struct rb_node *rb_parent)
584{
Michel Lespinassed3737182012-12-11 16:01:38 -0800585 /* Update tracking information for the gap following the new vma. */
586 if (vma->vm_next)
587 vma_gap_update(vma->vm_next);
588 else
589 mm->highest_vm_end = vma->vm_end;
590
591 /*
592 * vma->vm_prev wasn't known when we followed the rbtree to find the
593 * correct insertion point for that vma. As a result, we could not
594 * update the vma vm_rb parents rb_subtree_gap values on the way down.
595 * So, we first insert the vma with a zero rb_subtree_gap value
596 * (to be consistent with what we did on the way down), and then
597 * immediately update the gap to the correct value. Finally we
598 * rebalance the rbtree after all augmented values have been set.
599 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800601 vma->rb_subtree_gap = 0;
602 vma_gap_update(vma);
603 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700606static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
ZhenwenXu48aae422009-01-06 14:40:21 -0800608 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 file = vma->vm_file;
611 if (file) {
612 struct address_space *mapping = file->f_mapping;
613
614 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500615 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (vma->vm_flags & VM_SHARED)
617 mapping->i_mmap_writable++;
618
619 flush_dcache_mmap_lock(mapping);
620 if (unlikely(vma->vm_flags & VM_NONLINEAR))
621 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
622 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700623 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 flush_dcache_mmap_unlock(mapping);
625 }
626}
627
628static void
629__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
630 struct vm_area_struct *prev, struct rb_node **rb_link,
631 struct rb_node *rb_parent)
632{
633 __vma_link_list(mm, vma, prev, rb_parent);
634 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
637static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
638 struct vm_area_struct *prev, struct rb_node **rb_link,
639 struct rb_node *rb_parent)
640{
641 struct address_space *mapping = NULL;
642
643 if (vma->vm_file)
644 mapping = vma->vm_file->f_mapping;
645
Peter Zijlstra97a89412011-05-24 17:12:04 -0700646 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700647 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649 __vma_link(mm, vma, prev, rb_link, rb_parent);
650 __vma_link_file(vma);
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700653 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
655 mm->map_count++;
656 validate_mm(mm);
657}
658
659/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700660 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700661 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800663static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Hugh Dickins6597d782012-10-08 16:29:07 -0700665 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800666 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Hugh Dickins6597d782012-10-08 16:29:07 -0700668 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
669 &prev, &rb_link, &rb_parent))
670 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 __vma_link(mm, vma, prev, rb_link, rb_parent);
672 mm->map_count++;
673}
674
675static inline void
676__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
677 struct vm_area_struct *prev)
678{
Michel Lespinassed3737182012-12-11 16:01:38 -0800679 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700680
Michel Lespinassed3737182012-12-11 16:01:38 -0800681 vma_rb_erase(vma, &mm->mm_rb);
682 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700683 if (next)
684 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700685
686 /* Kill the cache */
687 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688}
689
690/*
691 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
692 * is already present in an i_mmap tree without adjusting the tree.
693 * The following helper function should be used when such adjustments
694 * are necessary. The "insert" vma (if any) is to be inserted
695 * before we drop the necessary locks.
696 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800697int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
699{
700 struct mm_struct *mm = vma->vm_mm;
701 struct vm_area_struct *next = vma->vm_next;
702 struct vm_area_struct *importer = NULL;
703 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700704 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700705 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800707 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 long adjust_next = 0;
709 int remove_next = 0;
710
711 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700712 struct vm_area_struct *exporter = NULL;
713
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 if (end >= next->vm_end) {
715 /*
716 * vma expands, overlapping all the next, and
717 * perhaps the one after too (mprotect case 6).
718 */
719again: remove_next = 1 + (end > next->vm_end);
720 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700721 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 importer = vma;
723 } else if (end > next->vm_start) {
724 /*
725 * vma expands, overlapping part of the next:
726 * mprotect case 5 shifting the boundary up.
727 */
728 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700729 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 importer = vma;
731 } else if (end < vma->vm_end) {
732 /*
733 * vma shrinks, and !insert tells it's not
734 * split_vma inserting another: so it must be
735 * mprotect case 4 shifting the boundary down.
736 */
737 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700738 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 importer = next;
740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Rik van Riel5beb4932010-03-05 13:42:07 -0800742 /*
743 * Easily overlooked: when mprotect shifts the boundary,
744 * make sure the expanding vma has anon_vma set if the
745 * shrinking vma had, to cover any anon pages imported.
746 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700747 if (exporter && exporter->anon_vma && !importer->anon_vma) {
748 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800749 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700750 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800751 }
752 }
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 if (file) {
755 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530756 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530758 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530759
760 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530761 uprobe_munmap(next, next->vm_start,
762 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530763 }
764
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700765 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700768 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 * are visible to arm/parisc __flush_dcache_page
770 * throughout; but we cannot insert into address
771 * space until vma start or end is updated.
772 */
773 __vma_link_file(insert);
774 }
775 }
776
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800777 vma_adjust_trans_huge(vma, start, end, adjust_next);
778
Michel Lespinassebf181b92012-10-08 16:31:39 -0700779 anon_vma = vma->anon_vma;
780 if (!anon_vma && adjust_next)
781 anon_vma = next->anon_vma;
782 if (anon_vma) {
Michel Lespinasseca42b262012-10-08 16:30:01 -0700783 VM_BUG_ON(adjust_next && next->anon_vma &&
784 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000785 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700786 anon_vma_interval_tree_pre_update_vma(vma);
787 if (adjust_next)
788 anon_vma_interval_tree_pre_update_vma(next);
789 }
Rik van Riel012f18002010-08-09 17:18:40 -0700790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if (root) {
792 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700793 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700795 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
797
Michel Lespinassed3737182012-12-11 16:01:38 -0800798 if (start != vma->vm_start) {
799 vma->vm_start = start;
800 start_changed = true;
801 }
802 if (end != vma->vm_end) {
803 vma->vm_end = end;
804 end_changed = true;
805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 vma->vm_pgoff = pgoff;
807 if (adjust_next) {
808 next->vm_start += adjust_next << PAGE_SHIFT;
809 next->vm_pgoff += adjust_next;
810 }
811
812 if (root) {
813 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700814 vma_interval_tree_insert(next, root);
815 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 flush_dcache_mmap_unlock(mapping);
817 }
818
819 if (remove_next) {
820 /*
821 * vma_merge has merged next into vma, and needs
822 * us to remove next before dropping the locks.
823 */
824 __vma_unlink(mm, next, vma);
825 if (file)
826 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 } else if (insert) {
828 /*
829 * split_vma has split insert from vma, and needs
830 * us to insert it before dropping the locks
831 * (it may either follow vma or precede it).
832 */
833 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800834 } else {
835 if (start_changed)
836 vma_gap_update(vma);
837 if (end_changed) {
838 if (!next)
839 mm->highest_vm_end = end;
840 else if (!adjust_next)
841 vma_gap_update(next);
842 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 }
844
Michel Lespinassebf181b92012-10-08 16:31:39 -0700845 if (anon_vma) {
846 anon_vma_interval_tree_post_update_vma(vma);
847 if (adjust_next)
848 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800849 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700852 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530854 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100855 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530856
857 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100858 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530859 }
860
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700862 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530863 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700865 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800866 if (next->anon_vma)
867 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700869 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 kmem_cache_free(vm_area_cachep, next);
871 /*
872 * In mprotect's case 6 (see comments on vma_merge),
873 * we must remove another next too. It would clutter
874 * up the code too much to do both in one go.
875 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800876 next = vma->vm_next;
877 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800879 else if (next)
880 vma_gap_update(next);
881 else
882 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530884 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100885 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800888
889 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890}
891
892/*
893 * If the vma has a ->close operation then the driver probably needs to release
894 * per-vma resources, so we don't attempt to merge those.
895 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896static inline int is_mergeable_vma(struct vm_area_struct *vma,
897 struct file *file, unsigned long vm_flags)
898{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800899 /*
900 * VM_SOFTDIRTY should not prevent from VMA merging, if we
901 * match the flags but dirty bit -- the caller should mark
902 * merged VMA as dirty. If dirty bit won't be excluded from
903 * comparison, we increase pressue on the memory system forcing
904 * the kernel to generate new VMAs when old one could be
905 * extended instead.
906 */
907 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return 0;
909 if (vma->vm_file != file)
910 return 0;
911 if (vma->vm_ops && vma->vm_ops->close)
912 return 0;
913 return 1;
914}
915
916static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700917 struct anon_vma *anon_vma2,
918 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919{
Shaohua Li965f55d2011-05-24 17:11:20 -0700920 /*
921 * The list_is_singular() test is to avoid merging VMA cloned from
922 * parents. This can improve scalability caused by anon_vma lock.
923 */
924 if ((!anon_vma1 || !anon_vma2) && (!vma ||
925 list_is_singular(&vma->anon_vma_chain)))
926 return 1;
927 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
930/*
931 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
932 * in front of (at a lower virtual address and file offset than) the vma.
933 *
934 * We cannot merge two vmas if they have differently assigned (non-NULL)
935 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
936 *
937 * We don't check here for the merged mmap wrapping around the end of pagecache
938 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
939 * wrap, nor mmaps which cover the final page at index -1UL.
940 */
941static int
942can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
943 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
944{
945 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700946 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (vma->vm_pgoff == vm_pgoff)
948 return 1;
949 }
950 return 0;
951}
952
953/*
954 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
955 * beyond (at a higher virtual address and file offset than) the vma.
956 *
957 * We cannot merge two vmas if they have differently assigned (non-NULL)
958 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
959 */
960static int
961can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
962 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
963{
964 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700965 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700967 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
969 return 1;
970 }
971 return 0;
972}
973
974/*
975 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
976 * whether that can be merged with its predecessor or its successor.
977 * Or both (it neatly fills a hole).
978 *
979 * In most cases - when called for mmap, brk or mremap - [addr,end) is
980 * certain not to be mapped by the time vma_merge is called; but when
981 * called for mprotect, it is certain to be already mapped (either at
982 * an offset within prev, or at the start of next), and the flags of
983 * this area are about to be changed to vm_flags - and the no-change
984 * case has already been eliminated.
985 *
986 * The following mprotect cases have to be considered, where AAAA is
987 * the area passed down from mprotect_fixup, never extending beyond one
988 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
989 *
990 * AAAA AAAA AAAA AAAA
991 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
992 * cannot merge might become might become might become
993 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
994 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
995 * mremap move: PPPPNNNNNNNN 8
996 * AAAA
997 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
998 * might become case 1 below case 2 below case 3 below
999 *
1000 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1001 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1002 */
1003struct vm_area_struct *vma_merge(struct mm_struct *mm,
1004 struct vm_area_struct *prev, unsigned long addr,
1005 unsigned long end, unsigned long vm_flags,
1006 struct anon_vma *anon_vma, struct file *file,
1007 pgoff_t pgoff, struct mempolicy *policy)
1008{
1009 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1010 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001011 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
1013 /*
1014 * We later require that vma->vm_flags == vm_flags,
1015 * so this tests vma->vm_flags & VM_SPECIAL, too.
1016 */
1017 if (vm_flags & VM_SPECIAL)
1018 return NULL;
1019
1020 if (prev)
1021 next = prev->vm_next;
1022 else
1023 next = mm->mmap;
1024 area = next;
1025 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1026 next = next->vm_next;
1027
1028 /*
1029 * Can it merge with the predecessor?
1030 */
1031 if (prev && prev->vm_end == addr &&
1032 mpol_equal(vma_policy(prev), policy) &&
1033 can_vma_merge_after(prev, vm_flags,
1034 anon_vma, file, pgoff)) {
1035 /*
1036 * OK, it can. Can we now merge in the successor as well?
1037 */
1038 if (next && end == next->vm_start &&
1039 mpol_equal(policy, vma_policy(next)) &&
1040 can_vma_merge_before(next, vm_flags,
1041 anon_vma, file, pgoff+pglen) &&
1042 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001043 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001045 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 next->vm_end, prev->vm_pgoff, NULL);
1047 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001048 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001050 if (err)
1051 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001052 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 return prev;
1054 }
1055
1056 /*
1057 * Can this new request be merged in front of next?
1058 */
1059 if (next && end == next->vm_start &&
1060 mpol_equal(policy, vma_policy(next)) &&
1061 can_vma_merge_before(next, vm_flags,
1062 anon_vma, file, pgoff+pglen)) {
1063 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001064 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 addr, prev->vm_pgoff, NULL);
1066 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001067 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001069 if (err)
1070 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001071 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 return area;
1073 }
1074
1075 return NULL;
1076}
1077
1078/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001079 * Rough compatbility check to quickly see if it's even worth looking
1080 * at sharing an anon_vma.
1081 *
1082 * They need to have the same vm_file, and the flags can only differ
1083 * in things that mprotect may change.
1084 *
1085 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1086 * we can merge the two vma's. For example, we refuse to merge a vma if
1087 * there is a vm_ops->close() function, because that indicates that the
1088 * driver is doing some kind of reference counting. But that doesn't
1089 * really matter for the anon_vma sharing case.
1090 */
1091static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1092{
1093 return a->vm_end == b->vm_start &&
1094 mpol_equal(vma_policy(a), vma_policy(b)) &&
1095 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001096 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001097 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1098}
1099
1100/*
1101 * Do some basic sanity checking to see if we can re-use the anon_vma
1102 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1103 * the same as 'old', the other will be the new one that is trying
1104 * to share the anon_vma.
1105 *
1106 * NOTE! This runs with mm_sem held for reading, so it is possible that
1107 * the anon_vma of 'old' is concurrently in the process of being set up
1108 * by another page fault trying to merge _that_. But that's ok: if it
1109 * is being set up, that automatically means that it will be a singleton
1110 * acceptable for merging, so we can do all of this optimistically. But
1111 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1112 *
1113 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1114 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1115 * is to return an anon_vma that is "complex" due to having gone through
1116 * a fork).
1117 *
1118 * We also make sure that the two vma's are compatible (adjacent,
1119 * and with the same memory policies). That's all stable, even with just
1120 * a read lock on the mm_sem.
1121 */
1122static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1123{
1124 if (anon_vma_compatible(a, b)) {
1125 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1126
1127 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1128 return anon_vma;
1129 }
1130 return NULL;
1131}
1132
1133/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1135 * neighbouring vmas for a suitable anon_vma, before it goes off
1136 * to allocate a new anon_vma. It checks because a repetitive
1137 * sequence of mprotects and faults may otherwise lead to distinct
1138 * anon_vmas being allocated, preventing vma merge in subsequent
1139 * mprotect.
1140 */
1141struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1142{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001143 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 near = vma->vm_next;
1147 if (!near)
1148 goto try_prev;
1149
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001150 anon_vma = reusable_anon_vma(near, vma, near);
1151 if (anon_vma)
1152 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001154 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 if (!near)
1156 goto none;
1157
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001158 anon_vma = reusable_anon_vma(near, near, vma);
1159 if (anon_vma)
1160 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161none:
1162 /*
1163 * There's no absolute need to look only at touching neighbours:
1164 * we could search further afield for "compatible" anon_vmas.
1165 * But it would probably just be a waste of time searching,
1166 * or lead to too many vmas hanging off the same anon_vma.
1167 * We're trying to allow mprotect remerging later on,
1168 * not trying to minimize memory used for anon_vmas.
1169 */
1170 return NULL;
1171}
1172
1173#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001174void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 struct file *file, long pages)
1176{
1177 const unsigned long stack_flags
1178 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1179
Huang Shijie44de9d02012-07-31 16:41:49 -07001180 mm->total_vm += pages;
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (file) {
1183 mm->shared_vm += pages;
1184 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1185 mm->exec_vm += pages;
1186 } else if (flags & stack_flags)
1187 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189#endif /* CONFIG_PROC_FS */
1190
1191/*
Al Viro40401532012-02-13 03:58:52 +00001192 * If a hint addr is less than mmap_min_addr change hint to be as
1193 * low as possible but still greater than mmap_min_addr
1194 */
1195static inline unsigned long round_hint_to_min(unsigned long hint)
1196{
1197 hint &= PAGE_MASK;
1198 if (((void *)hint != NULL) &&
1199 (hint < mmap_min_addr))
1200 return PAGE_ALIGN(mmap_min_addr);
1201 return hint;
1202}
1203
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001204static inline int mlock_future_check(struct mm_struct *mm,
1205 unsigned long flags,
1206 unsigned long len)
1207{
1208 unsigned long locked, lock_limit;
1209
1210 /* mlock MCL_FUTURE? */
1211 if (flags & VM_LOCKED) {
1212 locked = len >> PAGE_SHIFT;
1213 locked += mm->locked_vm;
1214 lock_limit = rlimit(RLIMIT_MEMLOCK);
1215 lock_limit >>= PAGE_SHIFT;
1216 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1217 return -EAGAIN;
1218 }
1219 return 0;
1220}
1221
Al Viro40401532012-02-13 03:58:52 +00001222/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001223 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 */
1225
Al Viroe3fc6292012-05-30 20:08:42 -04001226unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001228 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001229 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
1231 struct mm_struct * mm = current->mm;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001232 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Michel Lespinasse41badc12013-02-22 16:32:47 -08001234 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 /*
1237 * Does the application expect PROT_READ to imply PROT_EXEC?
1238 *
1239 * (the exception is when the underlying filesystem is noexec
1240 * mounted, in which case we dont add PROT_EXEC.)
1241 */
1242 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001243 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 prot |= PROT_EXEC;
1245
1246 if (!len)
1247 return -EINVAL;
1248
Eric Paris7cd94142007-11-26 18:47:40 -05001249 if (!(flags & MAP_FIXED))
1250 addr = round_hint_to_min(addr);
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 /* Careful about overflows.. */
1253 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001254 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 return -ENOMEM;
1256
1257 /* offset overflow? */
1258 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1259 return -EOVERFLOW;
1260
1261 /* Too many mappings? */
1262 if (mm->map_count > sysctl_max_map_count)
1263 return -ENOMEM;
1264
1265 /* Obtain the address to map to. we verify (or select) it and ensure
1266 * that it represents a valid section of the address space.
1267 */
1268 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1269 if (addr & ~PAGE_MASK)
1270 return addr;
1271
1272 /* Do simple checking here so the lower-level routines won't have
1273 * to. we assume access permissions have been handled by the open
1274 * of the memory object, so we don't do any here.
1275 */
1276 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1277 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1278
Huang Shijiecdf7b342009-09-21 17:03:36 -07001279 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 if (!can_do_mlock())
1281 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001282
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001283 if (mlock_future_check(mm, vm_flags, len))
1284 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001287 struct inode *inode = file_inode(file);
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 switch (flags & MAP_TYPE) {
1290 case MAP_SHARED:
1291 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1292 return -EACCES;
1293
1294 /*
1295 * Make sure we don't allow writing to an append-only
1296 * file..
1297 */
1298 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1299 return -EACCES;
1300
1301 /*
1302 * Make sure there are no mandatory locks on the file.
1303 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001304 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 return -EAGAIN;
1306
1307 vm_flags |= VM_SHARED | VM_MAYSHARE;
1308 if (!(file->f_mode & FMODE_WRITE))
1309 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1310
1311 /* fall through */
1312 case MAP_PRIVATE:
1313 if (!(file->f_mode & FMODE_READ))
1314 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001315 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001316 if (vm_flags & VM_EXEC)
1317 return -EPERM;
1318 vm_flags &= ~VM_MAYEXEC;
1319 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001320
Al Viro72c2d532013-09-22 16:27:52 -04001321 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001322 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001323 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1324 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 break;
1326
1327 default:
1328 return -EINVAL;
1329 }
1330 } else {
1331 switch (flags & MAP_TYPE) {
1332 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001333 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1334 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001335 /*
1336 * Ignore pgoff.
1337 */
1338 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 vm_flags |= VM_SHARED | VM_MAYSHARE;
1340 break;
1341 case MAP_PRIVATE:
1342 /*
1343 * Set pgoff according to addr for anon_vma.
1344 */
1345 pgoff = addr >> PAGE_SHIFT;
1346 break;
1347 default:
1348 return -EINVAL;
1349 }
1350 }
1351
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001352 /*
1353 * Set 'VM_NORESERVE' if we should not account for the
1354 * memory use of this mapping.
1355 */
1356 if (flags & MAP_NORESERVE) {
1357 /* We honor MAP_NORESERVE if allowed to overcommit */
1358 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1359 vm_flags |= VM_NORESERVE;
1360
1361 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1362 if (file && is_file_hugepages(file))
1363 vm_flags |= VM_NORESERVE;
1364 }
1365
1366 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001367 if (!IS_ERR_VALUE(addr) &&
1368 ((vm_flags & VM_LOCKED) ||
1369 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001370 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001371 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001372}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001373
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001374SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1375 unsigned long, prot, unsigned long, flags,
1376 unsigned long, fd, unsigned long, pgoff)
1377{
1378 struct file *file = NULL;
1379 unsigned long retval = -EBADF;
1380
1381 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001382 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001383 file = fget(fd);
1384 if (!file)
1385 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001386 if (is_file_hugepages(file))
1387 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001388 retval = -EINVAL;
1389 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1390 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001391 } else if (flags & MAP_HUGETLB) {
1392 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001393 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001394
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001395 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001396 if (!hs)
1397 return -EINVAL;
1398
1399 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001400 /*
1401 * VM_NORESERVE is used because the reservations will be
1402 * taken when vm_ops->mmap() is called
1403 * A dummy user value is used because we are not locking
1404 * memory so no accounting is necessary
1405 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001406 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001407 VM_NORESERVE,
1408 &user, HUGETLB_ANONHUGE_INODE,
1409 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001410 if (IS_ERR(file))
1411 return PTR_ERR(file);
1412 }
1413
1414 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1415
Al Viroeb36c582012-05-30 20:17:35 -04001416 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001417out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001418 if (file)
1419 fput(file);
1420out:
1421 return retval;
1422}
1423
Christoph Hellwiga4679372010-03-10 15:21:15 -08001424#ifdef __ARCH_WANT_SYS_OLD_MMAP
1425struct mmap_arg_struct {
1426 unsigned long addr;
1427 unsigned long len;
1428 unsigned long prot;
1429 unsigned long flags;
1430 unsigned long fd;
1431 unsigned long offset;
1432};
1433
1434SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1435{
1436 struct mmap_arg_struct a;
1437
1438 if (copy_from_user(&a, arg, sizeof(a)))
1439 return -EFAULT;
1440 if (a.offset & ~PAGE_MASK)
1441 return -EINVAL;
1442
1443 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1444 a.offset >> PAGE_SHIFT);
1445}
1446#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1447
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001448/*
1449 * Some shared mappigns will want the pages marked read-only
1450 * to track write events. If so, we'll downgrade vm_page_prot
1451 * to the private version (using protection_map[] without the
1452 * VM_SHARED bit).
1453 */
1454int vma_wants_writenotify(struct vm_area_struct *vma)
1455{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001456 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001457
1458 /* If it was private or non-writable, the write bit is already clear */
1459 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1460 return 0;
1461
1462 /* The backer wishes to know when pages are first written to? */
1463 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1464 return 1;
1465
1466 /* The open routine did something to the protections already? */
1467 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001468 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001469 return 0;
1470
1471 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001472 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001473 return 0;
1474
1475 /* Can the mapping track the dirty pages? */
1476 return vma->vm_file && vma->vm_file->f_mapping &&
1477 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1478}
1479
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001480/*
1481 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001482 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001483 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001484static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001485{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001486 /*
1487 * hugetlb has its own accounting separate from the core VM
1488 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1489 */
1490 if (file && is_file_hugepages(file))
1491 return 0;
1492
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001493 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1494}
1495
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001496unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001497 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001498{
1499 struct mm_struct *mm = current->mm;
1500 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001501 int error;
1502 struct rb_node **rb_link, *rb_parent;
1503 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001504
Cyril Hrubise8420a82013-04-29 15:08:33 -07001505 /* Check against address space limit. */
1506 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1507 unsigned long nr_pages;
1508
1509 /*
1510 * MAP_FIXED may remove pages of mappings that intersects with
1511 * requested mapping. Account for the pages it would unmap.
1512 */
1513 if (!(vm_flags & MAP_FIXED))
1514 return -ENOMEM;
1515
1516 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1517
1518 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1519 return -ENOMEM;
1520 }
1521
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 /* Clear old maps */
1523 error = -ENOMEM;
1524munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001525 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 if (do_munmap(mm, addr, len))
1527 return -ENOMEM;
1528 goto munmap_back;
1529 }
1530
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001531 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001532 * Private writable mapping: check memory availability
1533 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001534 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001535 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001536 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001537 return -ENOMEM;
1538 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 }
1540
1541 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001542 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001544 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1545 if (vma)
1546 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 /*
1549 * Determine the object being mapped and call the appropriate
1550 * specific mapper. the address has already been validated, but
1551 * not unmapped, but the maps are removed from the list.
1552 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001553 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 if (!vma) {
1555 error = -ENOMEM;
1556 goto unacct_error;
1557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
1559 vma->vm_mm = mm;
1560 vma->vm_start = addr;
1561 vma->vm_end = addr + len;
1562 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001563 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001565 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
1567 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 if (vm_flags & VM_DENYWRITE) {
1569 error = deny_write_access(file);
1570 if (error)
1571 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 }
Al Virocb0942b2012-08-27 14:48:26 -04001573 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 error = file->f_op->mmap(file, vma);
1575 if (error)
1576 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001577
1578 /* Can addr have changed??
1579 *
1580 * Answer: Yes, several device drivers can do it in their
1581 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001582 * Bug: If addr is changed, prev, rb_link, rb_parent should
1583 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001584 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001585 WARN_ON_ONCE(addr != vma->vm_start);
1586
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001587 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001588 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 } else if (vm_flags & VM_SHARED) {
1590 error = shmem_zero_setup(vma);
1591 if (error)
1592 goto free_vma;
1593 }
1594
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001595 if (vma_wants_writenotify(vma)) {
1596 pgprot_t pprot = vma->vm_page_prot;
1597
1598 /* Can vma->vm_page_prot have changed??
1599 *
1600 * Answer: Yes, drivers may have changed it in their
1601 * f_op->mmap method.
1602 *
1603 * Ensures that vmas marked as uncached stay that way.
1604 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001605 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001606 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1607 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1608 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001609
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001610 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001611 /* Once vma denies write, undo our temporary denial count */
Oleg Nesterove8686772013-09-11 14:20:20 -07001612 if (vm_flags & VM_DENYWRITE)
1613 allow_write_access(file);
1614 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001615out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001616 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001617
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001618 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001620 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1621 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001622 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001623 else
1624 vma->vm_flags &= ~VM_LOCKED;
1625 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301626
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001627 if (file)
1628 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301629
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001630 /*
1631 * New (or expanded) vma always get soft dirty status.
1632 * Otherwise user-space soft-dirty page tracker won't
1633 * be able to distinguish situation when vma area unmapped,
1634 * then new mapped in-place (which must be aimed as
1635 * a completely new data area).
1636 */
1637 vma->vm_flags |= VM_SOFTDIRTY;
1638
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 return addr;
1640
1641unmap_and_free_vma:
Oleg Nesterove8686772013-09-11 14:20:20 -07001642 if (vm_flags & VM_DENYWRITE)
1643 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 vma->vm_file = NULL;
1645 fput(file);
1646
1647 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001648 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1649 charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650free_vma:
1651 kmem_cache_free(vm_area_cachep, vma);
1652unacct_error:
1653 if (charged)
1654 vm_unacct_memory(charged);
1655 return error;
1656}
1657
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001658unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1659{
1660 /*
1661 * We implement the search by looking for an rbtree node that
1662 * immediately follows a suitable gap. That is,
1663 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1664 * - gap_end = vma->vm_start >= info->low_limit + length;
1665 * - gap_end - gap_start >= length
1666 */
1667
1668 struct mm_struct *mm = current->mm;
1669 struct vm_area_struct *vma;
1670 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1671
1672 /* Adjust search length to account for worst case alignment overhead */
1673 length = info->length + info->align_mask;
1674 if (length < info->length)
1675 return -ENOMEM;
1676
1677 /* Adjust search limits by the desired length */
1678 if (info->high_limit < length)
1679 return -ENOMEM;
1680 high_limit = info->high_limit - length;
1681
1682 if (info->low_limit > high_limit)
1683 return -ENOMEM;
1684 low_limit = info->low_limit + length;
1685
1686 /* Check if rbtree root looks promising */
1687 if (RB_EMPTY_ROOT(&mm->mm_rb))
1688 goto check_highest;
1689 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1690 if (vma->rb_subtree_gap < length)
1691 goto check_highest;
1692
1693 while (true) {
1694 /* Visit left subtree if it looks promising */
1695 gap_end = vma->vm_start;
1696 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1697 struct vm_area_struct *left =
1698 rb_entry(vma->vm_rb.rb_left,
1699 struct vm_area_struct, vm_rb);
1700 if (left->rb_subtree_gap >= length) {
1701 vma = left;
1702 continue;
1703 }
1704 }
1705
1706 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1707check_current:
1708 /* Check if current node has a suitable gap */
1709 if (gap_start > high_limit)
1710 return -ENOMEM;
1711 if (gap_end >= low_limit && gap_end - gap_start >= length)
1712 goto found;
1713
1714 /* Visit right subtree if it looks promising */
1715 if (vma->vm_rb.rb_right) {
1716 struct vm_area_struct *right =
1717 rb_entry(vma->vm_rb.rb_right,
1718 struct vm_area_struct, vm_rb);
1719 if (right->rb_subtree_gap >= length) {
1720 vma = right;
1721 continue;
1722 }
1723 }
1724
1725 /* Go back up the rbtree to find next candidate node */
1726 while (true) {
1727 struct rb_node *prev = &vma->vm_rb;
1728 if (!rb_parent(prev))
1729 goto check_highest;
1730 vma = rb_entry(rb_parent(prev),
1731 struct vm_area_struct, vm_rb);
1732 if (prev == vma->vm_rb.rb_left) {
1733 gap_start = vma->vm_prev->vm_end;
1734 gap_end = vma->vm_start;
1735 goto check_current;
1736 }
1737 }
1738 }
1739
1740check_highest:
1741 /* Check highest gap, which does not precede any rbtree node */
1742 gap_start = mm->highest_vm_end;
1743 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1744 if (gap_start > high_limit)
1745 return -ENOMEM;
1746
1747found:
1748 /* We found a suitable gap. Clip it with the original low_limit. */
1749 if (gap_start < info->low_limit)
1750 gap_start = info->low_limit;
1751
1752 /* Adjust gap address to the desired alignment */
1753 gap_start += (info->align_offset - gap_start) & info->align_mask;
1754
1755 VM_BUG_ON(gap_start + info->length > info->high_limit);
1756 VM_BUG_ON(gap_start + info->length > gap_end);
1757 return gap_start;
1758}
1759
1760unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1761{
1762 struct mm_struct *mm = current->mm;
1763 struct vm_area_struct *vma;
1764 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1765
1766 /* Adjust search length to account for worst case alignment overhead */
1767 length = info->length + info->align_mask;
1768 if (length < info->length)
1769 return -ENOMEM;
1770
1771 /*
1772 * Adjust search limits by the desired length.
1773 * See implementation comment at top of unmapped_area().
1774 */
1775 gap_end = info->high_limit;
1776 if (gap_end < length)
1777 return -ENOMEM;
1778 high_limit = gap_end - length;
1779
1780 if (info->low_limit > high_limit)
1781 return -ENOMEM;
1782 low_limit = info->low_limit + length;
1783
1784 /* Check highest gap, which does not precede any rbtree node */
1785 gap_start = mm->highest_vm_end;
1786 if (gap_start <= high_limit)
1787 goto found_highest;
1788
1789 /* Check if rbtree root looks promising */
1790 if (RB_EMPTY_ROOT(&mm->mm_rb))
1791 return -ENOMEM;
1792 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1793 if (vma->rb_subtree_gap < length)
1794 return -ENOMEM;
1795
1796 while (true) {
1797 /* Visit right subtree if it looks promising */
1798 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1799 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1800 struct vm_area_struct *right =
1801 rb_entry(vma->vm_rb.rb_right,
1802 struct vm_area_struct, vm_rb);
1803 if (right->rb_subtree_gap >= length) {
1804 vma = right;
1805 continue;
1806 }
1807 }
1808
1809check_current:
1810 /* Check if current node has a suitable gap */
1811 gap_end = vma->vm_start;
1812 if (gap_end < low_limit)
1813 return -ENOMEM;
1814 if (gap_start <= high_limit && gap_end - gap_start >= length)
1815 goto found;
1816
1817 /* Visit left subtree if it looks promising */
1818 if (vma->vm_rb.rb_left) {
1819 struct vm_area_struct *left =
1820 rb_entry(vma->vm_rb.rb_left,
1821 struct vm_area_struct, vm_rb);
1822 if (left->rb_subtree_gap >= length) {
1823 vma = left;
1824 continue;
1825 }
1826 }
1827
1828 /* Go back up the rbtree to find next candidate node */
1829 while (true) {
1830 struct rb_node *prev = &vma->vm_rb;
1831 if (!rb_parent(prev))
1832 return -ENOMEM;
1833 vma = rb_entry(rb_parent(prev),
1834 struct vm_area_struct, vm_rb);
1835 if (prev == vma->vm_rb.rb_right) {
1836 gap_start = vma->vm_prev ?
1837 vma->vm_prev->vm_end : 0;
1838 goto check_current;
1839 }
1840 }
1841 }
1842
1843found:
1844 /* We found a suitable gap. Clip it with the original high_limit. */
1845 if (gap_end > info->high_limit)
1846 gap_end = info->high_limit;
1847
1848found_highest:
1849 /* Compute highest gap address at the desired alignment */
1850 gap_end -= info->length;
1851 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1852
1853 VM_BUG_ON(gap_end < info->low_limit);
1854 VM_BUG_ON(gap_end < gap_start);
1855 return gap_end;
1856}
1857
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858/* Get an address range which is currently unmapped.
1859 * For shmat() with addr=0.
1860 *
1861 * Ugly calling convention alert:
1862 * Return value with the low bits set means error value,
1863 * ie
1864 * if (ret & ~PAGE_MASK)
1865 * error = ret;
1866 *
1867 * This function "knows" that -ENOMEM has the bits set.
1868 */
1869#ifndef HAVE_ARCH_UNMAPPED_AREA
1870unsigned long
1871arch_get_unmapped_area(struct file *filp, unsigned long addr,
1872 unsigned long len, unsigned long pgoff, unsigned long flags)
1873{
1874 struct mm_struct *mm = current->mm;
1875 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001876 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001878 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 return -ENOMEM;
1880
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001881 if (flags & MAP_FIXED)
1882 return addr;
1883
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 if (addr) {
1885 addr = PAGE_ALIGN(addr);
1886 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001887 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 (!vma || addr + len <= vma->vm_start))
1889 return addr;
1890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001892 info.flags = 0;
1893 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001894 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001895 info.high_limit = TASK_SIZE;
1896 info.align_mask = 0;
1897 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
1899#endif
1900
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901/*
1902 * This mmap-allocator allocates new areas top-down from below the
1903 * stack's low limit (the base):
1904 */
1905#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1906unsigned long
1907arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1908 const unsigned long len, const unsigned long pgoff,
1909 const unsigned long flags)
1910{
1911 struct vm_area_struct *vma;
1912 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001913 unsigned long addr = addr0;
1914 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001917 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 return -ENOMEM;
1919
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001920 if (flags & MAP_FIXED)
1921 return addr;
1922
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 /* requesting a specific address */
1924 if (addr) {
1925 addr = PAGE_ALIGN(addr);
1926 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001927 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 (!vma || addr + len <= vma->vm_start))
1929 return addr;
1930 }
1931
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001932 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1933 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001934 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001935 info.high_limit = mm->mmap_base;
1936 info.align_mask = 0;
1937 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 /*
1940 * A failed mmap() very likely causes application failure,
1941 * so fall back to the bottom-up function here. This scenario
1942 * can happen with large stack limits and large mmap()
1943 * allocations.
1944 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001945 if (addr & ~PAGE_MASK) {
1946 VM_BUG_ON(addr != -ENOMEM);
1947 info.flags = 0;
1948 info.low_limit = TASK_UNMAPPED_BASE;
1949 info.high_limit = TASK_SIZE;
1950 addr = vm_unmapped_area(&info);
1951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 return addr;
1954}
1955#endif
1956
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957unsigned long
1958get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1959 unsigned long pgoff, unsigned long flags)
1960{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001961 unsigned long (*get_area)(struct file *, unsigned long,
1962 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Al Viro9206de92009-12-03 15:23:11 -05001964 unsigned long error = arch_mmap_check(addr, len, flags);
1965 if (error)
1966 return error;
1967
1968 /* Careful about overflows.. */
1969 if (len > TASK_SIZE)
1970 return -ENOMEM;
1971
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001972 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04001973 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001974 get_area = file->f_op->get_unmapped_area;
1975 addr = get_area(file, addr, len, pgoff, flags);
1976 if (IS_ERR_VALUE(addr))
1977 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001979 if (addr > TASK_SIZE - len)
1980 return -ENOMEM;
1981 if (addr & ~PAGE_MASK)
1982 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001983
Al Viro9ac4ed42012-05-30 17:13:15 -04001984 addr = arch_rebalance_pgtables(addr, len);
1985 error = security_mmap_addr(addr);
1986 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987}
1988
1989EXPORT_SYMBOL(get_unmapped_area);
1990
1991/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001992struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001994 struct rb_node *rb_node;
1995 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001997 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001998 vma = vmacache_find(mm, addr);
1999 if (likely(vma))
2000 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002002 rb_node = mm->mm_rb.rb_node;
2003 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002005 while (rb_node) {
2006 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002008 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002009
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002010 if (tmp->vm_end > addr) {
2011 vma = tmp;
2012 if (tmp->vm_start <= addr)
2013 break;
2014 rb_node = rb_node->rb_left;
2015 } else
2016 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002018
2019 if (vma)
2020 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 return vma;
2022}
2023
2024EXPORT_SYMBOL(find_vma);
2025
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002026/*
2027 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002028 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029struct vm_area_struct *
2030find_vma_prev(struct mm_struct *mm, unsigned long addr,
2031 struct vm_area_struct **pprev)
2032{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002033 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002035 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002036 if (vma) {
2037 *pprev = vma->vm_prev;
2038 } else {
2039 struct rb_node *rb_node = mm->mm_rb.rb_node;
2040 *pprev = NULL;
2041 while (rb_node) {
2042 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2043 rb_node = rb_node->rb_right;
2044 }
2045 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002046 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047}
2048
2049/*
2050 * Verify that the stack growth is acceptable and
2051 * update accounting. This is shared with both the
2052 * grow-up and grow-down cases.
2053 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002054static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055{
2056 struct mm_struct *mm = vma->vm_mm;
2057 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08002058 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
2060 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002061 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 return -ENOMEM;
2063
2064 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08002065 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 return -ENOMEM;
2067
2068 /* mlock limit tests */
2069 if (vma->vm_flags & VM_LOCKED) {
2070 unsigned long locked;
2071 unsigned long limit;
2072 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002073 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2074 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 if (locked > limit && !capable(CAP_IPC_LOCK))
2076 return -ENOMEM;
2077 }
2078
Adam Litke0d59a012007-01-30 14:35:39 -08002079 /* Check to ensure the stack will not grow into a hugetlb-only region */
2080 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2081 vma->vm_end - size;
2082 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2083 return -EFAULT;
2084
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 /*
2086 * Overcommit.. This must be the final test, as it will
2087 * update security statistics.
2088 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002089 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 return -ENOMEM;
2091
2092 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 if (vma->vm_flags & VM_LOCKED)
2094 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002095 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 return 0;
2097}
2098
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002099#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002101 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2102 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002104int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105{
2106 int error;
2107
2108 if (!(vma->vm_flags & VM_GROWSUP))
2109 return -EFAULT;
2110
2111 /*
2112 * We must make sure the anon_vma is allocated
2113 * so that the anon_vma locking is not a noop.
2114 */
2115 if (unlikely(anon_vma_prepare(vma)))
2116 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002117 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 /*
2120 * vma->vm_start/vm_end cannot change under us because the caller
2121 * is required to hold the mmap_sem in read mode. We need the
2122 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002123 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 */
Helge Deller06b32f32006-12-19 19:28:33 +01002125 if (address < PAGE_ALIGN(address+4))
2126 address = PAGE_ALIGN(address+4);
2127 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002128 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002129 return -ENOMEM;
2130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 error = 0;
2132
2133 /* Somebody else might have raced and expanded it already */
2134 if (address > vma->vm_end) {
2135 unsigned long size, grow;
2136
2137 size = address - vma->vm_start;
2138 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2139
Hugh Dickins42c36f62011-05-09 17:44:42 -07002140 error = -ENOMEM;
2141 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2142 error = acct_stack_growth(vma, size, grow);
2143 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002144 /*
2145 * vma_gap_update() doesn't support concurrent
2146 * updates, but we only hold a shared mmap_sem
2147 * lock here, so we need to protect against
2148 * concurrent vma expansions.
2149 * vma_lock_anon_vma() doesn't help here, as
2150 * we don't guarantee that all growable vmas
2151 * in a mm share the same root anon vma.
2152 * So, we reuse mm->page_table_lock to guard
2153 * against concurrent vma expansions.
2154 */
2155 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002156 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002157 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002158 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002159 if (vma->vm_next)
2160 vma_gap_update(vma->vm_next);
2161 else
2162 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002163 spin_unlock(&vma->vm_mm->page_table_lock);
2164
Hugh Dickins42c36f62011-05-09 17:44:42 -07002165 perf_event_mmap(vma);
2166 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002169 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002170 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002171 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 return error;
2173}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002174#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176/*
2177 * vma is the first one with address < vma->vm_start. Have to extend vma.
2178 */
Michal Hockod05f3162011-05-24 17:11:44 -07002179int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002180 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181{
2182 int error;
2183
2184 /*
2185 * We must make sure the anon_vma is allocated
2186 * so that the anon_vma locking is not a noop.
2187 */
2188 if (unlikely(anon_vma_prepare(vma)))
2189 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002190
2191 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002192 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002193 if (error)
2194 return error;
2195
Rik van Rielbb4a3402010-08-09 17:18:37 -07002196 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
2198 /*
2199 * vma->vm_start/vm_end cannot change under us because the caller
2200 * is required to hold the mmap_sem in read mode. We need the
2201 * anon_vma lock to serialize against concurrent expand_stacks.
2202 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
2204 /* Somebody else might have raced and expanded it already */
2205 if (address < vma->vm_start) {
2206 unsigned long size, grow;
2207
2208 size = vma->vm_end - address;
2209 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2210
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002211 error = -ENOMEM;
2212 if (grow <= vma->vm_pgoff) {
2213 error = acct_stack_growth(vma, size, grow);
2214 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002215 /*
2216 * vma_gap_update() doesn't support concurrent
2217 * updates, but we only hold a shared mmap_sem
2218 * lock here, so we need to protect against
2219 * concurrent vma expansions.
2220 * vma_lock_anon_vma() doesn't help here, as
2221 * we don't guarantee that all growable vmas
2222 * in a mm share the same root anon vma.
2223 * So, we reuse mm->page_table_lock to guard
2224 * against concurrent vma expansions.
2225 */
2226 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002227 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002228 vma->vm_start = address;
2229 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002230 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002231 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002232 spin_unlock(&vma->vm_mm->page_table_lock);
2233
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002234 perf_event_mmap(vma);
2235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
2237 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002238 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002239 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002240 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 return error;
2242}
2243
Linus Torvalds09884962013-02-27 08:36:04 -08002244/*
2245 * Note how expand_stack() refuses to expand the stack all the way to
2246 * abut the next virtual mapping, *unless* that mapping itself is also
2247 * a stack mapping. We want to leave room for a guard page, after all
2248 * (the guard page itself is not added here, that is done by the
2249 * actual page faulting logic)
2250 *
2251 * This matches the behavior of the guard page logic (see mm/memory.c:
2252 * check_stack_guard_page()), which only allows the guard page to be
2253 * removed under these circumstances.
2254 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002255#ifdef CONFIG_STACK_GROWSUP
2256int expand_stack(struct vm_area_struct *vma, unsigned long address)
2257{
Linus Torvalds09884962013-02-27 08:36:04 -08002258 struct vm_area_struct *next;
2259
2260 address &= PAGE_MASK;
2261 next = vma->vm_next;
2262 if (next && next->vm_start == address + PAGE_SIZE) {
2263 if (!(next->vm_flags & VM_GROWSUP))
2264 return -ENOMEM;
2265 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002266 return expand_upwards(vma, address);
2267}
2268
2269struct vm_area_struct *
2270find_extend_vma(struct mm_struct *mm, unsigned long addr)
2271{
2272 struct vm_area_struct *vma, *prev;
2273
2274 addr &= PAGE_MASK;
2275 vma = find_vma_prev(mm, addr, &prev);
2276 if (vma && (vma->vm_start <= addr))
2277 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002278 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002279 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002280 if (prev->vm_flags & VM_LOCKED)
2281 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002282 return prev;
2283}
2284#else
2285int expand_stack(struct vm_area_struct *vma, unsigned long address)
2286{
Linus Torvalds09884962013-02-27 08:36:04 -08002287 struct vm_area_struct *prev;
2288
2289 address &= PAGE_MASK;
2290 prev = vma->vm_prev;
2291 if (prev && prev->vm_end == address) {
2292 if (!(prev->vm_flags & VM_GROWSDOWN))
2293 return -ENOMEM;
2294 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002295 return expand_downwards(vma, address);
2296}
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298struct vm_area_struct *
2299find_extend_vma(struct mm_struct * mm, unsigned long addr)
2300{
2301 struct vm_area_struct * vma;
2302 unsigned long start;
2303
2304 addr &= PAGE_MASK;
2305 vma = find_vma(mm,addr);
2306 if (!vma)
2307 return NULL;
2308 if (vma->vm_start <= addr)
2309 return vma;
2310 if (!(vma->vm_flags & VM_GROWSDOWN))
2311 return NULL;
2312 start = vma->vm_start;
2313 if (expand_stack(vma, addr))
2314 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002315 if (vma->vm_flags & VM_LOCKED)
2316 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 return vma;
2318}
2319#endif
2320
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002322 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002324 *
2325 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002327static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002329 unsigned long nr_accounted = 0;
2330
Hugh Dickins365e9c872005-10-29 18:16:18 -07002331 /* Update high watermark before we lower total_vm */
2332 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002334 long nrpages = vma_pages(vma);
2335
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002336 if (vma->vm_flags & VM_ACCOUNT)
2337 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002338 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002339 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002340 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002341 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 validate_mm(mm);
2343}
2344
2345/*
2346 * Get rid of page table information in the indicated region.
2347 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002348 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 */
2350static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002351 struct vm_area_struct *vma, struct vm_area_struct *prev,
2352 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353{
Hugh Dickinse0da3822005-04-19 13:29:15 -07002354 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002355 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
2357 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002358 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002359 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002360 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002361 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002362 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002363 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364}
2365
2366/*
2367 * Create a list of vma's touched by the unmap, removing them from the mm's
2368 * vma list as we go..
2369 */
2370static void
2371detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2372 struct vm_area_struct *prev, unsigned long end)
2373{
2374 struct vm_area_struct **insertion_point;
2375 struct vm_area_struct *tail_vma = NULL;
2376
2377 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002378 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002380 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 mm->map_count--;
2382 tail_vma = vma;
2383 vma = vma->vm_next;
2384 } while (vma && vma->vm_start < end);
2385 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002386 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002387 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002388 vma_gap_update(vma);
2389 } else
2390 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002392
2393 /* Kill the cache */
2394 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395}
2396
2397/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002398 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2399 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002401static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 unsigned long addr, int new_below)
2403{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08002405 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
Andi Kleena5516432008-07-23 21:27:41 -07002407 if (is_vm_hugetlb_page(vma) && (addr &
2408 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 return -EINVAL;
2410
Christoph Lametere94b1762006-12-06 20:33:17 -08002411 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002413 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414
2415 /* most fields are the same, copy all, and then fixup */
2416 *new = *vma;
2417
Rik van Riel5beb4932010-03-05 13:42:07 -08002418 INIT_LIST_HEAD(&new->anon_vma_chain);
2419
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 if (new_below)
2421 new->vm_end = addr;
2422 else {
2423 new->vm_start = addr;
2424 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2425 }
2426
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002427 err = vma_dup_policy(vma, new);
2428 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002429 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Rik van Riel5beb4932010-03-05 13:42:07 -08002431 if (anon_vma_clone(new, vma))
2432 goto out_free_mpol;
2433
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002434 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 get_file(new->vm_file);
2436
2437 if (new->vm_ops && new->vm_ops->open)
2438 new->vm_ops->open(new);
2439
2440 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002441 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2443 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002444 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Rik van Riel5beb4932010-03-05 13:42:07 -08002446 /* Success. */
2447 if (!err)
2448 return 0;
2449
2450 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002451 if (new->vm_ops && new->vm_ops->close)
2452 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002453 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002454 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002455 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002456 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002457 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002458 out_free_vma:
2459 kmem_cache_free(vm_area_cachep, new);
2460 out_err:
2461 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462}
2463
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002464/*
2465 * Split a vma into two pieces at address 'addr', a new vma is allocated
2466 * either for the first part or the tail.
2467 */
2468int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2469 unsigned long addr, int new_below)
2470{
2471 if (mm->map_count >= sysctl_max_map_count)
2472 return -ENOMEM;
2473
2474 return __split_vma(mm, vma, addr, new_below);
2475}
2476
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477/* Munmap is split into 2 main parts -- this part which finds
2478 * what needs doing, and the areas themselves, which do the
2479 * work. This now handles partial unmappings.
2480 * Jeremy Fitzhardinge <jeremy@goop.org>
2481 */
2482int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2483{
2484 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002485 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486
2487 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2488 return -EINVAL;
2489
2490 if ((len = PAGE_ALIGN(len)) == 0)
2491 return -EINVAL;
2492
2493 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002494 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002495 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002497 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002498 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499
2500 /* if it doesn't overlap, we have nothing.. */
2501 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002502 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 return 0;
2504
2505 /*
2506 * If we need to split any vma, do it now to save pain later.
2507 *
2508 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2509 * unmapped vm_area_struct will remain in use: so lower split_vma
2510 * places tmp vma above, and higher split_vma places tmp vma below.
2511 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002512 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002513 int error;
2514
2515 /*
2516 * Make sure that map_count on return from munmap() will
2517 * not exceed its limit; but let map_count go just above
2518 * its limit temporarily, to help free resources as expected.
2519 */
2520 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2521 return -ENOMEM;
2522
2523 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 if (error)
2525 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002526 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 }
2528
2529 /* Does it split the last one? */
2530 last = find_vma(mm, end);
2531 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002532 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 if (error)
2534 return error;
2535 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002536 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
2538 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002539 * unlock any mlock()ed ranges before detaching vmas
2540 */
2541 if (mm->locked_vm) {
2542 struct vm_area_struct *tmp = vma;
2543 while (tmp && tmp->vm_start < end) {
2544 if (tmp->vm_flags & VM_LOCKED) {
2545 mm->locked_vm -= vma_pages(tmp);
2546 munlock_vma_pages_all(tmp);
2547 }
2548 tmp = tmp->vm_next;
2549 }
2550 }
2551
2552 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 * Remove the vma's, and unmap the actual pages
2554 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002555 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2556 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
2558 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002559 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
2561 return 0;
2562}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
Al Virobfce2812012-04-20 21:57:04 -04002564int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002565{
2566 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002567 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002568
2569 down_write(&mm->mmap_sem);
2570 ret = do_munmap(mm, start, len);
2571 up_write(&mm->mmap_sem);
2572 return ret;
2573}
2574EXPORT_SYMBOL(vm_munmap);
2575
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002576SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002579 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580}
2581
2582static inline void verify_mm_writelocked(struct mm_struct *mm)
2583{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002584#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2586 WARN_ON(1);
2587 up_read(&mm->mmap_sem);
2588 }
2589#endif
2590}
2591
2592/*
2593 * this is really a simplified "do_mmap". it only handles
2594 * anonymous maps. eventually we may be able to do some
2595 * brk-specific accounting here.
2596 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002597static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598{
2599 struct mm_struct * mm = current->mm;
2600 struct vm_area_struct * vma, * prev;
2601 unsigned long flags;
2602 struct rb_node ** rb_link, * rb_parent;
2603 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002604 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
2606 len = PAGE_ALIGN(len);
2607 if (!len)
2608 return addr;
2609
Kirill Korotaev3a459752006-09-07 14:17:04 +04002610 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2611
Al Viro2c6a1012009-12-03 19:40:46 -05002612 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2613 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002614 return error;
2615
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002616 error = mlock_future_check(mm, mm->def_flags, len);
2617 if (error)
2618 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
2620 /*
2621 * mm->mmap_sem is required to protect against another thread
2622 * changing the mappings in case we sleep.
2623 */
2624 verify_mm_writelocked(mm);
2625
2626 /*
2627 * Clear old maps. this also does some error checking for us
2628 */
2629 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002630 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 if (do_munmap(mm, addr, len))
2632 return -ENOMEM;
2633 goto munmap_back;
2634 }
2635
2636 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002637 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 return -ENOMEM;
2639
2640 if (mm->map_count > sysctl_max_map_count)
2641 return -ENOMEM;
2642
Al Viro191c5422012-02-13 03:58:52 +00002643 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 return -ENOMEM;
2645
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002647 vma = vma_merge(mm, prev, addr, addr + len, flags,
2648 NULL, NULL, pgoff, NULL);
2649 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 goto out;
2651
2652 /*
2653 * create a vma struct for an anonymous mapping
2654 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002655 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 if (!vma) {
2657 vm_unacct_memory(len >> PAGE_SHIFT);
2658 return -ENOMEM;
2659 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
Rik van Riel5beb4932010-03-05 13:42:07 -08002661 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 vma->vm_mm = mm;
2663 vma->vm_start = addr;
2664 vma->vm_end = addr + len;
2665 vma->vm_pgoff = pgoff;
2666 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002667 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 vma_link(mm, vma, prev, rb_link, rb_parent);
2669out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002670 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002672 if (flags & VM_LOCKED)
2673 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002674 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 return addr;
2676}
2677
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002678unsigned long vm_brk(unsigned long addr, unsigned long len)
2679{
2680 struct mm_struct *mm = current->mm;
2681 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002682 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002683
2684 down_write(&mm->mmap_sem);
2685 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002686 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002687 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002688 if (populate)
2689 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002690 return ret;
2691}
2692EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693
2694/* Release all mmaps. */
2695void exit_mmap(struct mm_struct *mm)
2696{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002697 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002698 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 unsigned long nr_accounted = 0;
2700
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002701 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002702 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002703
Rik van Rielba470de2008-10-18 20:26:50 -07002704 if (mm->locked_vm) {
2705 vma = mm->mmap;
2706 while (vma) {
2707 if (vma->vm_flags & VM_LOCKED)
2708 munlock_vma_pages_all(vma);
2709 vma = vma->vm_next;
2710 }
2711 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002712
2713 arch_exit_mmap(mm);
2714
Rik van Rielba470de2008-10-18 20:26:50 -07002715 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002716 if (!vma) /* Can happen if dup_mmap() received an OOM */
2717 return;
2718
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002721 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002722 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002723 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002724 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002725
Hugh Dickins6ee86302013-04-29 15:07:44 -07002726 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002727 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002730 * Walk the list again, actually closing and freeing it,
2731 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002733 while (vma) {
2734 if (vma->vm_flags & VM_ACCOUNT)
2735 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002736 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002737 }
2738 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002739
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -08002740 WARN_ON(atomic_long_read(&mm->nr_ptes) >
2741 (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742}
2743
2744/* Insert vm structure into process list sorted by address
2745 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002746 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002748int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749{
Hugh Dickins6597d782012-10-08 16:29:07 -07002750 struct vm_area_struct *prev;
2751 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752
2753 /*
2754 * The vm_pgoff of a purely anonymous vma should be irrelevant
2755 * until its first write fault, when page's anon_vma and index
2756 * are set. But now set the vm_pgoff it will almost certainly
2757 * end up with (unless mremap moves it elsewhere before that
2758 * first wfault), so /proc/pid/maps tells a consistent story.
2759 *
2760 * By setting it to reflect the virtual start address of the
2761 * vma, merges and splits can happen in a seamless way, just
2762 * using the existing file pgoff checks and manipulations.
2763 * Similarly in do_mmap_pgoff and in do_brk.
2764 */
2765 if (!vma->vm_file) {
2766 BUG_ON(vma->anon_vma);
2767 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2768 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002769 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2770 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002772 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002773 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002774 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302775
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 vma_link(mm, vma, prev, rb_link, rb_parent);
2777 return 0;
2778}
2779
2780/*
2781 * Copy the vma structure to a new location in the same mm,
2782 * prior to moving page table entries, to effect an mremap move.
2783 */
2784struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002785 unsigned long addr, unsigned long len, pgoff_t pgoff,
2786 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787{
2788 struct vm_area_struct *vma = *vmap;
2789 unsigned long vma_start = vma->vm_start;
2790 struct mm_struct *mm = vma->vm_mm;
2791 struct vm_area_struct *new_vma, *prev;
2792 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002793 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
2795 /*
2796 * If anonymous vma has not yet been faulted, update new pgoff
2797 * to match new location, to increase its chance of merging.
2798 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002799 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002801 faulted_in_anon_vma = false;
2802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
Hugh Dickins6597d782012-10-08 16:29:07 -07002804 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2805 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2807 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2808 if (new_vma) {
2809 /*
2810 * Source vma may have been merged into new_vma
2811 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002812 if (unlikely(vma_start >= new_vma->vm_start &&
2813 vma_start < new_vma->vm_end)) {
2814 /*
2815 * The only way we can get a vma_merge with
2816 * self during an mremap is if the vma hasn't
2817 * been faulted in yet and we were allowed to
2818 * reset the dst vma->vm_pgoff to the
2819 * destination address of the mremap to allow
2820 * the merge to happen. mremap must change the
2821 * vm_pgoff linearity between src and dst vmas
2822 * (in turn preventing a vma_merge) to be
2823 * safe. It is only safe to keep the vm_pgoff
2824 * linear if there are no pages mapped yet.
2825 */
2826 VM_BUG_ON(faulted_in_anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002827 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002828 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002829 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002831 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 if (new_vma) {
2833 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 new_vma->vm_start = addr;
2835 new_vma->vm_end = addr + len;
2836 new_vma->vm_pgoff = pgoff;
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002837 if (vma_dup_policy(vma, new_vma))
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002838 goto out_free_vma;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002839 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2840 if (anon_vma_clone(new_vma, vma))
2841 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002842 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 get_file(new_vma->vm_file);
2844 if (new_vma->vm_ops && new_vma->vm_ops->open)
2845 new_vma->vm_ops->open(new_vma);
2846 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002847 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 }
2849 }
2850 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002851
2852 out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002853 mpol_put(vma_policy(new_vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002854 out_free_vma:
2855 kmem_cache_free(vm_area_cachep, new_vma);
2856 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002858
2859/*
2860 * Return true if the calling process may expand its vm space by the passed
2861 * number of pages
2862 */
2863int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2864{
2865 unsigned long cur = mm->total_vm; /* pages */
2866 unsigned long lim;
2867
Jiri Slaby59e99e52010-03-05 13:41:44 -08002868 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002869
2870 if (cur + npages > lim)
2871 return 0;
2872 return 1;
2873}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002874
2875
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002876static int special_mapping_fault(struct vm_area_struct *vma,
2877 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002878{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002879 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002880 struct page **pages;
2881
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002882 /*
2883 * special mappings have no vm_file, and in that case, the mm
2884 * uses vm_pgoff internally. So we have to subtract it from here.
2885 * We are allowed to do this because we are the mm; do not copy
2886 * this code into drivers!
2887 */
2888 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002889
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002890 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2891 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002892
2893 if (*pages) {
2894 struct page *page = *pages;
2895 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002896 vmf->page = page;
2897 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002898 }
2899
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002900 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002901}
2902
2903/*
2904 * Having a close hook prevents vma merging regardless of flags.
2905 */
2906static void special_mapping_close(struct vm_area_struct *vma)
2907{
2908}
2909
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002910static const struct vm_operations_struct special_mapping_vmops = {
Roland McGrathfa5dc222007-02-08 14:20:41 -08002911 .close = special_mapping_close,
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002912 .fault = special_mapping_fault,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002913};
2914
2915/*
2916 * Called with mm->mmap_sem held for writing.
2917 * Insert a new vma covering the given region, with the given flags.
2918 * Its pages are supplied by the given array of struct page *.
2919 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2920 * The region past the last page supplied will always produce SIGBUS.
2921 * The array pointer and the pages it points to are assumed to stay alive
2922 * for as long as this mapping might exist.
2923 */
Stefani Seibold3935ed62014-03-17 23:22:02 +01002924struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
Roland McGrathfa5dc222007-02-08 14:20:41 -08002925 unsigned long addr, unsigned long len,
2926 unsigned long vm_flags, struct page **pages)
2927{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002928 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002929 struct vm_area_struct *vma;
2930
2931 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2932 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01002933 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002934
Rik van Riel5beb4932010-03-05 13:42:07 -08002935 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002936 vma->vm_mm = mm;
2937 vma->vm_start = addr;
2938 vma->vm_end = addr + len;
2939
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002940 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07002941 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002942
2943 vma->vm_ops = &special_mapping_vmops;
2944 vma->vm_private_data = pages;
2945
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002946 ret = insert_vm_struct(mm, vma);
2947 if (ret)
2948 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002949
2950 mm->total_vm += len >> PAGE_SHIFT;
2951
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002952 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002953
Stefani Seibold3935ed62014-03-17 23:22:02 +01002954 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002955
2956out:
2957 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01002958 return ERR_PTR(ret);
2959}
2960
2961int install_special_mapping(struct mm_struct *mm,
2962 unsigned long addr, unsigned long len,
2963 unsigned long vm_flags, struct page **pages)
2964{
2965 struct vm_area_struct *vma = _install_special_mapping(mm,
2966 addr, len, vm_flags, pages);
2967
2968 if (IS_ERR(vma))
2969 return PTR_ERR(vma);
2970 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002971}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002972
2973static DEFINE_MUTEX(mm_all_locks_mutex);
2974
Peter Zijlstra454ed842008-08-11 09:30:25 +02002975static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002976{
Michel Lespinassebf181b92012-10-08 16:31:39 -07002977 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002978 /*
2979 * The LSB of head.next can't change from under us
2980 * because we hold the mm_all_locks_mutex.
2981 */
Jiri Kosina572043c2013-01-11 14:31:59 -08002982 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002983 /*
2984 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00002985 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002986 * the same anon_vma we won't take it again.
2987 *
2988 * No need of atomic instructions here, head.next
2989 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00002990 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002991 */
2992 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07002993 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002994 BUG();
2995 }
2996}
2997
Peter Zijlstra454ed842008-08-11 09:30:25 +02002998static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07002999{
3000 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3001 /*
3002 * AS_MM_ALL_LOCKS can't change from under us because
3003 * we hold the mm_all_locks_mutex.
3004 *
3005 * Operations on ->flags have to be atomic because
3006 * even if AS_MM_ALL_LOCKS is stable thanks to the
3007 * mm_all_locks_mutex, there may be other cpus
3008 * changing other bitflags in parallel to us.
3009 */
3010 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3011 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003012 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003013 }
3014}
3015
3016/*
3017 * This operation locks against the VM for all pte/vma/mm related
3018 * operations that could ever happen on a certain mm. This includes
3019 * vmtruncate, try_to_unmap, and all page faults.
3020 *
3021 * The caller must take the mmap_sem in write mode before calling
3022 * mm_take_all_locks(). The caller isn't allowed to release the
3023 * mmap_sem until mm_drop_all_locks() returns.
3024 *
3025 * mmap_sem in write mode is required in order to block all operations
3026 * that could modify pagetables and free pages without need of
3027 * altering the vma layout (for example populate_range() with
3028 * nonlinear vmas). It's also needed in write mode to avoid new
3029 * anon_vmas to be associated with existing vmas.
3030 *
3031 * A single task can't take more than one mm_take_all_locks() in a row
3032 * or it would deadlock.
3033 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003034 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003035 * mapping->flags avoid to take the same lock twice, if more than one
3036 * vma in this mm is backed by the same anon_vma or address_space.
3037 *
3038 * We can take all the locks in random order because the VM code
Yuanhan Liu631b0cf2013-02-04 14:28:48 -08003039 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003040 * takes more than one of them in a row. Secondly we're protected
3041 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3042 *
3043 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3044 * that may have to take thousand of locks.
3045 *
3046 * mm_take_all_locks() can fail if it's interrupted by signals.
3047 */
3048int mm_take_all_locks(struct mm_struct *mm)
3049{
3050 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003051 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003052
3053 BUG_ON(down_read_trylock(&mm->mmap_sem));
3054
3055 mutex_lock(&mm_all_locks_mutex);
3056
3057 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3058 if (signal_pending(current))
3059 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003060 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003061 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003062 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003063
3064 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3065 if (signal_pending(current))
3066 goto out_unlock;
3067 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003068 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3069 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003070 }
3071
Kautuk Consul584cff52011-10-31 17:08:59 -07003072 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003073
3074out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003075 mm_drop_all_locks(mm);
3076 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003077}
3078
3079static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3080{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003081 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003082 /*
3083 * The LSB of head.next can't change to 0 from under
3084 * us because we hold the mm_all_locks_mutex.
3085 *
3086 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003087 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003088 * never see our bitflag.
3089 *
3090 * No need of atomic instructions here, head.next
3091 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003092 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003093 */
3094 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003095 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003096 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003097 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003098 }
3099}
3100
3101static void vm_unlock_mapping(struct address_space *mapping)
3102{
3103 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3104 /*
3105 * AS_MM_ALL_LOCKS can't change to 0 from under us
3106 * because we hold the mm_all_locks_mutex.
3107 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003108 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003109 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3110 &mapping->flags))
3111 BUG();
3112 }
3113}
3114
3115/*
3116 * The mmap_sem cannot be released by the caller until
3117 * mm_drop_all_locks() returns.
3118 */
3119void mm_drop_all_locks(struct mm_struct *mm)
3120{
3121 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003122 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003123
3124 BUG_ON(down_read_trylock(&mm->mmap_sem));
3125 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3126
3127 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3128 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003129 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3130 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003131 if (vma->vm_file && vma->vm_file->f_mapping)
3132 vm_unlock_mapping(vma->vm_file->f_mapping);
3133 }
3134
3135 mutex_unlock(&mm_all_locks_mutex);
3136}
David Howells8feae132009-01-08 12:04:47 +00003137
3138/*
3139 * initialise the VMA slab
3140 */
3141void __init mmap_init(void)
3142{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003143 int ret;
3144
3145 ret = percpu_counter_init(&vm_committed_as, 0);
3146 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003147}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003148
3149/*
3150 * Initialise sysctl_user_reserve_kbytes.
3151 *
3152 * This is intended to prevent a user from starting a single memory hogging
3153 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3154 * mode.
3155 *
3156 * The default value is min(3% of free memory, 128MB)
3157 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3158 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003159static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003160{
3161 unsigned long free_kbytes;
3162
3163 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3164
3165 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3166 return 0;
3167}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003168subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003169
3170/*
3171 * Initialise sysctl_admin_reserve_kbytes.
3172 *
3173 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3174 * to log in and kill a memory hogging process.
3175 *
3176 * Systems with more than 256MB will reserve 8MB, enough to recover
3177 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3178 * only reserve 3% of free pages by default.
3179 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003180static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003181{
3182 unsigned long free_kbytes;
3183
3184 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3185
3186 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3187 return 0;
3188}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003189subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003190
3191/*
3192 * Reinititalise user and admin reserves if memory is added or removed.
3193 *
3194 * The default user reserve max is 128MB, and the default max for the
3195 * admin reserve is 8MB. These are usually, but not always, enough to
3196 * enable recovery from a memory hogging process using login/sshd, a shell,
3197 * and tools like top. It may make sense to increase or even disable the
3198 * reserve depending on the existence of swap or variations in the recovery
3199 * tools. So, the admin may have changed them.
3200 *
3201 * If memory is added and the reserves have been eliminated or increased above
3202 * the default max, then we'll trust the admin.
3203 *
3204 * If memory is removed and there isn't enough free memory, then we
3205 * need to reset the reserves.
3206 *
3207 * Otherwise keep the reserve set by the admin.
3208 */
3209static int reserve_mem_notifier(struct notifier_block *nb,
3210 unsigned long action, void *data)
3211{
3212 unsigned long tmp, free_kbytes;
3213
3214 switch (action) {
3215 case MEM_ONLINE:
3216 /* Default max is 128MB. Leave alone if modified by operator. */
3217 tmp = sysctl_user_reserve_kbytes;
3218 if (0 < tmp && tmp < (1UL << 17))
3219 init_user_reserve();
3220
3221 /* Default max is 8MB. Leave alone if modified by operator. */
3222 tmp = sysctl_admin_reserve_kbytes;
3223 if (0 < tmp && tmp < (1UL << 13))
3224 init_admin_reserve();
3225
3226 break;
3227 case MEM_OFFLINE:
3228 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3229
3230 if (sysctl_user_reserve_kbytes > free_kbytes) {
3231 init_user_reserve();
3232 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3233 sysctl_user_reserve_kbytes);
3234 }
3235
3236 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3237 init_admin_reserve();
3238 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3239 sysctl_admin_reserve_kbytes);
3240 }
3241 break;
3242 default:
3243 break;
3244 }
3245 return NOTIFY_OK;
3246}
3247
3248static struct notifier_block reserve_mem_nb = {
3249 .notifier_call = reserve_mem_notifier,
3250};
3251
3252static int __meminit init_reserve_notifier(void)
3253{
3254 if (register_hotmemory_notifier(&reserve_mem_nb))
3255 printk("Failed registering memory add/remove notifier for admin reserve");
3256
3257 return 0;
3258}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003259subsys_initcall(init_reserve_notifier);