blob: c1f2ea4a0b9960d39940c82f4ba329dad2df74df [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
Alan Cox046c6882009-01-05 14:06:29 +00006 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 */
8
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07009#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
Cyril Hrubise8420a82013-04-29 15:08:33 -070011#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/slab.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070013#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/mm.h>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070015#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/shm.h>
17#include <linux/mman.h>
18#include <linux/pagemap.h>
19#include <linux/swap.h>
20#include <linux/syscalls.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080021#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/init.h>
23#include <linux/file.h>
24#include <linux/fs.h>
25#include <linux/personality.h>
26#include <linux/security.h>
27#include <linux/hugetlb.h>
28#include <linux/profile.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040029#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/mount.h>
31#include <linux/mempolicy.h>
32#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070033#include <linux/mmu_notifier.h>
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -070034#include <linux/mmdebug.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020035#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040036#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080037#include <linux/khugepaged.h>
Srikar Dronamraju2b144492012-02-09 14:56:42 +053038#include <linux/uprobes.h>
Michel Lespinassed3737182012-12-11 16:01:38 -080039#include <linux/rbtree_augmented.h>
Clark Williamscf4aebc22013-02-07 09:46:59 -060040#include <linux/sched/sysctl.h>
Andrew Shewmaker16408792013-04-29 15:08:12 -070041#include <linux/notifier.h>
42#include <linux/memory.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070043#include <linux/printk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#include <asm/uaccess.h>
46#include <asm/cacheflush.h>
47#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020048#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Jan Beulich42b77722008-07-23 21:27:10 -070050#include "internal.h"
51
Kirill Korotaev3a459752006-09-07 14:17:04 +040052#ifndef arch_mmap_check
53#define arch_mmap_check(addr, len, flags) (0)
54#endif
55
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080056#ifndef arch_rebalance_pgtables
57#define arch_rebalance_pgtables(addr, len) (addr)
58#endif
59
Hugh Dickinse0da3822005-04-19 13:29:15 -070060static void unmap_region(struct mm_struct *mm,
61 struct vm_area_struct *vma, struct vm_area_struct *prev,
62 unsigned long start, unsigned long end);
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/* description of effects of mapping type and prot in current implementation.
65 * this is due to the limited x86 page protection hardware. The expected
66 * behavior is in parens:
67 *
68 * map_type prot
69 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
70 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (yes) yes w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
73 *
74 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
75 * w: (no) no w: (no) no w: (copy) copy w: (no) no
76 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
77 *
78 */
79pgprot_t protection_map[16] = {
80 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
81 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
82};
83
Hugh Dickins804af2c2006-07-26 21:39:49 +010084pgprot_t vm_get_page_prot(unsigned long vm_flags)
85{
Dave Kleikampb845f312008-07-08 00:28:51 +100086 return __pgprot(pgprot_val(protection_map[vm_flags &
87 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
88 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +010089}
90EXPORT_SYMBOL(vm_get_page_prot);
91
Shaohua Li34679d72011-05-24 17:11:18 -070092int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
93int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -080094unsigned long sysctl_overcommit_kbytes __read_mostly;
Christoph Lameterc3d8c142005-09-06 15:16:33 -070095int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070096unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070097unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Shaohua Li34679d72011-05-24 17:11:18 -070098/*
99 * Make sure vm_committed_as in one cacheline and not cacheline shared with
100 * other variables. It can be updated by several CPUs frequently.
101 */
102struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104/*
K. Y. Srinivasan997071b2012-11-15 14:34:42 -0800105 * The global memory commitment made in the system can be a metric
106 * that can be used to drive ballooning decisions when Linux is hosted
107 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
108 * balancing memory across competing virtual machines that are hosted.
109 * Several metrics drive this policy engine including the guest reported
110 * memory commitment.
111 */
112unsigned long vm_memory_committed(void)
113{
114 return percpu_counter_read_positive(&vm_committed_as);
115}
116EXPORT_SYMBOL_GPL(vm_memory_committed);
117
118/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 * Check that a process has enough memory to allocate a new virtual
120 * mapping. 0 means there is enough memory for the allocation to
121 * succeed and -ENOMEM implies there is not.
122 *
123 * We currently support three overcommit policies, which are set via the
124 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
125 *
126 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
127 * Additional code 2002 Jul 20 by Robert Love.
128 *
129 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
130 *
131 * Note this is a helper function intended to be used by LSMs which
132 * wish to use this logic.
133 */
Alan Cox34b4e4a2007-08-22 14:01:28 -0700134int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700136 unsigned long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -0700138 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
139 -(s64)vm_committed_as_batch * num_online_cpus(),
140 "memory commitment underflow");
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 vm_acct_memory(pages);
143
144 /*
145 * Sometimes we want to use more memory than we have
146 */
147 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
148 return 0;
149
150 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -0700151 free = global_page_state(NR_FREE_PAGES);
152 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Dmitry Finkc15bef32011-07-25 17:12:19 -0700154 /*
155 * shmem pages shouldn't be counted as free in this
156 * case, they can't be purged, only swapped out, and
157 * that won't affect the overall amount of available
158 * memory in the system.
159 */
160 free -= global_page_state(NR_SHMEM);
161
Shaohua Liec8acf22013-02-22 16:34:38 -0800162 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164 /*
165 * Any slabs which are created with the
166 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
167 * which are reclaimable, under pressure. The dentry
168 * cache and most inode caches should fall into this
169 */
Christoph Lameter972d1a72006-09-25 23:31:51 -0700170 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
172 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -0700173 * Leave reserved pages. The pages are not for anonymous pages.
174 */
175 if (free <= totalreserve_pages)
176 goto error;
177 else
178 free -= totalreserve_pages;
179
180 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700181 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 */
183 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700184 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
186 if (free > pages)
187 return 0;
188
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700189 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 }
191
Jerome Marchand00619bc2013-11-12 15:08:31 -0800192 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700194 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 */
196 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -0700197 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -0700199 /*
200 * Don't let a single process grow so big a user can't recover
201 */
202 if (mm) {
203 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
204 allowed -= min(mm->total_vm / 32, reserve);
205 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700207 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return 0;
Hideo AOKI6d9f7832006-04-10 22:53:00 -0700209error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 vm_unacct_memory(pages);
211
212 return -ENOMEM;
213}
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215/*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700216 * Requires inode->i_mapping->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 */
218static void __remove_shared_vm_struct(struct vm_area_struct *vma,
219 struct file *file, struct address_space *mapping)
220{
221 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500222 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700224 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
226 flush_dcache_mmap_lock(mapping);
227 if (unlikely(vma->vm_flags & VM_NONLINEAR))
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700228 list_del_init(&vma->shared.nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700230 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 flush_dcache_mmap_unlock(mapping);
232}
233
234/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700235 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700236 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700238void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 struct file *file = vma->vm_file;
241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 if (file) {
243 struct address_space *mapping = file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700244 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 __remove_shared_vm_struct(vma, file, mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700246 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700248}
249
250/*
251 * Close a vm structure and free it, returning the next.
252 */
253static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
254{
255 struct vm_area_struct *next = vma->vm_next;
256
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700257 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 if (vma->vm_ops && vma->vm_ops->close)
259 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700260 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700261 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700262 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700264 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700267static unsigned long do_brk(unsigned long addr, unsigned long len);
268
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100269SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
271 unsigned long rlim, retval;
272 unsigned long newbrk, oldbrk;
273 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700274 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800275 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277 down_write(&mm->mmap_sem);
278
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700279#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800280 /*
281 * CONFIG_COMPAT_BRK can still be overridden by setting
282 * randomize_va_space to 2, which will still cause mm->start_brk
283 * to be arbitrarily shifted
284 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700285 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800286 min_brk = mm->start_brk;
287 else
288 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700289#else
290 min_brk = mm->start_brk;
291#endif
292 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700294
295 /*
296 * Check against rlimit here. If this check is done later after the test
297 * of oldbrk with newbrk then it can escape the test and let the data
298 * segment grow beyond its set limit the in case where the limit is
299 * not page aligned -Ram Gupta
300 */
Jiri Slaby59e99e52010-03-05 13:41:44 -0800301 rlim = rlimit(RLIMIT_DATA);
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100302 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
303 (mm->end_data - mm->start_data) > rlim)
Ram Gupta1e624192006-04-10 22:52:57 -0700304 goto out;
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 newbrk = PAGE_ALIGN(brk);
307 oldbrk = PAGE_ALIGN(mm->brk);
308 if (oldbrk == newbrk)
309 goto set_brk;
310
311 /* Always allow shrinking brk. */
312 if (brk <= mm->brk) {
313 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
314 goto set_brk;
315 goto out;
316 }
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 /* Check against existing mmap mappings. */
319 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
320 goto out;
321
322 /* Ok, looks good - let it rip. */
323 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
324 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326set_brk:
327 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800328 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
329 up_write(&mm->mmap_sem);
330 if (populate)
331 mm_populate(oldbrk, newbrk - oldbrk);
332 return brk;
333
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334out:
335 retval = mm->brk;
336 up_write(&mm->mmap_sem);
337 return retval;
338}
339
Michel Lespinassed3737182012-12-11 16:01:38 -0800340static long vma_compute_subtree_gap(struct vm_area_struct *vma)
341{
342 unsigned long max, subtree_gap;
343 max = vma->vm_start;
344 if (vma->vm_prev)
345 max -= vma->vm_prev->vm_end;
346 if (vma->vm_rb.rb_left) {
347 subtree_gap = rb_entry(vma->vm_rb.rb_left,
348 struct vm_area_struct, vm_rb)->rb_subtree_gap;
349 if (subtree_gap > max)
350 max = subtree_gap;
351 }
352 if (vma->vm_rb.rb_right) {
353 subtree_gap = rb_entry(vma->vm_rb.rb_right,
354 struct vm_area_struct, vm_rb)->rb_subtree_gap;
355 if (subtree_gap > max)
356 max = subtree_gap;
357 }
358 return max;
359}
360
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700361#ifdef CONFIG_DEBUG_VM_RB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362static int browse_rb(struct rb_root *root)
363{
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800364 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 struct rb_node *nd, *pn = NULL;
366 unsigned long prev = 0, pend = 0;
367
368 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
369 struct vm_area_struct *vma;
370 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800371 if (vma->vm_start < prev) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700372 pr_info("vm_start %lx prev %lx\n", vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800373 bug = 1;
374 }
375 if (vma->vm_start < pend) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700376 pr_info("vm_start %lx pend %lx\n", vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800377 bug = 1;
378 }
379 if (vma->vm_start > vma->vm_end) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700380 pr_info("vm_end %lx < vm_start %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800381 vma->vm_end, vma->vm_start);
382 bug = 1;
383 }
384 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700385 pr_info("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800386 vma->rb_subtree_gap,
387 vma_compute_subtree_gap(vma));
388 bug = 1;
389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 i++;
391 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800392 prev = vma->vm_start;
393 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
395 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800396 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800398 if (i != j) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700399 pr_info("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800400 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800402 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403}
404
Michel Lespinassed3737182012-12-11 16:01:38 -0800405static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
406{
407 struct rb_node *nd;
408
409 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
410 struct vm_area_struct *vma;
411 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
412 BUG_ON(vma != ignore &&
413 vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700417static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
419 int bug = 0;
420 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800421 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700422 struct vm_area_struct *vma = mm->mmap;
423 while (vma) {
424 struct anon_vma_chain *avc;
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800425 vma_lock_anon_vma(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700426 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
427 anon_vma_interval_tree_verify(avc);
Michel Lespinasse63c3b902012-11-16 14:14:47 -0800428 vma_unlock_anon_vma(vma);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800429 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700430 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 i++;
432 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800433 if (i != mm->map_count) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700434 pr_info("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800435 bug = 1;
436 }
437 if (highest_address != mm->highest_vm_end) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700438 pr_info("mm->highest_vm_end %lx, found %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800439 mm->highest_vm_end, highest_address);
440 bug = 1;
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 i = browse_rb(&mm->mm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800443 if (i != mm->map_count) {
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700444 pr_info("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800445 bug = 1;
446 }
Eric Sesterhenn46a350e2006-04-01 01:23:29 +0200447 BUG_ON(bug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800450#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451#define validate_mm(mm) do { } while (0)
452#endif
453
Michel Lespinassed3737182012-12-11 16:01:38 -0800454RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
455 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
456
457/*
458 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
459 * vma->vm_prev->vm_end values changed, without modifying the vma's position
460 * in the rbtree.
461 */
462static void vma_gap_update(struct vm_area_struct *vma)
463{
464 /*
465 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
466 * function that does exacltly what we want.
467 */
468 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
469}
470
471static inline void vma_rb_insert(struct vm_area_struct *vma,
472 struct rb_root *root)
473{
474 /* All rb_subtree_gap values must be consistent prior to insertion */
475 validate_mm_rb(root, NULL);
476
477 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
478}
479
480static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
481{
482 /*
483 * All rb_subtree_gap values must be consistent prior to erase,
484 * with the possible exception of the vma being erased.
485 */
486 validate_mm_rb(root, vma);
487
488 /*
489 * Note rb_erase_augmented is a fairly large inline function,
490 * so make sure we instantiate it only once with our desired
491 * augmented rbtree callbacks.
492 */
493 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
494}
495
Michel Lespinassebf181b92012-10-08 16:31:39 -0700496/*
497 * vma has some anon_vma assigned, and is already inserted on that
498 * anon_vma's interval trees.
499 *
500 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
501 * vma must be removed from the anon_vma's interval trees using
502 * anon_vma_interval_tree_pre_update_vma().
503 *
504 * After the update, the vma will be reinserted using
505 * anon_vma_interval_tree_post_update_vma().
506 *
507 * The entire update must be protected by exclusive mmap_sem and by
508 * the root anon_vma's mutex.
509 */
510static inline void
511anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
512{
513 struct anon_vma_chain *avc;
514
515 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
516 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
517}
518
519static inline void
520anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
521{
522 struct anon_vma_chain *avc;
523
524 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
525 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
526}
527
Hugh Dickins6597d782012-10-08 16:29:07 -0700528static int find_vma_links(struct mm_struct *mm, unsigned long addr,
529 unsigned long end, struct vm_area_struct **pprev,
530 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
Hugh Dickins6597d782012-10-08 16:29:07 -0700532 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534 __rb_link = &mm->mm_rb.rb_node;
535 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
537 while (*__rb_link) {
538 struct vm_area_struct *vma_tmp;
539
540 __rb_parent = *__rb_link;
541 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
542
543 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700544 /* Fail if an existing vma overlaps the area */
545 if (vma_tmp->vm_start < end)
546 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 __rb_link = &__rb_parent->rb_left;
548 } else {
549 rb_prev = __rb_parent;
550 __rb_link = &__rb_parent->rb_right;
551 }
552 }
553
554 *pprev = NULL;
555 if (rb_prev)
556 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
557 *rb_link = __rb_link;
558 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700559 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
Cyril Hrubise8420a82013-04-29 15:08:33 -0700562static unsigned long count_vma_pages_range(struct mm_struct *mm,
563 unsigned long addr, unsigned long end)
564{
565 unsigned long nr_pages = 0;
566 struct vm_area_struct *vma;
567
568 /* Find first overlaping mapping */
569 vma = find_vma_intersection(mm, addr, end);
570 if (!vma)
571 return 0;
572
573 nr_pages = (min(end, vma->vm_end) -
574 max(addr, vma->vm_start)) >> PAGE_SHIFT;
575
576 /* Iterate over the rest of the overlaps */
577 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
578 unsigned long overlap_len;
579
580 if (vma->vm_start > end)
581 break;
582
583 overlap_len = min(end, vma->vm_end) - vma->vm_start;
584 nr_pages += overlap_len >> PAGE_SHIFT;
585 }
586
587 return nr_pages;
588}
589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
591 struct rb_node **rb_link, struct rb_node *rb_parent)
592{
Michel Lespinassed3737182012-12-11 16:01:38 -0800593 /* Update tracking information for the gap following the new vma. */
594 if (vma->vm_next)
595 vma_gap_update(vma->vm_next);
596 else
597 mm->highest_vm_end = vma->vm_end;
598
599 /*
600 * vma->vm_prev wasn't known when we followed the rbtree to find the
601 * correct insertion point for that vma. As a result, we could not
602 * update the vma vm_rb parents rb_subtree_gap values on the way down.
603 * So, we first insert the vma with a zero rb_subtree_gap value
604 * (to be consistent with what we did on the way down), and then
605 * immediately update the gap to the correct value. Finally we
606 * rebalance the rbtree after all augmented values have been set.
607 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800609 vma->rb_subtree_gap = 0;
610 vma_gap_update(vma);
611 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700614static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
ZhenwenXu48aae422009-01-06 14:40:21 -0800616 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
618 file = vma->vm_file;
619 if (file) {
620 struct address_space *mapping = file->f_mapping;
621
622 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500623 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700625 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
627 flush_dcache_mmap_lock(mapping);
628 if (unlikely(vma->vm_flags & VM_NONLINEAR))
629 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
630 else
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700631 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 flush_dcache_mmap_unlock(mapping);
633 }
634}
635
636static void
637__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 __vma_link_list(mm, vma, prev, rb_parent);
642 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643}
644
645static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
646 struct vm_area_struct *prev, struct rb_node **rb_link,
647 struct rb_node *rb_parent)
648{
649 struct address_space *mapping = NULL;
650
Huang Shijie64ac4942014-06-04 16:07:33 -0700651 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 mapping = vma->vm_file->f_mapping;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700653 mutex_lock(&mapping->i_mmap_mutex);
Huang Shijie64ac4942014-06-04 16:07:33 -0700654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
656 __vma_link(mm, vma, prev, rb_link, rb_parent);
657 __vma_link_file(vma);
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700660 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
662 mm->map_count++;
663 validate_mm(mm);
664}
665
666/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700667 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700668 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800670static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671{
Hugh Dickins6597d782012-10-08 16:29:07 -0700672 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800673 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Hugh Dickins6597d782012-10-08 16:29:07 -0700675 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
676 &prev, &rb_link, &rb_parent))
677 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 __vma_link(mm, vma, prev, rb_link, rb_parent);
679 mm->map_count++;
680}
681
682static inline void
683__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
684 struct vm_area_struct *prev)
685{
Michel Lespinassed3737182012-12-11 16:01:38 -0800686 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700687
Michel Lespinassed3737182012-12-11 16:01:38 -0800688 vma_rb_erase(vma, &mm->mm_rb);
689 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700690 if (next)
691 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700692
693 /* Kill the cache */
694 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
697/*
698 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
699 * is already present in an i_mmap tree without adjusting the tree.
700 * The following helper function should be used when such adjustments
701 * are necessary. The "insert" vma (if any) is to be inserted
702 * before we drop the necessary locks.
703 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800704int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
706{
707 struct mm_struct *mm = vma->vm_mm;
708 struct vm_area_struct *next = vma->vm_next;
709 struct vm_area_struct *importer = NULL;
710 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700711 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700712 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800714 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 long adjust_next = 0;
716 int remove_next = 0;
717
718 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700719 struct vm_area_struct *exporter = NULL;
720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 if (end >= next->vm_end) {
722 /*
723 * vma expands, overlapping all the next, and
724 * perhaps the one after too (mprotect case 6).
725 */
726again: remove_next = 1 + (end > next->vm_end);
727 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700728 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 importer = vma;
730 } else if (end > next->vm_start) {
731 /*
732 * vma expands, overlapping part of the next:
733 * mprotect case 5 shifting the boundary up.
734 */
735 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700736 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 importer = vma;
738 } else if (end < vma->vm_end) {
739 /*
740 * vma shrinks, and !insert tells it's not
741 * split_vma inserting another: so it must be
742 * mprotect case 4 shifting the boundary down.
743 */
744 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700745 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 importer = next;
747 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Rik van Riel5beb4932010-03-05 13:42:07 -0800749 /*
750 * Easily overlooked: when mprotect shifts the boundary,
751 * make sure the expanding vma has anon_vma set if the
752 * shrinking vma had, to cover any anon pages imported.
753 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700754 if (exporter && exporter->anon_vma && !importer->anon_vma) {
755 if (anon_vma_clone(importer, exporter))
Rik van Riel5beb4932010-03-05 13:42:07 -0800756 return -ENOMEM;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700757 importer->anon_vma = exporter->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800758 }
759 }
760
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 if (file) {
762 mapping = file->f_mapping;
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530763 if (!(vma->vm_flags & VM_NONLINEAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 root = &mapping->i_mmap;
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530765 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530766
767 if (adjust_next)
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530768 uprobe_munmap(next, next->vm_start,
769 next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530770 }
771
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700772 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700775 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 * are visible to arm/parisc __flush_dcache_page
777 * throughout; but we cannot insert into address
778 * space until vma start or end is updated.
779 */
780 __vma_link_file(insert);
781 }
782 }
783
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800784 vma_adjust_trans_huge(vma, start, end, adjust_next);
785
Michel Lespinassebf181b92012-10-08 16:31:39 -0700786 anon_vma = vma->anon_vma;
787 if (!anon_vma && adjust_next)
788 anon_vma = next->anon_vma;
789 if (anon_vma) {
Michel Lespinasseca42b262012-10-08 16:30:01 -0700790 VM_BUG_ON(adjust_next && next->anon_vma &&
791 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000792 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700793 anon_vma_interval_tree_pre_update_vma(vma);
794 if (adjust_next)
795 anon_vma_interval_tree_pre_update_vma(next);
796 }
Rik van Riel012f18002010-08-09 17:18:40 -0700797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 if (root) {
799 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700800 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700802 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
804
Michel Lespinassed3737182012-12-11 16:01:38 -0800805 if (start != vma->vm_start) {
806 vma->vm_start = start;
807 start_changed = true;
808 }
809 if (end != vma->vm_end) {
810 vma->vm_end = end;
811 end_changed = true;
812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 vma->vm_pgoff = pgoff;
814 if (adjust_next) {
815 next->vm_start += adjust_next << PAGE_SHIFT;
816 next->vm_pgoff += adjust_next;
817 }
818
819 if (root) {
820 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700821 vma_interval_tree_insert(next, root);
822 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 flush_dcache_mmap_unlock(mapping);
824 }
825
826 if (remove_next) {
827 /*
828 * vma_merge has merged next into vma, and needs
829 * us to remove next before dropping the locks.
830 */
831 __vma_unlink(mm, next, vma);
832 if (file)
833 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 } else if (insert) {
835 /*
836 * split_vma has split insert from vma, and needs
837 * us to insert it before dropping the locks
838 * (it may either follow vma or precede it).
839 */
840 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800841 } else {
842 if (start_changed)
843 vma_gap_update(vma);
844 if (end_changed) {
845 if (!next)
846 mm->highest_vm_end = end;
847 else if (!adjust_next)
848 vma_gap_update(next);
849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851
Michel Lespinassebf181b92012-10-08 16:31:39 -0700852 if (anon_vma) {
853 anon_vma_interval_tree_post_update_vma(vma);
854 if (adjust_next)
855 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800856 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700857 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700859 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530861 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100862 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530863
864 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100865 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530866 }
867
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700869 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530870 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700872 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800873 if (next->anon_vma)
874 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700876 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 kmem_cache_free(vm_area_cachep, next);
878 /*
879 * In mprotect's case 6 (see comments on vma_merge),
880 * we must remove another next too. It would clutter
881 * up the code too much to do both in one go.
882 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800883 next = vma->vm_next;
884 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800886 else if (next)
887 vma_gap_update(next);
888 else
889 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530891 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100892 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
894 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800895
896 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
899/*
900 * If the vma has a ->close operation then the driver probably needs to release
901 * per-vma resources, so we don't attempt to merge those.
902 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903static inline int is_mergeable_vma(struct vm_area_struct *vma,
904 struct file *file, unsigned long vm_flags)
905{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800906 /*
907 * VM_SOFTDIRTY should not prevent from VMA merging, if we
908 * match the flags but dirty bit -- the caller should mark
909 * merged VMA as dirty. If dirty bit won't be excluded from
910 * comparison, we increase pressue on the memory system forcing
911 * the kernel to generate new VMAs when old one could be
912 * extended instead.
913 */
914 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 return 0;
916 if (vma->vm_file != file)
917 return 0;
918 if (vma->vm_ops && vma->vm_ops->close)
919 return 0;
920 return 1;
921}
922
923static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700924 struct anon_vma *anon_vma2,
925 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926{
Shaohua Li965f55d2011-05-24 17:11:20 -0700927 /*
928 * The list_is_singular() test is to avoid merging VMA cloned from
929 * parents. This can improve scalability caused by anon_vma lock.
930 */
931 if ((!anon_vma1 || !anon_vma2) && (!vma ||
932 list_is_singular(&vma->anon_vma_chain)))
933 return 1;
934 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935}
936
937/*
938 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
939 * in front of (at a lower virtual address and file offset than) the vma.
940 *
941 * We cannot merge two vmas if they have differently assigned (non-NULL)
942 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
943 *
944 * We don't check here for the merged mmap wrapping around the end of pagecache
945 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
946 * wrap, nor mmaps which cover the final page at index -1UL.
947 */
948static int
949can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
950 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
951{
952 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700953 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 if (vma->vm_pgoff == vm_pgoff)
955 return 1;
956 }
957 return 0;
958}
959
960/*
961 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
962 * beyond (at a higher virtual address and file offset than) the vma.
963 *
964 * We cannot merge two vmas if they have differently assigned (non-NULL)
965 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
966 */
967static int
968can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
969 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
970{
971 if (is_mergeable_vma(vma, file, vm_flags) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700972 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700974 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
976 return 1;
977 }
978 return 0;
979}
980
981/*
982 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
983 * whether that can be merged with its predecessor or its successor.
984 * Or both (it neatly fills a hole).
985 *
986 * In most cases - when called for mmap, brk or mremap - [addr,end) is
987 * certain not to be mapped by the time vma_merge is called; but when
988 * called for mprotect, it is certain to be already mapped (either at
989 * an offset within prev, or at the start of next), and the flags of
990 * this area are about to be changed to vm_flags - and the no-change
991 * case has already been eliminated.
992 *
993 * The following mprotect cases have to be considered, where AAAA is
994 * the area passed down from mprotect_fixup, never extending beyond one
995 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
996 *
997 * AAAA AAAA AAAA AAAA
998 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
999 * cannot merge might become might become might become
1000 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1001 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1002 * mremap move: PPPPNNNNNNNN 8
1003 * AAAA
1004 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1005 * might become case 1 below case 2 below case 3 below
1006 *
1007 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1008 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1009 */
1010struct vm_area_struct *vma_merge(struct mm_struct *mm,
1011 struct vm_area_struct *prev, unsigned long addr,
1012 unsigned long end, unsigned long vm_flags,
1013 struct anon_vma *anon_vma, struct file *file,
1014 pgoff_t pgoff, struct mempolicy *policy)
1015{
1016 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1017 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001018 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 /*
1021 * We later require that vma->vm_flags == vm_flags,
1022 * so this tests vma->vm_flags & VM_SPECIAL, too.
1023 */
1024 if (vm_flags & VM_SPECIAL)
1025 return NULL;
1026
1027 if (prev)
1028 next = prev->vm_next;
1029 else
1030 next = mm->mmap;
1031 area = next;
1032 if (next && next->vm_end == end) /* cases 6, 7, 8 */
1033 next = next->vm_next;
1034
1035 /*
1036 * Can it merge with the predecessor?
1037 */
1038 if (prev && prev->vm_end == addr &&
1039 mpol_equal(vma_policy(prev), policy) &&
1040 can_vma_merge_after(prev, vm_flags,
1041 anon_vma, file, pgoff)) {
1042 /*
1043 * OK, it can. Can we now merge in the successor as well?
1044 */
1045 if (next && end == next->vm_start &&
1046 mpol_equal(policy, vma_policy(next)) &&
1047 can_vma_merge_before(next, vm_flags,
1048 anon_vma, file, pgoff+pglen) &&
1049 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001050 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001052 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 next->vm_end, prev->vm_pgoff, NULL);
1054 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001055 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001057 if (err)
1058 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001059 khugepaged_enter_vma_merge(prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 return prev;
1061 }
1062
1063 /*
1064 * Can this new request be merged in front of next?
1065 */
1066 if (next && end == next->vm_start &&
1067 mpol_equal(policy, vma_policy(next)) &&
1068 can_vma_merge_before(next, vm_flags,
1069 anon_vma, file, pgoff+pglen)) {
1070 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001071 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 addr, prev->vm_pgoff, NULL);
1073 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001074 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001076 if (err)
1077 return NULL;
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08001078 khugepaged_enter_vma_merge(area);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return area;
1080 }
1081
1082 return NULL;
1083}
1084
1085/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001086 * Rough compatbility check to quickly see if it's even worth looking
1087 * at sharing an anon_vma.
1088 *
1089 * They need to have the same vm_file, and the flags can only differ
1090 * in things that mprotect may change.
1091 *
1092 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1093 * we can merge the two vma's. For example, we refuse to merge a vma if
1094 * there is a vm_ops->close() function, because that indicates that the
1095 * driver is doing some kind of reference counting. But that doesn't
1096 * really matter for the anon_vma sharing case.
1097 */
1098static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1099{
1100 return a->vm_end == b->vm_start &&
1101 mpol_equal(vma_policy(a), vma_policy(b)) &&
1102 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001103 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001104 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1105}
1106
1107/*
1108 * Do some basic sanity checking to see if we can re-use the anon_vma
1109 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1110 * the same as 'old', the other will be the new one that is trying
1111 * to share the anon_vma.
1112 *
1113 * NOTE! This runs with mm_sem held for reading, so it is possible that
1114 * the anon_vma of 'old' is concurrently in the process of being set up
1115 * by another page fault trying to merge _that_. But that's ok: if it
1116 * is being set up, that automatically means that it will be a singleton
1117 * acceptable for merging, so we can do all of this optimistically. But
1118 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1119 *
1120 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1121 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1122 * is to return an anon_vma that is "complex" due to having gone through
1123 * a fork).
1124 *
1125 * We also make sure that the two vma's are compatible (adjacent,
1126 * and with the same memory policies). That's all stable, even with just
1127 * a read lock on the mm_sem.
1128 */
1129static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1130{
1131 if (anon_vma_compatible(a, b)) {
1132 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1133
1134 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1135 return anon_vma;
1136 }
1137 return NULL;
1138}
1139
1140/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1142 * neighbouring vmas for a suitable anon_vma, before it goes off
1143 * to allocate a new anon_vma. It checks because a repetitive
1144 * sequence of mprotects and faults may otherwise lead to distinct
1145 * anon_vmas being allocated, preventing vma merge in subsequent
1146 * mprotect.
1147 */
1148struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1149{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001150 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
1153 near = vma->vm_next;
1154 if (!near)
1155 goto try_prev;
1156
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001157 anon_vma = reusable_anon_vma(near, vma, near);
1158 if (anon_vma)
1159 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001161 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 if (!near)
1163 goto none;
1164
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001165 anon_vma = reusable_anon_vma(near, near, vma);
1166 if (anon_vma)
1167 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168none:
1169 /*
1170 * There's no absolute need to look only at touching neighbours:
1171 * we could search further afield for "compatible" anon_vmas.
1172 * But it would probably just be a waste of time searching,
1173 * or lead to too many vmas hanging off the same anon_vma.
1174 * We're trying to allow mprotect remerging later on,
1175 * not trying to minimize memory used for anon_vmas.
1176 */
1177 return NULL;
1178}
1179
1180#ifdef CONFIG_PROC_FS
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001181void vm_stat_account(struct mm_struct *mm, unsigned long flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 struct file *file, long pages)
1183{
1184 const unsigned long stack_flags
1185 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1186
Huang Shijie44de9d02012-07-31 16:41:49 -07001187 mm->total_vm += pages;
1188
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 if (file) {
1190 mm->shared_vm += pages;
1191 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1192 mm->exec_vm += pages;
1193 } else if (flags & stack_flags)
1194 mm->stack_vm += pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195}
1196#endif /* CONFIG_PROC_FS */
1197
1198/*
Al Viro40401532012-02-13 03:58:52 +00001199 * If a hint addr is less than mmap_min_addr change hint to be as
1200 * low as possible but still greater than mmap_min_addr
1201 */
1202static inline unsigned long round_hint_to_min(unsigned long hint)
1203{
1204 hint &= PAGE_MASK;
1205 if (((void *)hint != NULL) &&
1206 (hint < mmap_min_addr))
1207 return PAGE_ALIGN(mmap_min_addr);
1208 return hint;
1209}
1210
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001211static inline int mlock_future_check(struct mm_struct *mm,
1212 unsigned long flags,
1213 unsigned long len)
1214{
1215 unsigned long locked, lock_limit;
1216
1217 /* mlock MCL_FUTURE? */
1218 if (flags & VM_LOCKED) {
1219 locked = len >> PAGE_SHIFT;
1220 locked += mm->locked_vm;
1221 lock_limit = rlimit(RLIMIT_MEMLOCK);
1222 lock_limit >>= PAGE_SHIFT;
1223 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1224 return -EAGAIN;
1225 }
1226 return 0;
1227}
1228
Al Viro40401532012-02-13 03:58:52 +00001229/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001230 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 */
1232
Al Viroe3fc6292012-05-30 20:08:42 -04001233unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 unsigned long len, unsigned long prot,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001235 unsigned long flags, unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001236 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
1238 struct mm_struct * mm = current->mm;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001239 vm_flags_t vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Michel Lespinasse41badc12013-02-22 16:32:47 -08001241 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 /*
1244 * Does the application expect PROT_READ to imply PROT_EXEC?
1245 *
1246 * (the exception is when the underlying filesystem is noexec
1247 * mounted, in which case we dont add PROT_EXEC.)
1248 */
1249 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001250 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 prot |= PROT_EXEC;
1252
1253 if (!len)
1254 return -EINVAL;
1255
Eric Paris7cd94142007-11-26 18:47:40 -05001256 if (!(flags & MAP_FIXED))
1257 addr = round_hint_to_min(addr);
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 /* Careful about overflows.. */
1260 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001261 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 return -ENOMEM;
1263
1264 /* offset overflow? */
1265 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1266 return -EOVERFLOW;
1267
1268 /* Too many mappings? */
1269 if (mm->map_count > sysctl_max_map_count)
1270 return -ENOMEM;
1271
1272 /* Obtain the address to map to. we verify (or select) it and ensure
1273 * that it represents a valid section of the address space.
1274 */
1275 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1276 if (addr & ~PAGE_MASK)
1277 return addr;
1278
1279 /* Do simple checking here so the lower-level routines won't have
1280 * to. we assume access permissions have been handled by the open
1281 * of the memory object, so we don't do any here.
1282 */
1283 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1284 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1285
Huang Shijiecdf7b342009-09-21 17:03:36 -07001286 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (!can_do_mlock())
1288 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001289
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001290 if (mlock_future_check(mm, vm_flags, len))
1291 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001294 struct inode *inode = file_inode(file);
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 switch (flags & MAP_TYPE) {
1297 case MAP_SHARED:
1298 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1299 return -EACCES;
1300
1301 /*
1302 * Make sure we don't allow writing to an append-only
1303 * file..
1304 */
1305 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1306 return -EACCES;
1307
1308 /*
1309 * Make sure there are no mandatory locks on the file.
1310 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001311 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 return -EAGAIN;
1313
1314 vm_flags |= VM_SHARED | VM_MAYSHARE;
1315 if (!(file->f_mode & FMODE_WRITE))
1316 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1317
1318 /* fall through */
1319 case MAP_PRIVATE:
1320 if (!(file->f_mode & FMODE_READ))
1321 return -EACCES;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001322 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001323 if (vm_flags & VM_EXEC)
1324 return -EPERM;
1325 vm_flags &= ~VM_MAYEXEC;
1326 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001327
Al Viro72c2d532013-09-22 16:27:52 -04001328 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001329 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001330 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1331 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 break;
1333
1334 default:
1335 return -EINVAL;
1336 }
1337 } else {
1338 switch (flags & MAP_TYPE) {
1339 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001340 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1341 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001342 /*
1343 * Ignore pgoff.
1344 */
1345 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 vm_flags |= VM_SHARED | VM_MAYSHARE;
1347 break;
1348 case MAP_PRIVATE:
1349 /*
1350 * Set pgoff according to addr for anon_vma.
1351 */
1352 pgoff = addr >> PAGE_SHIFT;
1353 break;
1354 default:
1355 return -EINVAL;
1356 }
1357 }
1358
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001359 /*
1360 * Set 'VM_NORESERVE' if we should not account for the
1361 * memory use of this mapping.
1362 */
1363 if (flags & MAP_NORESERVE) {
1364 /* We honor MAP_NORESERVE if allowed to overcommit */
1365 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1366 vm_flags |= VM_NORESERVE;
1367
1368 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1369 if (file && is_file_hugepages(file))
1370 vm_flags |= VM_NORESERVE;
1371 }
1372
1373 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001374 if (!IS_ERR_VALUE(addr) &&
1375 ((vm_flags & VM_LOCKED) ||
1376 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001377 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001378 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001379}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001380
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001381SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1382 unsigned long, prot, unsigned long, flags,
1383 unsigned long, fd, unsigned long, pgoff)
1384{
1385 struct file *file = NULL;
1386 unsigned long retval = -EBADF;
1387
1388 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001389 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001390 file = fget(fd);
1391 if (!file)
1392 goto out;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001393 if (is_file_hugepages(file))
1394 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001395 retval = -EINVAL;
1396 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1397 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001398 } else if (flags & MAP_HUGETLB) {
1399 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001400 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001401
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001402 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001403 if (!hs)
1404 return -EINVAL;
1405
1406 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001407 /*
1408 * VM_NORESERVE is used because the reservations will be
1409 * taken when vm_ops->mmap() is called
1410 * A dummy user value is used because we are not locking
1411 * memory so no accounting is necessary
1412 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001413 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001414 VM_NORESERVE,
1415 &user, HUGETLB_ANONHUGE_INODE,
1416 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001417 if (IS_ERR(file))
1418 return PTR_ERR(file);
1419 }
1420
1421 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1422
Al Viroeb36c582012-05-30 20:17:35 -04001423 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001424out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001425 if (file)
1426 fput(file);
1427out:
1428 return retval;
1429}
1430
Christoph Hellwiga4679372010-03-10 15:21:15 -08001431#ifdef __ARCH_WANT_SYS_OLD_MMAP
1432struct mmap_arg_struct {
1433 unsigned long addr;
1434 unsigned long len;
1435 unsigned long prot;
1436 unsigned long flags;
1437 unsigned long fd;
1438 unsigned long offset;
1439};
1440
1441SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1442{
1443 struct mmap_arg_struct a;
1444
1445 if (copy_from_user(&a, arg, sizeof(a)))
1446 return -EFAULT;
1447 if (a.offset & ~PAGE_MASK)
1448 return -EINVAL;
1449
1450 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1451 a.offset >> PAGE_SHIFT);
1452}
1453#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1454
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001455/*
1456 * Some shared mappigns will want the pages marked read-only
1457 * to track write events. If so, we'll downgrade vm_page_prot
1458 * to the private version (using protection_map[] without the
1459 * VM_SHARED bit).
1460 */
1461int vma_wants_writenotify(struct vm_area_struct *vma)
1462{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001463 vm_flags_t vm_flags = vma->vm_flags;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001464
1465 /* If it was private or non-writable, the write bit is already clear */
1466 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1467 return 0;
1468
1469 /* The backer wishes to know when pages are first written to? */
1470 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1471 return 1;
1472
1473 /* The open routine did something to the protections already? */
1474 if (pgprot_val(vma->vm_page_prot) !=
Coly Li3ed75eb2007-10-18 23:39:15 -07001475 pgprot_val(vm_get_page_prot(vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001476 return 0;
1477
1478 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001479 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001480 return 0;
1481
1482 /* Can the mapping track the dirty pages? */
1483 return vma->vm_file && vma->vm_file->f_mapping &&
1484 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1485}
1486
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001487/*
1488 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001489 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001490 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001491static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001492{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001493 /*
1494 * hugetlb has its own accounting separate from the core VM
1495 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1496 */
1497 if (file && is_file_hugepages(file))
1498 return 0;
1499
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001500 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1501}
1502
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001503unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001504 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001505{
1506 struct mm_struct *mm = current->mm;
1507 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001508 int error;
1509 struct rb_node **rb_link, *rb_parent;
1510 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001511
Cyril Hrubise8420a82013-04-29 15:08:33 -07001512 /* Check against address space limit. */
1513 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1514 unsigned long nr_pages;
1515
1516 /*
1517 * MAP_FIXED may remove pages of mappings that intersects with
1518 * requested mapping. Account for the pages it would unmap.
1519 */
1520 if (!(vm_flags & MAP_FIXED))
1521 return -ENOMEM;
1522
1523 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1524
1525 if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1526 return -ENOMEM;
1527 }
1528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 /* Clear old maps */
1530 error = -ENOMEM;
1531munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07001532 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 if (do_munmap(mm, addr, len))
1534 return -ENOMEM;
1535 goto munmap_back;
1536 }
1537
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001538 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001539 * Private writable mapping: check memory availability
1540 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001541 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001542 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001543 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001544 return -ENOMEM;
1545 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 }
1547
1548 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001549 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 */
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001551 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1552 if (vma)
1553 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
1555 /*
1556 * Determine the object being mapped and call the appropriate
1557 * specific mapper. the address has already been validated, but
1558 * not unmapped, but the maps are removed from the list.
1559 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001560 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (!vma) {
1562 error = -ENOMEM;
1563 goto unacct_error;
1564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
1566 vma->vm_mm = mm;
1567 vma->vm_start = addr;
1568 vma->vm_end = addr + len;
1569 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001570 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001572 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
1574 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 if (vm_flags & VM_DENYWRITE) {
1576 error = deny_write_access(file);
1577 if (error)
1578 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001580 if (vm_flags & VM_SHARED) {
1581 error = mapping_map_writable(file->f_mapping);
1582 if (error)
1583 goto allow_write_and_free_vma;
1584 }
1585
1586 /* ->mmap() can change vma->vm_file, but must guarantee that
1587 * vma_link() below can deny write-access if VM_DENYWRITE is set
1588 * and map writably if VM_SHARED is set. This usually means the
1589 * new file must not have been exposed to user-space, yet.
1590 */
Al Virocb0942b2012-08-27 14:48:26 -04001591 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 error = file->f_op->mmap(file, vma);
1593 if (error)
1594 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001595
1596 /* Can addr have changed??
1597 *
1598 * Answer: Yes, several device drivers can do it in their
1599 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001600 * Bug: If addr is changed, prev, rb_link, rb_parent should
1601 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001602 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001603 WARN_ON_ONCE(addr != vma->vm_start);
1604
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001605 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001606 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 } else if (vm_flags & VM_SHARED) {
1608 error = shmem_zero_setup(vma);
1609 if (error)
1610 goto free_vma;
1611 }
1612
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001613 if (vma_wants_writenotify(vma)) {
1614 pgprot_t pprot = vma->vm_page_prot;
1615
1616 /* Can vma->vm_page_prot have changed??
1617 *
1618 * Answer: Yes, drivers may have changed it in their
1619 * f_op->mmap method.
1620 *
1621 * Ensures that vmas marked as uncached stay that way.
1622 */
Hugh Dickins1ddd4392007-10-22 20:45:12 -07001623 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
Magnus Dammc9d0bf22009-12-14 17:59:49 -08001624 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1625 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1626 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001627
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001628 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001629 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001630 if (file) {
1631 if (vm_flags & VM_SHARED)
1632 mapping_unmap_writable(file->f_mapping);
1633 if (vm_flags & VM_DENYWRITE)
1634 allow_write_access(file);
1635 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001636 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001637out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001638 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001639
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07001640 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001642 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1643 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001644 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001645 else
1646 vma->vm_flags &= ~VM_LOCKED;
1647 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301648
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001649 if (file)
1650 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301651
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001652 /*
1653 * New (or expanded) vma always get soft dirty status.
1654 * Otherwise user-space soft-dirty page tracker won't
1655 * be able to distinguish situation when vma area unmapped,
1656 * then new mapped in-place (which must be aimed as
1657 * a completely new data area).
1658 */
1659 vma->vm_flags |= VM_SOFTDIRTY;
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return addr;
1662
1663unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 vma->vm_file = NULL;
1665 fput(file);
1666
1667 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001668 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1669 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001670 if (vm_flags & VM_SHARED)
1671 mapping_unmap_writable(file->f_mapping);
1672allow_write_and_free_vma:
1673 if (vm_flags & VM_DENYWRITE)
1674 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675free_vma:
1676 kmem_cache_free(vm_area_cachep, vma);
1677unacct_error:
1678 if (charged)
1679 vm_unacct_memory(charged);
1680 return error;
1681}
1682
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001683unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1684{
1685 /*
1686 * We implement the search by looking for an rbtree node that
1687 * immediately follows a suitable gap. That is,
1688 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1689 * - gap_end = vma->vm_start >= info->low_limit + length;
1690 * - gap_end - gap_start >= length
1691 */
1692
1693 struct mm_struct *mm = current->mm;
1694 struct vm_area_struct *vma;
1695 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1696
1697 /* Adjust search length to account for worst case alignment overhead */
1698 length = info->length + info->align_mask;
1699 if (length < info->length)
1700 return -ENOMEM;
1701
1702 /* Adjust search limits by the desired length */
1703 if (info->high_limit < length)
1704 return -ENOMEM;
1705 high_limit = info->high_limit - length;
1706
1707 if (info->low_limit > high_limit)
1708 return -ENOMEM;
1709 low_limit = info->low_limit + length;
1710
1711 /* Check if rbtree root looks promising */
1712 if (RB_EMPTY_ROOT(&mm->mm_rb))
1713 goto check_highest;
1714 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1715 if (vma->rb_subtree_gap < length)
1716 goto check_highest;
1717
1718 while (true) {
1719 /* Visit left subtree if it looks promising */
1720 gap_end = vma->vm_start;
1721 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1722 struct vm_area_struct *left =
1723 rb_entry(vma->vm_rb.rb_left,
1724 struct vm_area_struct, vm_rb);
1725 if (left->rb_subtree_gap >= length) {
1726 vma = left;
1727 continue;
1728 }
1729 }
1730
1731 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1732check_current:
1733 /* Check if current node has a suitable gap */
1734 if (gap_start > high_limit)
1735 return -ENOMEM;
1736 if (gap_end >= low_limit && gap_end - gap_start >= length)
1737 goto found;
1738
1739 /* Visit right subtree if it looks promising */
1740 if (vma->vm_rb.rb_right) {
1741 struct vm_area_struct *right =
1742 rb_entry(vma->vm_rb.rb_right,
1743 struct vm_area_struct, vm_rb);
1744 if (right->rb_subtree_gap >= length) {
1745 vma = right;
1746 continue;
1747 }
1748 }
1749
1750 /* Go back up the rbtree to find next candidate node */
1751 while (true) {
1752 struct rb_node *prev = &vma->vm_rb;
1753 if (!rb_parent(prev))
1754 goto check_highest;
1755 vma = rb_entry(rb_parent(prev),
1756 struct vm_area_struct, vm_rb);
1757 if (prev == vma->vm_rb.rb_left) {
1758 gap_start = vma->vm_prev->vm_end;
1759 gap_end = vma->vm_start;
1760 goto check_current;
1761 }
1762 }
1763 }
1764
1765check_highest:
1766 /* Check highest gap, which does not precede any rbtree node */
1767 gap_start = mm->highest_vm_end;
1768 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1769 if (gap_start > high_limit)
1770 return -ENOMEM;
1771
1772found:
1773 /* We found a suitable gap. Clip it with the original low_limit. */
1774 if (gap_start < info->low_limit)
1775 gap_start = info->low_limit;
1776
1777 /* Adjust gap address to the desired alignment */
1778 gap_start += (info->align_offset - gap_start) & info->align_mask;
1779
1780 VM_BUG_ON(gap_start + info->length > info->high_limit);
1781 VM_BUG_ON(gap_start + info->length > gap_end);
1782 return gap_start;
1783}
1784
1785unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1786{
1787 struct mm_struct *mm = current->mm;
1788 struct vm_area_struct *vma;
1789 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1790
1791 /* Adjust search length to account for worst case alignment overhead */
1792 length = info->length + info->align_mask;
1793 if (length < info->length)
1794 return -ENOMEM;
1795
1796 /*
1797 * Adjust search limits by the desired length.
1798 * See implementation comment at top of unmapped_area().
1799 */
1800 gap_end = info->high_limit;
1801 if (gap_end < length)
1802 return -ENOMEM;
1803 high_limit = gap_end - length;
1804
1805 if (info->low_limit > high_limit)
1806 return -ENOMEM;
1807 low_limit = info->low_limit + length;
1808
1809 /* Check highest gap, which does not precede any rbtree node */
1810 gap_start = mm->highest_vm_end;
1811 if (gap_start <= high_limit)
1812 goto found_highest;
1813
1814 /* Check if rbtree root looks promising */
1815 if (RB_EMPTY_ROOT(&mm->mm_rb))
1816 return -ENOMEM;
1817 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1818 if (vma->rb_subtree_gap < length)
1819 return -ENOMEM;
1820
1821 while (true) {
1822 /* Visit right subtree if it looks promising */
1823 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1824 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1825 struct vm_area_struct *right =
1826 rb_entry(vma->vm_rb.rb_right,
1827 struct vm_area_struct, vm_rb);
1828 if (right->rb_subtree_gap >= length) {
1829 vma = right;
1830 continue;
1831 }
1832 }
1833
1834check_current:
1835 /* Check if current node has a suitable gap */
1836 gap_end = vma->vm_start;
1837 if (gap_end < low_limit)
1838 return -ENOMEM;
1839 if (gap_start <= high_limit && gap_end - gap_start >= length)
1840 goto found;
1841
1842 /* Visit left subtree if it looks promising */
1843 if (vma->vm_rb.rb_left) {
1844 struct vm_area_struct *left =
1845 rb_entry(vma->vm_rb.rb_left,
1846 struct vm_area_struct, vm_rb);
1847 if (left->rb_subtree_gap >= length) {
1848 vma = left;
1849 continue;
1850 }
1851 }
1852
1853 /* Go back up the rbtree to find next candidate node */
1854 while (true) {
1855 struct rb_node *prev = &vma->vm_rb;
1856 if (!rb_parent(prev))
1857 return -ENOMEM;
1858 vma = rb_entry(rb_parent(prev),
1859 struct vm_area_struct, vm_rb);
1860 if (prev == vma->vm_rb.rb_right) {
1861 gap_start = vma->vm_prev ?
1862 vma->vm_prev->vm_end : 0;
1863 goto check_current;
1864 }
1865 }
1866 }
1867
1868found:
1869 /* We found a suitable gap. Clip it with the original high_limit. */
1870 if (gap_end > info->high_limit)
1871 gap_end = info->high_limit;
1872
1873found_highest:
1874 /* Compute highest gap address at the desired alignment */
1875 gap_end -= info->length;
1876 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1877
1878 VM_BUG_ON(gap_end < info->low_limit);
1879 VM_BUG_ON(gap_end < gap_start);
1880 return gap_end;
1881}
1882
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883/* Get an address range which is currently unmapped.
1884 * For shmat() with addr=0.
1885 *
1886 * Ugly calling convention alert:
1887 * Return value with the low bits set means error value,
1888 * ie
1889 * if (ret & ~PAGE_MASK)
1890 * error = ret;
1891 *
1892 * This function "knows" that -ENOMEM has the bits set.
1893 */
1894#ifndef HAVE_ARCH_UNMAPPED_AREA
1895unsigned long
1896arch_get_unmapped_area(struct file *filp, unsigned long addr,
1897 unsigned long len, unsigned long pgoff, unsigned long flags)
1898{
1899 struct mm_struct *mm = current->mm;
1900 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001901 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001903 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 return -ENOMEM;
1905
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001906 if (flags & MAP_FIXED)
1907 return addr;
1908
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (addr) {
1910 addr = PAGE_ALIGN(addr);
1911 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001912 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 (!vma || addr + len <= vma->vm_start))
1914 return addr;
1915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001917 info.flags = 0;
1918 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001919 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001920 info.high_limit = TASK_SIZE;
1921 info.align_mask = 0;
1922 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923}
1924#endif
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926/*
1927 * This mmap-allocator allocates new areas top-down from below the
1928 * stack's low limit (the base):
1929 */
1930#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1931unsigned long
1932arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1933 const unsigned long len, const unsigned long pgoff,
1934 const unsigned long flags)
1935{
1936 struct vm_area_struct *vma;
1937 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001938 unsigned long addr = addr0;
1939 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
1941 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001942 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 return -ENOMEM;
1944
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001945 if (flags & MAP_FIXED)
1946 return addr;
1947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 /* requesting a specific address */
1949 if (addr) {
1950 addr = PAGE_ALIGN(addr);
1951 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001952 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 (!vma || addr + len <= vma->vm_start))
1954 return addr;
1955 }
1956
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001957 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1958 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001959 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001960 info.high_limit = mm->mmap_base;
1961 info.align_mask = 0;
1962 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001963
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 /*
1965 * A failed mmap() very likely causes application failure,
1966 * so fall back to the bottom-up function here. This scenario
1967 * can happen with large stack limits and large mmap()
1968 * allocations.
1969 */
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001970 if (addr & ~PAGE_MASK) {
1971 VM_BUG_ON(addr != -ENOMEM);
1972 info.flags = 0;
1973 info.low_limit = TASK_UNMAPPED_BASE;
1974 info.high_limit = TASK_SIZE;
1975 addr = vm_unmapped_area(&info);
1976 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
1978 return addr;
1979}
1980#endif
1981
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982unsigned long
1983get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1984 unsigned long pgoff, unsigned long flags)
1985{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001986 unsigned long (*get_area)(struct file *, unsigned long,
1987 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Al Viro9206de92009-12-03 15:23:11 -05001989 unsigned long error = arch_mmap_check(addr, len, flags);
1990 if (error)
1991 return error;
1992
1993 /* Careful about overflows.. */
1994 if (len > TASK_SIZE)
1995 return -ENOMEM;
1996
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001997 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04001998 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001999 get_area = file->f_op->get_unmapped_area;
2000 addr = get_area(file, addr, len, pgoff, flags);
2001 if (IS_ERR_VALUE(addr))
2002 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002004 if (addr > TASK_SIZE - len)
2005 return -ENOMEM;
2006 if (addr & ~PAGE_MASK)
2007 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002008
Al Viro9ac4ed42012-05-30 17:13:15 -04002009 addr = arch_rebalance_pgtables(addr, len);
2010 error = security_mmap_addr(addr);
2011 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012}
2013
2014EXPORT_SYMBOL(get_unmapped_area);
2015
2016/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002017struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002019 struct rb_node *rb_node;
2020 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002022 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002023 vma = vmacache_find(mm, addr);
2024 if (likely(vma))
2025 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002027 rb_node = mm->mm_rb.rb_node;
2028 vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002030 while (rb_node) {
2031 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002033 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002034
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002035 if (tmp->vm_end > addr) {
2036 vma = tmp;
2037 if (tmp->vm_start <= addr)
2038 break;
2039 rb_node = rb_node->rb_left;
2040 } else
2041 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002043
2044 if (vma)
2045 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 return vma;
2047}
2048
2049EXPORT_SYMBOL(find_vma);
2050
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002051/*
2052 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002053 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054struct vm_area_struct *
2055find_vma_prev(struct mm_struct *mm, unsigned long addr,
2056 struct vm_area_struct **pprev)
2057{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002058 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002060 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002061 if (vma) {
2062 *pprev = vma->vm_prev;
2063 } else {
2064 struct rb_node *rb_node = mm->mm_rb.rb_node;
2065 *pprev = NULL;
2066 while (rb_node) {
2067 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2068 rb_node = rb_node->rb_right;
2069 }
2070 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002071 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072}
2073
2074/*
2075 * Verify that the stack growth is acceptable and
2076 * update accounting. This is shared with both the
2077 * grow-up and grow-down cases.
2078 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002079static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080{
2081 struct mm_struct *mm = vma->vm_mm;
2082 struct rlimit *rlim = current->signal->rlim;
Adam Litke0d59a012007-01-30 14:35:39 -08002083 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084
2085 /* address space limit tests */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002086 if (!may_expand_vm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 return -ENOMEM;
2088
2089 /* Stack limit test */
Jiri Slaby59e99e52010-03-05 13:41:44 -08002090 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 return -ENOMEM;
2092
2093 /* mlock limit tests */
2094 if (vma->vm_flags & VM_LOCKED) {
2095 unsigned long locked;
2096 unsigned long limit;
2097 locked = mm->locked_vm + grow;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002098 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2099 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 if (locked > limit && !capable(CAP_IPC_LOCK))
2101 return -ENOMEM;
2102 }
2103
Adam Litke0d59a012007-01-30 14:35:39 -08002104 /* Check to ensure the stack will not grow into a hugetlb-only region */
2105 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2106 vma->vm_end - size;
2107 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2108 return -EFAULT;
2109
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 /*
2111 * Overcommit.. This must be the final test, as it will
2112 * update security statistics.
2113 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002114 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 return -ENOMEM;
2116
2117 /* Ok, everything looks good - let it rip */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 if (vma->vm_flags & VM_LOCKED)
2119 mm->locked_vm += grow;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -07002120 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 return 0;
2122}
2123
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002124#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002126 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2127 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002129int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130{
2131 int error;
2132
2133 if (!(vma->vm_flags & VM_GROWSUP))
2134 return -EFAULT;
2135
2136 /*
2137 * We must make sure the anon_vma is allocated
2138 * so that the anon_vma locking is not a noop.
2139 */
2140 if (unlikely(anon_vma_prepare(vma)))
2141 return -ENOMEM;
Rik van Rielbb4a3402010-08-09 17:18:37 -07002142 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143
2144 /*
2145 * vma->vm_start/vm_end cannot change under us because the caller
2146 * is required to hold the mmap_sem in read mode. We need the
2147 * anon_vma lock to serialize against concurrent expand_stacks.
Helge Deller06b32f32006-12-19 19:28:33 +01002148 * Also guard against wrapping around to address 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 */
Helge Deller06b32f32006-12-19 19:28:33 +01002150 if (address < PAGE_ALIGN(address+4))
2151 address = PAGE_ALIGN(address+4);
2152 else {
Rik van Rielbb4a3402010-08-09 17:18:37 -07002153 vma_unlock_anon_vma(vma);
Helge Deller06b32f32006-12-19 19:28:33 +01002154 return -ENOMEM;
2155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 error = 0;
2157
2158 /* Somebody else might have raced and expanded it already */
2159 if (address > vma->vm_end) {
2160 unsigned long size, grow;
2161
2162 size = address - vma->vm_start;
2163 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2164
Hugh Dickins42c36f62011-05-09 17:44:42 -07002165 error = -ENOMEM;
2166 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2167 error = acct_stack_growth(vma, size, grow);
2168 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002169 /*
2170 * vma_gap_update() doesn't support concurrent
2171 * updates, but we only hold a shared mmap_sem
2172 * lock here, so we need to protect against
2173 * concurrent vma expansions.
2174 * vma_lock_anon_vma() doesn't help here, as
2175 * we don't guarantee that all growable vmas
2176 * in a mm share the same root anon vma.
2177 * So, we reuse mm->page_table_lock to guard
2178 * against concurrent vma expansions.
2179 */
2180 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002181 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002182 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002183 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002184 if (vma->vm_next)
2185 vma_gap_update(vma->vm_next);
2186 else
2187 vma->vm_mm->highest_vm_end = address;
Michel Lespinasse41289972012-12-12 13:52:25 -08002188 spin_unlock(&vma->vm_mm->page_table_lock);
2189
Hugh Dickins42c36f62011-05-09 17:44:42 -07002190 perf_event_mmap(vma);
2191 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002194 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002195 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002196 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 return error;
2198}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002199#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201/*
2202 * vma is the first one with address < vma->vm_start. Have to extend vma.
2203 */
Michal Hockod05f3162011-05-24 17:11:44 -07002204int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002205 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206{
2207 int error;
2208
2209 /*
2210 * We must make sure the anon_vma is allocated
2211 * so that the anon_vma locking is not a noop.
2212 */
2213 if (unlikely(anon_vma_prepare(vma)))
2214 return -ENOMEM;
Eric Paris88694772007-11-26 18:47:26 -05002215
2216 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002217 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002218 if (error)
2219 return error;
2220
Rik van Rielbb4a3402010-08-09 17:18:37 -07002221 vma_lock_anon_vma(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
2223 /*
2224 * vma->vm_start/vm_end cannot change under us because the caller
2225 * is required to hold the mmap_sem in read mode. We need the
2226 * anon_vma lock to serialize against concurrent expand_stacks.
2227 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
2229 /* Somebody else might have raced and expanded it already */
2230 if (address < vma->vm_start) {
2231 unsigned long size, grow;
2232
2233 size = vma->vm_end - address;
2234 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2235
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002236 error = -ENOMEM;
2237 if (grow <= vma->vm_pgoff) {
2238 error = acct_stack_growth(vma, size, grow);
2239 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002240 /*
2241 * vma_gap_update() doesn't support concurrent
2242 * updates, but we only hold a shared mmap_sem
2243 * lock here, so we need to protect against
2244 * concurrent vma expansions.
2245 * vma_lock_anon_vma() doesn't help here, as
2246 * we don't guarantee that all growable vmas
2247 * in a mm share the same root anon vma.
2248 * So, we reuse mm->page_table_lock to guard
2249 * against concurrent vma expansions.
2250 */
2251 spin_lock(&vma->vm_mm->page_table_lock);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002252 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002253 vma->vm_start = address;
2254 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002255 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002256 vma_gap_update(vma);
Michel Lespinasse41289972012-12-12 13:52:25 -08002257 spin_unlock(&vma->vm_mm->page_table_lock);
2258
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002259 perf_event_mmap(vma);
2260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 }
2262 }
Rik van Rielbb4a3402010-08-09 17:18:37 -07002263 vma_unlock_anon_vma(vma);
Andrea Arcangelib15d00b2011-01-13 15:46:59 -08002264 khugepaged_enter_vma_merge(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -07002265 validate_mm(vma->vm_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 return error;
2267}
2268
Linus Torvalds09884962013-02-27 08:36:04 -08002269/*
2270 * Note how expand_stack() refuses to expand the stack all the way to
2271 * abut the next virtual mapping, *unless* that mapping itself is also
2272 * a stack mapping. We want to leave room for a guard page, after all
2273 * (the guard page itself is not added here, that is done by the
2274 * actual page faulting logic)
2275 *
2276 * This matches the behavior of the guard page logic (see mm/memory.c:
2277 * check_stack_guard_page()), which only allows the guard page to be
2278 * removed under these circumstances.
2279 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002280#ifdef CONFIG_STACK_GROWSUP
2281int expand_stack(struct vm_area_struct *vma, unsigned long address)
2282{
Linus Torvalds09884962013-02-27 08:36:04 -08002283 struct vm_area_struct *next;
2284
2285 address &= PAGE_MASK;
2286 next = vma->vm_next;
2287 if (next && next->vm_start == address + PAGE_SIZE) {
2288 if (!(next->vm_flags & VM_GROWSUP))
2289 return -ENOMEM;
2290 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002291 return expand_upwards(vma, address);
2292}
2293
2294struct vm_area_struct *
2295find_extend_vma(struct mm_struct *mm, unsigned long addr)
2296{
2297 struct vm_area_struct *vma, *prev;
2298
2299 addr &= PAGE_MASK;
2300 vma = find_vma_prev(mm, addr, &prev);
2301 if (vma && (vma->vm_start <= addr))
2302 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002303 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002304 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002305 if (prev->vm_flags & VM_LOCKED)
2306 __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002307 return prev;
2308}
2309#else
2310int expand_stack(struct vm_area_struct *vma, unsigned long address)
2311{
Linus Torvalds09884962013-02-27 08:36:04 -08002312 struct vm_area_struct *prev;
2313
2314 address &= PAGE_MASK;
2315 prev = vma->vm_prev;
2316 if (prev && prev->vm_end == address) {
2317 if (!(prev->vm_flags & VM_GROWSDOWN))
2318 return -ENOMEM;
2319 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002320 return expand_downwards(vma, address);
2321}
2322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323struct vm_area_struct *
2324find_extend_vma(struct mm_struct * mm, unsigned long addr)
2325{
2326 struct vm_area_struct * vma;
2327 unsigned long start;
2328
2329 addr &= PAGE_MASK;
2330 vma = find_vma(mm,addr);
2331 if (!vma)
2332 return NULL;
2333 if (vma->vm_start <= addr)
2334 return vma;
2335 if (!(vma->vm_flags & VM_GROWSDOWN))
2336 return NULL;
2337 start = vma->vm_start;
2338 if (expand_stack(vma, addr))
2339 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002340 if (vma->vm_flags & VM_LOCKED)
2341 __mlock_vma_pages_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 return vma;
2343}
2344#endif
2345
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002347 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002349 *
2350 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002352static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002354 unsigned long nr_accounted = 0;
2355
Hugh Dickins365e9c872005-10-29 18:16:18 -07002356 /* Update high watermark before we lower total_vm */
2357 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002359 long nrpages = vma_pages(vma);
2360
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002361 if (vma->vm_flags & VM_ACCOUNT)
2362 nr_accounted += nrpages;
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002363 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002364 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002365 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002366 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 validate_mm(mm);
2368}
2369
2370/*
2371 * Get rid of page table information in the indicated region.
2372 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002373 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 */
2375static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002376 struct vm_area_struct *vma, struct vm_area_struct *prev,
2377 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378{
Hugh Dickinse0da3822005-04-19 13:29:15 -07002379 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002380 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
2382 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002383 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002384 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002385 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002386 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002387 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002388 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389}
2390
2391/*
2392 * Create a list of vma's touched by the unmap, removing them from the mm's
2393 * vma list as we go..
2394 */
2395static void
2396detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2397 struct vm_area_struct *prev, unsigned long end)
2398{
2399 struct vm_area_struct **insertion_point;
2400 struct vm_area_struct *tail_vma = NULL;
2401
2402 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002403 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002405 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 mm->map_count--;
2407 tail_vma = vma;
2408 vma = vma->vm_next;
2409 } while (vma && vma->vm_start < end);
2410 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002411 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002412 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002413 vma_gap_update(vma);
2414 } else
2415 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002417
2418 /* Kill the cache */
2419 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420}
2421
2422/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002423 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2424 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 */
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002426static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 unsigned long addr, int new_below)
2428{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 struct vm_area_struct *new;
Rik van Riel5beb4932010-03-05 13:42:07 -08002430 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
Andi Kleena5516432008-07-23 21:27:41 -07002432 if (is_vm_hugetlb_page(vma) && (addr &
2433 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 return -EINVAL;
2435
Christoph Lametere94b1762006-12-06 20:33:17 -08002436 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 if (!new)
Rik van Riel5beb4932010-03-05 13:42:07 -08002438 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
2440 /* most fields are the same, copy all, and then fixup */
2441 *new = *vma;
2442
Rik van Riel5beb4932010-03-05 13:42:07 -08002443 INIT_LIST_HEAD(&new->anon_vma_chain);
2444
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 if (new_below)
2446 new->vm_end = addr;
2447 else {
2448 new->vm_start = addr;
2449 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2450 }
2451
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002452 err = vma_dup_policy(vma, new);
2453 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002454 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
Rik van Riel5beb4932010-03-05 13:42:07 -08002456 if (anon_vma_clone(new, vma))
2457 goto out_free_mpol;
2458
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002459 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 get_file(new->vm_file);
2461
2462 if (new->vm_ops && new->vm_ops->open)
2463 new->vm_ops->open(new);
2464
2465 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002466 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2468 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002469 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
Rik van Riel5beb4932010-03-05 13:42:07 -08002471 /* Success. */
2472 if (!err)
2473 return 0;
2474
2475 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002476 if (new->vm_ops && new->vm_ops->close)
2477 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002478 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002479 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002480 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002481 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002482 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002483 out_free_vma:
2484 kmem_cache_free(vm_area_cachep, new);
2485 out_err:
2486 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487}
2488
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002489/*
2490 * Split a vma into two pieces at address 'addr', a new vma is allocated
2491 * either for the first part or the tail.
2492 */
2493int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2494 unsigned long addr, int new_below)
2495{
2496 if (mm->map_count >= sysctl_max_map_count)
2497 return -ENOMEM;
2498
2499 return __split_vma(mm, vma, addr, new_below);
2500}
2501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502/* Munmap is split into 2 main parts -- this part which finds
2503 * what needs doing, and the areas themselves, which do the
2504 * work. This now handles partial unmappings.
2505 * Jeremy Fitzhardinge <jeremy@goop.org>
2506 */
2507int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2508{
2509 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002510 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
2512 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2513 return -EINVAL;
2514
2515 if ((len = PAGE_ALIGN(len)) == 0)
2516 return -EINVAL;
2517
2518 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002519 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002520 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002522 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002523 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
2525 /* if it doesn't overlap, we have nothing.. */
2526 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002527 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 return 0;
2529
2530 /*
2531 * If we need to split any vma, do it now to save pain later.
2532 *
2533 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2534 * unmapped vm_area_struct will remain in use: so lower split_vma
2535 * places tmp vma above, and higher split_vma places tmp vma below.
2536 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002537 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002538 int error;
2539
2540 /*
2541 * Make sure that map_count on return from munmap() will
2542 * not exceed its limit; but let map_count go just above
2543 * its limit temporarily, to help free resources as expected.
2544 */
2545 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2546 return -ENOMEM;
2547
2548 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 if (error)
2550 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002551 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 }
2553
2554 /* Does it split the last one? */
2555 last = find_vma(mm, end);
2556 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002557 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 if (error)
2559 return error;
2560 }
Hugh Dickins146425a2005-04-19 13:29:18 -07002561 vma = prev? prev->vm_next: mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
2563 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002564 * unlock any mlock()ed ranges before detaching vmas
2565 */
2566 if (mm->locked_vm) {
2567 struct vm_area_struct *tmp = vma;
2568 while (tmp && tmp->vm_start < end) {
2569 if (tmp->vm_flags & VM_LOCKED) {
2570 mm->locked_vm -= vma_pages(tmp);
2571 munlock_vma_pages_all(tmp);
2572 }
2573 tmp = tmp->vm_next;
2574 }
2575 }
2576
2577 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 * Remove the vma's, and unmap the actual pages
2579 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002580 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2581 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
2583 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002584 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
2586 return 0;
2587}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
Al Virobfce2812012-04-20 21:57:04 -04002589int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002590{
2591 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002592 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002593
2594 down_write(&mm->mmap_sem);
2595 ret = do_munmap(mm, start, len);
2596 up_write(&mm->mmap_sem);
2597 return ret;
2598}
2599EXPORT_SYMBOL(vm_munmap);
2600
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002601SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002604 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605}
2606
2607static inline void verify_mm_writelocked(struct mm_struct *mm)
2608{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002609#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2611 WARN_ON(1);
2612 up_read(&mm->mmap_sem);
2613 }
2614#endif
2615}
2616
2617/*
2618 * this is really a simplified "do_mmap". it only handles
2619 * anonymous maps. eventually we may be able to do some
2620 * brk-specific accounting here.
2621 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002622static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623{
2624 struct mm_struct * mm = current->mm;
2625 struct vm_area_struct * vma, * prev;
2626 unsigned long flags;
2627 struct rb_node ** rb_link, * rb_parent;
2628 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002629 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
2631 len = PAGE_ALIGN(len);
2632 if (!len)
2633 return addr;
2634
Kirill Korotaev3a459752006-09-07 14:17:04 +04002635 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2636
Al Viro2c6a1012009-12-03 19:40:46 -05002637 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2638 if (error & ~PAGE_MASK)
Kirill Korotaev3a459752006-09-07 14:17:04 +04002639 return error;
2640
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002641 error = mlock_future_check(mm, mm->def_flags, len);
2642 if (error)
2643 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
2645 /*
2646 * mm->mmap_sem is required to protect against another thread
2647 * changing the mappings in case we sleep.
2648 */
2649 verify_mm_writelocked(mm);
2650
2651 /*
2652 * Clear old maps. this also does some error checking for us
2653 */
2654 munmap_back:
Hugh Dickins6597d782012-10-08 16:29:07 -07002655 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 if (do_munmap(mm, addr, len))
2657 return -ENOMEM;
2658 goto munmap_back;
2659 }
2660
2661 /* Check against address space limits *after* clearing old maps... */
akpm@osdl.org119f6572005-05-01 08:58:35 -07002662 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 return -ENOMEM;
2664
2665 if (mm->map_count > sysctl_max_map_count)
2666 return -ENOMEM;
2667
Al Viro191c5422012-02-13 03:58:52 +00002668 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 return -ENOMEM;
2670
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002672 vma = vma_merge(mm, prev, addr, addr + len, flags,
2673 NULL, NULL, pgoff, NULL);
2674 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 goto out;
2676
2677 /*
2678 * create a vma struct for an anonymous mapping
2679 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002680 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 if (!vma) {
2682 vm_unacct_memory(len >> PAGE_SHIFT);
2683 return -ENOMEM;
2684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685
Rik van Riel5beb4932010-03-05 13:42:07 -08002686 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 vma->vm_mm = mm;
2688 vma->vm_start = addr;
2689 vma->vm_end = addr + len;
2690 vma->vm_pgoff = pgoff;
2691 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002692 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 vma_link(mm, vma, prev, rb_link, rb_parent);
2694out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002695 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 mm->total_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002697 if (flags & VM_LOCKED)
2698 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002699 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 return addr;
2701}
2702
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002703unsigned long vm_brk(unsigned long addr, unsigned long len)
2704{
2705 struct mm_struct *mm = current->mm;
2706 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002707 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002708
2709 down_write(&mm->mmap_sem);
2710 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002711 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002712 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002713 if (populate)
2714 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002715 return ret;
2716}
2717EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
2719/* Release all mmaps. */
2720void exit_mmap(struct mm_struct *mm)
2721{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002722 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002723 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 unsigned long nr_accounted = 0;
2725
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002726 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002727 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002728
Rik van Rielba470de2008-10-18 20:26:50 -07002729 if (mm->locked_vm) {
2730 vma = mm->mmap;
2731 while (vma) {
2732 if (vma->vm_flags & VM_LOCKED)
2733 munlock_vma_pages_all(vma);
2734 vma = vma->vm_next;
2735 }
2736 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002737
2738 arch_exit_mmap(mm);
2739
Rik van Rielba470de2008-10-18 20:26:50 -07002740 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002741 if (!vma) /* Can happen if dup_mmap() received an OOM */
2742 return;
2743
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002746 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002747 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002748 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002749 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002750
Hugh Dickins6ee86302013-04-29 15:07:44 -07002751 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002752 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002755 * Walk the list again, actually closing and freeing it,
2756 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002758 while (vma) {
2759 if (vma->vm_flags & VM_ACCOUNT)
2760 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002761 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002762 }
2763 vm_unacct_memory(nr_accounted);
Hugh Dickinse0da3822005-04-19 13:29:15 -07002764
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -08002765 WARN_ON(atomic_long_read(&mm->nr_ptes) >
2766 (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767}
2768
2769/* Insert vm structure into process list sorted by address
2770 * and into the inode's i_mmap tree. If vm_file is non-NULL
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002771 * then i_mmap_mutex is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002773int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774{
Hugh Dickins6597d782012-10-08 16:29:07 -07002775 struct vm_area_struct *prev;
2776 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
2778 /*
2779 * The vm_pgoff of a purely anonymous vma should be irrelevant
2780 * until its first write fault, when page's anon_vma and index
2781 * are set. But now set the vm_pgoff it will almost certainly
2782 * end up with (unless mremap moves it elsewhere before that
2783 * first wfault), so /proc/pid/maps tells a consistent story.
2784 *
2785 * By setting it to reflect the virtual start address of the
2786 * vma, merges and splits can happen in a seamless way, just
2787 * using the existing file pgoff checks and manipulations.
2788 * Similarly in do_mmap_pgoff and in do_brk.
2789 */
2790 if (!vma->vm_file) {
2791 BUG_ON(vma->anon_vma);
2792 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2793 }
Hugh Dickins6597d782012-10-08 16:29:07 -07002794 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2795 &prev, &rb_link, &rb_parent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 return -ENOMEM;
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002797 if ((vma->vm_flags & VM_ACCOUNT) &&
Alan Cox34b4e4a2007-08-22 14:01:28 -07002798 security_vm_enough_memory_mm(mm, vma_pages(vma)))
Hugh Dickins2fd4ef82005-09-14 06:13:02 +01002799 return -ENOMEM;
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302800
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 vma_link(mm, vma, prev, rb_link, rb_parent);
2802 return 0;
2803}
2804
2805/*
2806 * Copy the vma structure to a new location in the same mm,
2807 * prior to moving page table entries, to effect an mremap move.
2808 */
2809struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002810 unsigned long addr, unsigned long len, pgoff_t pgoff,
2811 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812{
2813 struct vm_area_struct *vma = *vmap;
2814 unsigned long vma_start = vma->vm_start;
2815 struct mm_struct *mm = vma->vm_mm;
2816 struct vm_area_struct *new_vma, *prev;
2817 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002818 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819
2820 /*
2821 * If anonymous vma has not yet been faulted, update new pgoff
2822 * to match new location, to increase its chance of merging.
2823 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002824 if (unlikely(!vma->vm_file && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002826 faulted_in_anon_vma = false;
2827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828
Hugh Dickins6597d782012-10-08 16:29:07 -07002829 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2830 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2832 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2833 if (new_vma) {
2834 /*
2835 * Source vma may have been merged into new_vma
2836 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002837 if (unlikely(vma_start >= new_vma->vm_start &&
2838 vma_start < new_vma->vm_end)) {
2839 /*
2840 * The only way we can get a vma_merge with
2841 * self during an mremap is if the vma hasn't
2842 * been faulted in yet and we were allowed to
2843 * reset the dst vma->vm_pgoff to the
2844 * destination address of the mremap to allow
2845 * the merge to happen. mremap must change the
2846 * vm_pgoff linearity between src and dst vmas
2847 * (in turn preventing a vma_merge) to be
2848 * safe. It is only safe to keep the vm_pgoff
2849 * linear if there are no pages mapped yet.
2850 */
2851 VM_BUG_ON(faulted_in_anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002852 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002853 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002854 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002856 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 if (new_vma) {
2858 *new_vma = *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 new_vma->vm_start = addr;
2860 new_vma->vm_end = addr + len;
2861 new_vma->vm_pgoff = pgoff;
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002862 if (vma_dup_policy(vma, new_vma))
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002863 goto out_free_vma;
Michel Lespinasse523d4e22012-10-08 16:31:48 -07002864 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2865 if (anon_vma_clone(new_vma, vma))
2866 goto out_free_mempol;
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002867 if (new_vma->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 get_file(new_vma->vm_file);
2869 if (new_vma->vm_ops && new_vma->vm_ops->open)
2870 new_vma->vm_ops->open(new_vma);
2871 vma_link(mm, new_vma, prev, rb_link, rb_parent);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002872 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 }
2874 }
2875 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002876
2877 out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002878 mpol_put(vma_policy(new_vma));
Rik van Riel5beb4932010-03-05 13:42:07 -08002879 out_free_vma:
2880 kmem_cache_free(vm_area_cachep, new_vma);
2881 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002883
2884/*
2885 * Return true if the calling process may expand its vm space by the passed
2886 * number of pages
2887 */
2888int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2889{
2890 unsigned long cur = mm->total_vm; /* pages */
2891 unsigned long lim;
2892
Jiri Slaby59e99e52010-03-05 13:41:44 -08002893 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002894
2895 if (cur + npages > lim)
2896 return 0;
2897 return 1;
2898}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002899
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002900static int special_mapping_fault(struct vm_area_struct *vma,
2901 struct vm_fault *vmf);
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
2910static const char *special_mapping_name(struct vm_area_struct *vma)
2911{
2912 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
2913}
2914
2915static const struct vm_operations_struct special_mapping_vmops = {
2916 .close = special_mapping_close,
2917 .fault = special_mapping_fault,
2918 .name = special_mapping_name,
2919};
2920
2921static const struct vm_operations_struct legacy_special_mapping_vmops = {
2922 .close = special_mapping_close,
2923 .fault = special_mapping_fault,
2924};
Roland McGrathfa5dc222007-02-08 14:20:41 -08002925
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002926static int special_mapping_fault(struct vm_area_struct *vma,
2927 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002928{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002929 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002930 struct page **pages;
2931
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002932 /*
2933 * special mappings have no vm_file, and in that case, the mm
2934 * uses vm_pgoff internally. So we have to subtract it from here.
2935 * We are allowed to do this because we are the mm; do not copy
2936 * this code into drivers!
2937 */
2938 pgoff = vmf->pgoff - vma->vm_pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002939
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002940 if (vma->vm_ops == &legacy_special_mapping_vmops)
2941 pages = vma->vm_private_data;
2942 else
2943 pages = ((struct vm_special_mapping *)vma->vm_private_data)->
2944 pages;
2945
2946 for (; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002947 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002948
2949 if (*pages) {
2950 struct page *page = *pages;
2951 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002952 vmf->page = page;
2953 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002954 }
2955
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002956 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002957}
2958
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002959static struct vm_area_struct *__install_special_mapping(
2960 struct mm_struct *mm,
2961 unsigned long addr, unsigned long len,
2962 unsigned long vm_flags, const struct vm_operations_struct *ops,
2963 void *priv)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002964{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002965 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002966 struct vm_area_struct *vma;
2967
2968 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2969 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01002970 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002971
Rik van Riel5beb4932010-03-05 13:42:07 -08002972 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002973 vma->vm_mm = mm;
2974 vma->vm_start = addr;
2975 vma->vm_end = addr + len;
2976
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002977 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07002978 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002979
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002980 vma->vm_ops = ops;
2981 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002982
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002983 ret = insert_vm_struct(mm, vma);
2984 if (ret)
2985 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002986
2987 mm->total_vm += len >> PAGE_SHIFT;
2988
Ingo Molnarcdd6c482009-09-21 12:02:48 +02002989 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02002990
Stefani Seibold3935ed62014-03-17 23:22:02 +01002991 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002992
2993out:
2994 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01002995 return ERR_PTR(ret);
2996}
2997
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002998/*
2999 * Called with mm->mmap_sem held for writing.
3000 * Insert a new vma covering the given region, with the given flags.
3001 * Its pages are supplied by the given array of struct page *.
3002 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3003 * The region past the last page supplied will always produce SIGBUS.
3004 * The array pointer and the pages it points to are assumed to stay alive
3005 * for as long as this mapping might exist.
3006 */
3007struct vm_area_struct *_install_special_mapping(
3008 struct mm_struct *mm,
3009 unsigned long addr, unsigned long len,
3010 unsigned long vm_flags, const struct vm_special_mapping *spec)
3011{
3012 return __install_special_mapping(mm, addr, len, vm_flags,
3013 &special_mapping_vmops, (void *)spec);
3014}
3015
Stefani Seibold3935ed62014-03-17 23:22:02 +01003016int install_special_mapping(struct mm_struct *mm,
3017 unsigned long addr, unsigned long len,
3018 unsigned long vm_flags, struct page **pages)
3019{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003020 struct vm_area_struct *vma = __install_special_mapping(
3021 mm, addr, len, vm_flags, &legacy_special_mapping_vmops,
3022 (void *)pages);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003023
Duan Jiong14bd5b42014-06-04 16:07:05 -07003024 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003025}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003026
3027static DEFINE_MUTEX(mm_all_locks_mutex);
3028
Peter Zijlstra454ed842008-08-11 09:30:25 +02003029static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003030{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003031 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003032 /*
3033 * The LSB of head.next can't change from under us
3034 * because we hold the mm_all_locks_mutex.
3035 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003036 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003037 /*
3038 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003039 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003040 * the same anon_vma we won't take it again.
3041 *
3042 * No need of atomic instructions here, head.next
3043 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003044 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003045 */
3046 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003047 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003048 BUG();
3049 }
3050}
3051
Peter Zijlstra454ed842008-08-11 09:30:25 +02003052static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003053{
3054 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3055 /*
3056 * AS_MM_ALL_LOCKS can't change from under us because
3057 * we hold the mm_all_locks_mutex.
3058 *
3059 * Operations on ->flags have to be atomic because
3060 * even if AS_MM_ALL_LOCKS is stable thanks to the
3061 * mm_all_locks_mutex, there may be other cpus
3062 * changing other bitflags in parallel to us.
3063 */
3064 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3065 BUG();
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003066 mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003067 }
3068}
3069
3070/*
3071 * This operation locks against the VM for all pte/vma/mm related
3072 * operations that could ever happen on a certain mm. This includes
3073 * vmtruncate, try_to_unmap, and all page faults.
3074 *
3075 * The caller must take the mmap_sem in write mode before calling
3076 * mm_take_all_locks(). The caller isn't allowed to release the
3077 * mmap_sem until mm_drop_all_locks() returns.
3078 *
3079 * mmap_sem in write mode is required in order to block all operations
3080 * that could modify pagetables and free pages without need of
3081 * altering the vma layout (for example populate_range() with
3082 * nonlinear vmas). It's also needed in write mode to avoid new
3083 * anon_vmas to be associated with existing vmas.
3084 *
3085 * A single task can't take more than one mm_take_all_locks() in a row
3086 * or it would deadlock.
3087 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003088 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003089 * mapping->flags avoid to take the same lock twice, if more than one
3090 * vma in this mm is backed by the same anon_vma or address_space.
3091 *
3092 * We can take all the locks in random order because the VM code
Yuanhan Liu631b0cf2013-02-04 14:28:48 -08003093 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003094 * takes more than one of them in a row. Secondly we're protected
3095 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3096 *
3097 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3098 * that may have to take thousand of locks.
3099 *
3100 * mm_take_all_locks() can fail if it's interrupted by signals.
3101 */
3102int mm_take_all_locks(struct mm_struct *mm)
3103{
3104 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003105 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003106
3107 BUG_ON(down_read_trylock(&mm->mmap_sem));
3108
3109 mutex_lock(&mm_all_locks_mutex);
3110
3111 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3112 if (signal_pending(current))
3113 goto out_unlock;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003114 if (vma->vm_file && vma->vm_file->f_mapping)
Peter Zijlstra454ed842008-08-11 09:30:25 +02003115 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003116 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003117
3118 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3119 if (signal_pending(current))
3120 goto out_unlock;
3121 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003122 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3123 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003124 }
3125
Kautuk Consul584cff52011-10-31 17:08:59 -07003126 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003127
3128out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003129 mm_drop_all_locks(mm);
3130 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003131}
3132
3133static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3134{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003135 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003136 /*
3137 * The LSB of head.next can't change to 0 from under
3138 * us because we hold the mm_all_locks_mutex.
3139 *
3140 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003141 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003142 * never see our bitflag.
3143 *
3144 * No need of atomic instructions here, head.next
3145 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003146 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003147 */
3148 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003149 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003150 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003151 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003152 }
3153}
3154
3155static void vm_unlock_mapping(struct address_space *mapping)
3156{
3157 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3158 /*
3159 * AS_MM_ALL_LOCKS can't change to 0 from under us
3160 * because we hold the mm_all_locks_mutex.
3161 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07003162 mutex_unlock(&mapping->i_mmap_mutex);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003163 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3164 &mapping->flags))
3165 BUG();
3166 }
3167}
3168
3169/*
3170 * The mmap_sem cannot be released by the caller until
3171 * mm_drop_all_locks() returns.
3172 */
3173void mm_drop_all_locks(struct mm_struct *mm)
3174{
3175 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003176 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003177
3178 BUG_ON(down_read_trylock(&mm->mmap_sem));
3179 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3180
3181 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3182 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003183 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3184 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003185 if (vma->vm_file && vma->vm_file->f_mapping)
3186 vm_unlock_mapping(vma->vm_file->f_mapping);
3187 }
3188
3189 mutex_unlock(&mm_all_locks_mutex);
3190}
David Howells8feae132009-01-08 12:04:47 +00003191
3192/*
3193 * initialise the VMA slab
3194 */
3195void __init mmap_init(void)
3196{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003197 int ret;
3198
3199 ret = percpu_counter_init(&vm_committed_as, 0);
3200 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003201}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003202
3203/*
3204 * Initialise sysctl_user_reserve_kbytes.
3205 *
3206 * This is intended to prevent a user from starting a single memory hogging
3207 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3208 * mode.
3209 *
3210 * The default value is min(3% of free memory, 128MB)
3211 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3212 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003213static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003214{
3215 unsigned long free_kbytes;
3216
3217 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3218
3219 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3220 return 0;
3221}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003222subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003223
3224/*
3225 * Initialise sysctl_admin_reserve_kbytes.
3226 *
3227 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3228 * to log in and kill a memory hogging process.
3229 *
3230 * Systems with more than 256MB will reserve 8MB, enough to recover
3231 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3232 * only reserve 3% of free pages by default.
3233 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003234static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003235{
3236 unsigned long free_kbytes;
3237
3238 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3239
3240 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3241 return 0;
3242}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003243subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003244
3245/*
3246 * Reinititalise user and admin reserves if memory is added or removed.
3247 *
3248 * The default user reserve max is 128MB, and the default max for the
3249 * admin reserve is 8MB. These are usually, but not always, enough to
3250 * enable recovery from a memory hogging process using login/sshd, a shell,
3251 * and tools like top. It may make sense to increase or even disable the
3252 * reserve depending on the existence of swap or variations in the recovery
3253 * tools. So, the admin may have changed them.
3254 *
3255 * If memory is added and the reserves have been eliminated or increased above
3256 * the default max, then we'll trust the admin.
3257 *
3258 * If memory is removed and there isn't enough free memory, then we
3259 * need to reset the reserves.
3260 *
3261 * Otherwise keep the reserve set by the admin.
3262 */
3263static int reserve_mem_notifier(struct notifier_block *nb,
3264 unsigned long action, void *data)
3265{
3266 unsigned long tmp, free_kbytes;
3267
3268 switch (action) {
3269 case MEM_ONLINE:
3270 /* Default max is 128MB. Leave alone if modified by operator. */
3271 tmp = sysctl_user_reserve_kbytes;
3272 if (0 < tmp && tmp < (1UL << 17))
3273 init_user_reserve();
3274
3275 /* Default max is 8MB. Leave alone if modified by operator. */
3276 tmp = sysctl_admin_reserve_kbytes;
3277 if (0 < tmp && tmp < (1UL << 13))
3278 init_admin_reserve();
3279
3280 break;
3281 case MEM_OFFLINE:
3282 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3283
3284 if (sysctl_user_reserve_kbytes > free_kbytes) {
3285 init_user_reserve();
3286 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3287 sysctl_user_reserve_kbytes);
3288 }
3289
3290 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3291 init_admin_reserve();
3292 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3293 sysctl_admin_reserve_kbytes);
3294 }
3295 break;
3296 default:
3297 break;
3298 }
3299 return NOTIFY_OK;
3300}
3301
3302static struct notifier_block reserve_mem_nb = {
3303 .notifier_call = reserve_mem_notifier,
3304};
3305
3306static int __meminit init_reserve_notifier(void)
3307{
3308 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003309 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003310
3311 return 0;
3312}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003313subsys_initcall(init_reserve_notifier);