blob: 234edffec1d0791251c631fc596b798dc49b7c91 [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>
Andrew Shewmaker16408792013-04-29 15:08:12 -070040#include <linux/notifier.h>
41#include <linux/memory.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070042#include <linux/printk.h>
Andrea Arcangeli19a809a2015-09-04 15:46:24 -070043#include <linux/userfaultfd_k.h>
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080044#include <linux/moduleparam.h>
Dave Hansen62b5f7d2016-02-12 13:02:40 -080045#include <linux/pkeys.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47#include <asm/uaccess.h>
48#include <asm/cacheflush.h>
49#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020050#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Jan Beulich42b77722008-07-23 21:27:10 -070052#include "internal.h"
53
Kirill Korotaev3a459752006-09-07 14:17:04 +040054#ifndef arch_mmap_check
55#define arch_mmap_check(addr, len, flags) (0)
56#endif
57
Daniel Cashmand07e2252016-01-14 15:19:53 -080058#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
59const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
60const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
61int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
62#endif
63#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
64const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
65const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
66int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
67#endif
68
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -070069static bool ignore_rlimit_data;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080070core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
Daniel Cashmand07e2252016-01-14 15:19:53 -080071
Hugh Dickinse0da3822005-04-19 13:29:15 -070072static void unmap_region(struct mm_struct *mm,
73 struct vm_area_struct *vma, struct vm_area_struct *prev,
74 unsigned long start, unsigned long end);
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* description of effects of mapping type and prot in current implementation.
77 * this is due to the limited x86 page protection hardware. The expected
78 * behavior is in parens:
79 *
80 * map_type prot
81 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
82 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
83 * w: (no) no w: (no) no w: (yes) yes w: (no) no
84 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070085 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 * w: (no) no w: (no) no w: (copy) copy w: (no) no
88 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
89 *
90 */
91pgprot_t protection_map[16] = {
92 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
93 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
94};
95
Hugh Dickins804af2c2006-07-26 21:39:49 +010096pgprot_t vm_get_page_prot(unsigned long vm_flags)
97{
Dave Kleikampb845f312008-07-08 00:28:51 +100098 return __pgprot(pgprot_val(protection_map[vm_flags &
99 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
100 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100101}
102EXPORT_SYMBOL(vm_get_page_prot);
103
Peter Feiner64e45502014-10-13 15:55:46 -0700104static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
105{
106 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
107}
108
109/* Update vma->vm_page_prot to reflect vma->vm_flags. */
110void vma_set_page_prot(struct vm_area_struct *vma)
111{
112 unsigned long vm_flags = vma->vm_flags;
113
114 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
115 if (vma_wants_writenotify(vma)) {
116 vm_flags &= ~VM_SHARED;
117 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
118 vm_flags);
119 }
120}
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800123 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 */
125static void __remove_shared_vm_struct(struct vm_area_struct *vma,
126 struct file *file, struct address_space *mapping)
127{
128 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500129 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700131 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800134 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 flush_dcache_mmap_unlock(mapping);
136}
137
138/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700139 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700140 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700142void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
144 struct file *file = vma->vm_file;
145
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 if (file) {
147 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800148 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800150 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700152}
153
154/*
155 * Close a vm structure and free it, returning the next.
156 */
157static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
158{
159 struct vm_area_struct *next = vma->vm_next;
160
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700161 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 if (vma->vm_ops && vma->vm_ops->close)
163 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700164 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700165 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700166 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700168 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169}
170
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700171static int do_brk(unsigned long addr, unsigned long len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700172
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100173SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700175 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 unsigned long newbrk, oldbrk;
177 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700178 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800179 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700181 if (down_write_killable(&mm->mmap_sem))
182 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700184#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800185 /*
186 * CONFIG_COMPAT_BRK can still be overridden by setting
187 * randomize_va_space to 2, which will still cause mm->start_brk
188 * to be arbitrarily shifted
189 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700190 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800191 min_brk = mm->start_brk;
192 else
193 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700194#else
195 min_brk = mm->start_brk;
196#endif
197 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700199
200 /*
201 * Check against rlimit here. If this check is done later after the test
202 * of oldbrk with newbrk then it can escape the test and let the data
203 * segment grow beyond its set limit the in case where the limit is
204 * not page aligned -Ram Gupta
205 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700206 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
207 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700208 goto out;
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 newbrk = PAGE_ALIGN(brk);
211 oldbrk = PAGE_ALIGN(mm->brk);
212 if (oldbrk == newbrk)
213 goto set_brk;
214
215 /* Always allow shrinking brk. */
216 if (brk <= mm->brk) {
217 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
218 goto set_brk;
219 goto out;
220 }
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 /* Check against existing mmap mappings. */
223 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
224 goto out;
225
226 /* Ok, looks good - let it rip. */
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700227 if (do_brk(oldbrk, newbrk-oldbrk) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230set_brk:
231 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800232 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
233 up_write(&mm->mmap_sem);
234 if (populate)
235 mm_populate(oldbrk, newbrk - oldbrk);
236 return brk;
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238out:
239 retval = mm->brk;
240 up_write(&mm->mmap_sem);
241 return retval;
242}
243
Michel Lespinassed3737182012-12-11 16:01:38 -0800244static long vma_compute_subtree_gap(struct vm_area_struct *vma)
245{
246 unsigned long max, subtree_gap;
247 max = vma->vm_start;
248 if (vma->vm_prev)
249 max -= vma->vm_prev->vm_end;
250 if (vma->vm_rb.rb_left) {
251 subtree_gap = rb_entry(vma->vm_rb.rb_left,
252 struct vm_area_struct, vm_rb)->rb_subtree_gap;
253 if (subtree_gap > max)
254 max = subtree_gap;
255 }
256 if (vma->vm_rb.rb_right) {
257 subtree_gap = rb_entry(vma->vm_rb.rb_right,
258 struct vm_area_struct, vm_rb)->rb_subtree_gap;
259 if (subtree_gap > max)
260 max = subtree_gap;
261 }
262 return max;
263}
264
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700265#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800266static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800268 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800269 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 struct rb_node *nd, *pn = NULL;
271 unsigned long prev = 0, pend = 0;
272
273 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
274 struct vm_area_struct *vma;
275 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800276 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700277 pr_emerg("vm_start %lx < prev %lx\n",
278 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800279 bug = 1;
280 }
281 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700282 pr_emerg("vm_start %lx < pend %lx\n",
283 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800284 bug = 1;
285 }
286 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700287 pr_emerg("vm_start %lx > vm_end %lx\n",
288 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800289 bug = 1;
290 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800291 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800292 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700293 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800294 vma->rb_subtree_gap,
295 vma_compute_subtree_gap(vma));
296 bug = 1;
297 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800298 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 i++;
300 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800301 prev = vma->vm_start;
302 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 }
304 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800305 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800307 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700308 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800309 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800311 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312}
313
Michel Lespinassed3737182012-12-11 16:01:38 -0800314static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
315{
316 struct rb_node *nd;
317
318 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
319 struct vm_area_struct *vma;
320 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700321 VM_BUG_ON_VMA(vma != ignore &&
322 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
323 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700327static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328{
329 int bug = 0;
330 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800331 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700332 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700333
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700334 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800335 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700336 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700337
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800338 if (anon_vma) {
339 anon_vma_lock_read(anon_vma);
340 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
341 anon_vma_interval_tree_verify(avc);
342 anon_vma_unlock_read(anon_vma);
343 }
344
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800345 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700346 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 i++;
348 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800349 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700350 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800351 bug = 1;
352 }
353 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700354 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700355 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800356 bug = 1;
357 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800358 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800359 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700360 if (i != -1)
361 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800362 bug = 1;
363 }
Sasha Levin96dad672014-10-09 15:28:39 -0700364 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365}
366#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800367#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368#define validate_mm(mm) do { } while (0)
369#endif
370
Michel Lespinassed3737182012-12-11 16:01:38 -0800371RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
372 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
373
374/*
375 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
376 * vma->vm_prev->vm_end values changed, without modifying the vma's position
377 * in the rbtree.
378 */
379static void vma_gap_update(struct vm_area_struct *vma)
380{
381 /*
382 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
383 * function that does exacltly what we want.
384 */
385 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
386}
387
388static inline void vma_rb_insert(struct vm_area_struct *vma,
389 struct rb_root *root)
390{
391 /* All rb_subtree_gap values must be consistent prior to insertion */
392 validate_mm_rb(root, NULL);
393
394 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
395}
396
397static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
398{
399 /*
400 * All rb_subtree_gap values must be consistent prior to erase,
401 * with the possible exception of the vma being erased.
402 */
403 validate_mm_rb(root, vma);
404
405 /*
406 * Note rb_erase_augmented is a fairly large inline function,
407 * so make sure we instantiate it only once with our desired
408 * augmented rbtree callbacks.
409 */
410 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
411}
412
Michel Lespinassebf181b92012-10-08 16:31:39 -0700413/*
414 * vma has some anon_vma assigned, and is already inserted on that
415 * anon_vma's interval trees.
416 *
417 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
418 * vma must be removed from the anon_vma's interval trees using
419 * anon_vma_interval_tree_pre_update_vma().
420 *
421 * After the update, the vma will be reinserted using
422 * anon_vma_interval_tree_post_update_vma().
423 *
424 * The entire update must be protected by exclusive mmap_sem and by
425 * the root anon_vma's mutex.
426 */
427static inline void
428anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
429{
430 struct anon_vma_chain *avc;
431
432 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
433 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
434}
435
436static inline void
437anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
438{
439 struct anon_vma_chain *avc;
440
441 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
442 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
443}
444
Hugh Dickins6597d782012-10-08 16:29:07 -0700445static int find_vma_links(struct mm_struct *mm, unsigned long addr,
446 unsigned long end, struct vm_area_struct **pprev,
447 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
Hugh Dickins6597d782012-10-08 16:29:07 -0700449 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 __rb_link = &mm->mm_rb.rb_node;
452 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 while (*__rb_link) {
455 struct vm_area_struct *vma_tmp;
456
457 __rb_parent = *__rb_link;
458 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
459
460 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700461 /* Fail if an existing vma overlaps the area */
462 if (vma_tmp->vm_start < end)
463 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 __rb_link = &__rb_parent->rb_left;
465 } else {
466 rb_prev = __rb_parent;
467 __rb_link = &__rb_parent->rb_right;
468 }
469 }
470
471 *pprev = NULL;
472 if (rb_prev)
473 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
474 *rb_link = __rb_link;
475 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700476 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
Cyril Hrubise8420a82013-04-29 15:08:33 -0700479static unsigned long count_vma_pages_range(struct mm_struct *mm,
480 unsigned long addr, unsigned long end)
481{
482 unsigned long nr_pages = 0;
483 struct vm_area_struct *vma;
484
485 /* Find first overlaping mapping */
486 vma = find_vma_intersection(mm, addr, end);
487 if (!vma)
488 return 0;
489
490 nr_pages = (min(end, vma->vm_end) -
491 max(addr, vma->vm_start)) >> PAGE_SHIFT;
492
493 /* Iterate over the rest of the overlaps */
494 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
495 unsigned long overlap_len;
496
497 if (vma->vm_start > end)
498 break;
499
500 overlap_len = min(end, vma->vm_end) - vma->vm_start;
501 nr_pages += overlap_len >> PAGE_SHIFT;
502 }
503
504 return nr_pages;
505}
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
508 struct rb_node **rb_link, struct rb_node *rb_parent)
509{
Michel Lespinassed3737182012-12-11 16:01:38 -0800510 /* Update tracking information for the gap following the new vma. */
511 if (vma->vm_next)
512 vma_gap_update(vma->vm_next);
513 else
514 mm->highest_vm_end = vma->vm_end;
515
516 /*
517 * vma->vm_prev wasn't known when we followed the rbtree to find the
518 * correct insertion point for that vma. As a result, we could not
519 * update the vma vm_rb parents rb_subtree_gap values on the way down.
520 * So, we first insert the vma with a zero rb_subtree_gap value
521 * (to be consistent with what we did on the way down), and then
522 * immediately update the gap to the correct value. Finally we
523 * rebalance the rbtree after all augmented values have been set.
524 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800526 vma->rb_subtree_gap = 0;
527 vma_gap_update(vma);
528 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529}
530
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700531static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532{
ZhenwenXu48aae422009-01-06 14:40:21 -0800533 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 file = vma->vm_file;
536 if (file) {
537 struct address_space *mapping = file->f_mapping;
538
539 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500540 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700542 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800545 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 flush_dcache_mmap_unlock(mapping);
547 }
548}
549
550static void
551__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
552 struct vm_area_struct *prev, struct rb_node **rb_link,
553 struct rb_node *rb_parent)
554{
555 __vma_link_list(mm, vma, prev, rb_parent);
556 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
559static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
560 struct vm_area_struct *prev, struct rb_node **rb_link,
561 struct rb_node *rb_parent)
562{
563 struct address_space *mapping = NULL;
564
Huang Shijie64ac4942014-06-04 16:07:33 -0700565 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800567 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 __vma_link(mm, vma, prev, rb_link, rb_parent);
571 __vma_link_file(vma);
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800574 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
576 mm->map_count++;
577 validate_mm(mm);
578}
579
580/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700581 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700582 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800584static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Hugh Dickins6597d782012-10-08 16:29:07 -0700586 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800587 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Hugh Dickins6597d782012-10-08 16:29:07 -0700589 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
590 &prev, &rb_link, &rb_parent))
591 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 __vma_link(mm, vma, prev, rb_link, rb_parent);
593 mm->map_count++;
594}
595
596static inline void
597__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
598 struct vm_area_struct *prev)
599{
Michel Lespinassed3737182012-12-11 16:01:38 -0800600 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700601
Michel Lespinassed3737182012-12-11 16:01:38 -0800602 vma_rb_erase(vma, &mm->mm_rb);
603 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700604 if (next)
605 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700606
607 /* Kill the cache */
608 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
611/*
612 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
613 * is already present in an i_mmap tree without adjusting the tree.
614 * The following helper function should be used when such adjustments
615 * are necessary. The "insert" vma (if any) is to be inserted
616 * before we drop the necessary locks.
617 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800618int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
620{
621 struct mm_struct *mm = vma->vm_mm;
622 struct vm_area_struct *next = vma->vm_next;
623 struct vm_area_struct *importer = NULL;
624 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700625 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700626 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800628 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 long adjust_next = 0;
630 int remove_next = 0;
631
632 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700633 struct vm_area_struct *exporter = NULL;
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 if (end >= next->vm_end) {
636 /*
637 * vma expands, overlapping all the next, and
638 * perhaps the one after too (mprotect case 6).
639 */
640again: remove_next = 1 + (end > next->vm_end);
641 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700642 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 importer = vma;
644 } else if (end > next->vm_start) {
645 /*
646 * vma expands, overlapping part of the next:
647 * mprotect case 5 shifting the boundary up.
648 */
649 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700650 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 importer = vma;
652 } else if (end < vma->vm_end) {
653 /*
654 * vma shrinks, and !insert tells it's not
655 * split_vma inserting another: so it must be
656 * mprotect case 4 shifting the boundary down.
657 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700658 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700659 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 importer = next;
661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Rik van Riel5beb4932010-03-05 13:42:07 -0800663 /*
664 * Easily overlooked: when mprotect shifts the boundary,
665 * make sure the expanding vma has anon_vma set if the
666 * shrinking vma had, to cover any anon pages imported.
667 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700668 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800669 int error;
670
Linus Torvalds287d97a2010-04-10 15:22:30 -0700671 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300672 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700673 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300674 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800675 }
676 }
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 if (file) {
679 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800680 root = &mapping->i_mmap;
681 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530682
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800683 if (adjust_next)
684 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530685
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800686 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700689 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 * are visible to arm/parisc __flush_dcache_page
691 * throughout; but we cannot insert into address
692 * space until vma start or end is updated.
693 */
694 __vma_link_file(insert);
695 }
696 }
697
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800698 vma_adjust_trans_huge(vma, start, end, adjust_next);
699
Michel Lespinassebf181b92012-10-08 16:31:39 -0700700 anon_vma = vma->anon_vma;
701 if (!anon_vma && adjust_next)
702 anon_vma = next->anon_vma;
703 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700704 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
705 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000706 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700707 anon_vma_interval_tree_pre_update_vma(vma);
708 if (adjust_next)
709 anon_vma_interval_tree_pre_update_vma(next);
710 }
Rik van Riel012f18002010-08-09 17:18:40 -0700711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if (root) {
713 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700714 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700716 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 }
718
Michel Lespinassed3737182012-12-11 16:01:38 -0800719 if (start != vma->vm_start) {
720 vma->vm_start = start;
721 start_changed = true;
722 }
723 if (end != vma->vm_end) {
724 vma->vm_end = end;
725 end_changed = true;
726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 vma->vm_pgoff = pgoff;
728 if (adjust_next) {
729 next->vm_start += adjust_next << PAGE_SHIFT;
730 next->vm_pgoff += adjust_next;
731 }
732
733 if (root) {
734 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700735 vma_interval_tree_insert(next, root);
736 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 flush_dcache_mmap_unlock(mapping);
738 }
739
740 if (remove_next) {
741 /*
742 * vma_merge has merged next into vma, and needs
743 * us to remove next before dropping the locks.
744 */
745 __vma_unlink(mm, next, vma);
746 if (file)
747 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 } else if (insert) {
749 /*
750 * split_vma has split insert from vma, and needs
751 * us to insert it before dropping the locks
752 * (it may either follow vma or precede it).
753 */
754 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800755 } else {
756 if (start_changed)
757 vma_gap_update(vma);
758 if (end_changed) {
759 if (!next)
760 mm->highest_vm_end = end;
761 else if (!adjust_next)
762 vma_gap_update(next);
763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 }
765
Michel Lespinassebf181b92012-10-08 16:31:39 -0700766 if (anon_vma) {
767 anon_vma_interval_tree_post_update_vma(vma);
768 if (adjust_next)
769 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800770 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800773 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530775 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100776 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530777
778 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100779 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530780 }
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700783 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530784 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700786 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800787 if (next->anon_vma)
788 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700790 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 kmem_cache_free(vm_area_cachep, next);
792 /*
793 * In mprotect's case 6 (see comments on vma_merge),
794 * we must remove another next too. It would clutter
795 * up the code too much to do both in one go.
796 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800797 next = vma->vm_next;
798 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800800 else if (next)
801 vma_gap_update(next);
802 else
803 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530805 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100806 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
808 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800809
810 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
813/*
814 * If the vma has a ->close operation then the driver probably needs to release
815 * per-vma resources, so we don't attempt to merge those.
816 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700818 struct file *file, unsigned long vm_flags,
819 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800821 /*
822 * VM_SOFTDIRTY should not prevent from VMA merging, if we
823 * match the flags but dirty bit -- the caller should mark
824 * merged VMA as dirty. If dirty bit won't be excluded from
825 * comparison, we increase pressue on the memory system forcing
826 * the kernel to generate new VMAs when old one could be
827 * extended instead.
828 */
829 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 return 0;
831 if (vma->vm_file != file)
832 return 0;
833 if (vma->vm_ops && vma->vm_ops->close)
834 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700835 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
836 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return 1;
838}
839
840static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700841 struct anon_vma *anon_vma2,
842 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
Shaohua Li965f55d2011-05-24 17:11:20 -0700844 /*
845 * The list_is_singular() test is to avoid merging VMA cloned from
846 * parents. This can improve scalability caused by anon_vma lock.
847 */
848 if ((!anon_vma1 || !anon_vma2) && (!vma ||
849 list_is_singular(&vma->anon_vma_chain)))
850 return 1;
851 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852}
853
854/*
855 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
856 * in front of (at a lower virtual address and file offset than) the vma.
857 *
858 * We cannot merge two vmas if they have differently assigned (non-NULL)
859 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
860 *
861 * We don't check here for the merged mmap wrapping around the end of pagecache
862 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
863 * wrap, nor mmaps which cover the final page at index -1UL.
864 */
865static int
866can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700867 struct anon_vma *anon_vma, struct file *file,
868 pgoff_t vm_pgoff,
869 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700871 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700872 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (vma->vm_pgoff == vm_pgoff)
874 return 1;
875 }
876 return 0;
877}
878
879/*
880 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
881 * beyond (at a higher virtual address and file offset than) the vma.
882 *
883 * We cannot merge two vmas if they have differently assigned (non-NULL)
884 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
885 */
886static int
887can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700888 struct anon_vma *anon_vma, struct file *file,
889 pgoff_t vm_pgoff,
890 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700892 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700893 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700895 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
897 return 1;
898 }
899 return 0;
900}
901
902/*
903 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
904 * whether that can be merged with its predecessor or its successor.
905 * Or both (it neatly fills a hole).
906 *
907 * In most cases - when called for mmap, brk or mremap - [addr,end) is
908 * certain not to be mapped by the time vma_merge is called; but when
909 * called for mprotect, it is certain to be already mapped (either at
910 * an offset within prev, or at the start of next), and the flags of
911 * this area are about to be changed to vm_flags - and the no-change
912 * case has already been eliminated.
913 *
914 * The following mprotect cases have to be considered, where AAAA is
915 * the area passed down from mprotect_fixup, never extending beyond one
916 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
917 *
918 * AAAA AAAA AAAA AAAA
919 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
920 * cannot merge might become might become might become
921 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
922 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
923 * mremap move: PPPPNNNNNNNN 8
924 * AAAA
925 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
926 * might become case 1 below case 2 below case 3 below
927 *
928 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
929 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
930 */
931struct vm_area_struct *vma_merge(struct mm_struct *mm,
932 struct vm_area_struct *prev, unsigned long addr,
933 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -0700934 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700935 pgoff_t pgoff, struct mempolicy *policy,
936 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
938 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
939 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800940 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
942 /*
943 * We later require that vma->vm_flags == vm_flags,
944 * so this tests vma->vm_flags & VM_SPECIAL, too.
945 */
946 if (vm_flags & VM_SPECIAL)
947 return NULL;
948
949 if (prev)
950 next = prev->vm_next;
951 else
952 next = mm->mmap;
953 area = next;
954 if (next && next->vm_end == end) /* cases 6, 7, 8 */
955 next = next->vm_next;
956
957 /*
958 * Can it merge with the predecessor?
959 */
960 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -0700961 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700963 anon_vma, file, pgoff,
964 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 /*
966 * OK, it can. Can we now merge in the successor as well?
967 */
968 if (next && end == next->vm_start &&
969 mpol_equal(policy, vma_policy(next)) &&
970 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700971 anon_vma, file,
972 pgoff+pglen,
973 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700975 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800977 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 next->vm_end, prev->vm_pgoff, NULL);
979 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800980 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800982 if (err)
983 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -0700984 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 return prev;
986 }
987
988 /*
989 * Can this new request be merged in front of next?
990 */
991 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -0700992 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700994 anon_vma, file, pgoff+pglen,
995 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800997 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 addr, prev->vm_pgoff, NULL);
999 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001000 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001002 if (err)
1003 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001004 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 return area;
1006 }
1007
1008 return NULL;
1009}
1010
1011/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001012 * Rough compatbility check to quickly see if it's even worth looking
1013 * at sharing an anon_vma.
1014 *
1015 * They need to have the same vm_file, and the flags can only differ
1016 * in things that mprotect may change.
1017 *
1018 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1019 * we can merge the two vma's. For example, we refuse to merge a vma if
1020 * there is a vm_ops->close() function, because that indicates that the
1021 * driver is doing some kind of reference counting. But that doesn't
1022 * really matter for the anon_vma sharing case.
1023 */
1024static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1025{
1026 return a->vm_end == b->vm_start &&
1027 mpol_equal(vma_policy(a), vma_policy(b)) &&
1028 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001029 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001030 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1031}
1032
1033/*
1034 * Do some basic sanity checking to see if we can re-use the anon_vma
1035 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1036 * the same as 'old', the other will be the new one that is trying
1037 * to share the anon_vma.
1038 *
1039 * NOTE! This runs with mm_sem held for reading, so it is possible that
1040 * the anon_vma of 'old' is concurrently in the process of being set up
1041 * by another page fault trying to merge _that_. But that's ok: if it
1042 * is being set up, that automatically means that it will be a singleton
1043 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001044 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001045 *
1046 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1047 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1048 * is to return an anon_vma that is "complex" due to having gone through
1049 * a fork).
1050 *
1051 * We also make sure that the two vma's are compatible (adjacent,
1052 * and with the same memory policies). That's all stable, even with just
1053 * a read lock on the mm_sem.
1054 */
1055static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1056{
1057 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001058 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001059
1060 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1061 return anon_vma;
1062 }
1063 return NULL;
1064}
1065
1066/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1068 * neighbouring vmas for a suitable anon_vma, before it goes off
1069 * to allocate a new anon_vma. It checks because a repetitive
1070 * sequence of mprotects and faults may otherwise lead to distinct
1071 * anon_vmas being allocated, preventing vma merge in subsequent
1072 * mprotect.
1073 */
1074struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1075{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001076 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 near = vma->vm_next;
1080 if (!near)
1081 goto try_prev;
1082
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001083 anon_vma = reusable_anon_vma(near, vma, near);
1084 if (anon_vma)
1085 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001087 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 if (!near)
1089 goto none;
1090
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001091 anon_vma = reusable_anon_vma(near, near, vma);
1092 if (anon_vma)
1093 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094none:
1095 /*
1096 * There's no absolute need to look only at touching neighbours:
1097 * we could search further afield for "compatible" anon_vmas.
1098 * But it would probably just be a waste of time searching,
1099 * or lead to too many vmas hanging off the same anon_vma.
1100 * We're trying to allow mprotect remerging later on,
1101 * not trying to minimize memory used for anon_vmas.
1102 */
1103 return NULL;
1104}
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106/*
Al Viro40401532012-02-13 03:58:52 +00001107 * If a hint addr is less than mmap_min_addr change hint to be as
1108 * low as possible but still greater than mmap_min_addr
1109 */
1110static inline unsigned long round_hint_to_min(unsigned long hint)
1111{
1112 hint &= PAGE_MASK;
1113 if (((void *)hint != NULL) &&
1114 (hint < mmap_min_addr))
1115 return PAGE_ALIGN(mmap_min_addr);
1116 return hint;
1117}
1118
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001119static inline int mlock_future_check(struct mm_struct *mm,
1120 unsigned long flags,
1121 unsigned long len)
1122{
1123 unsigned long locked, lock_limit;
1124
1125 /* mlock MCL_FUTURE? */
1126 if (flags & VM_LOCKED) {
1127 locked = len >> PAGE_SHIFT;
1128 locked += mm->locked_vm;
1129 lock_limit = rlimit(RLIMIT_MEMLOCK);
1130 lock_limit >>= PAGE_SHIFT;
1131 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1132 return -EAGAIN;
1133 }
1134 return 0;
1135}
1136
Al Viro40401532012-02-13 03:58:52 +00001137/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001138 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001140unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001142 unsigned long flags, vm_flags_t vm_flags,
1143 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
vishnu.pscc71aba2014-10-09 15:26:29 -07001145 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001146 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Michel Lespinasse41badc12013-02-22 16:32:47 -08001148 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001149
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001150 if (!len)
1151 return -EINVAL;
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 /*
1154 * Does the application expect PROT_READ to imply PROT_EXEC?
1155 *
1156 * (the exception is when the underlying filesystem is noexec
1157 * mounted, in which case we dont add PROT_EXEC.)
1158 */
1159 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001160 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 prot |= PROT_EXEC;
1162
Eric Paris7cd94142007-11-26 18:47:40 -05001163 if (!(flags & MAP_FIXED))
1164 addr = round_hint_to_min(addr);
1165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 /* Careful about overflows.. */
1167 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001168 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 return -ENOMEM;
1170
1171 /* offset overflow? */
1172 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001173 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175 /* Too many mappings? */
1176 if (mm->map_count > sysctl_max_map_count)
1177 return -ENOMEM;
1178
1179 /* Obtain the address to map to. we verify (or select) it and ensure
1180 * that it represents a valid section of the address space.
1181 */
1182 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001183 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 return addr;
1185
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001186 if (prot == PROT_EXEC) {
1187 pkey = execute_only_pkey(mm);
1188 if (pkey < 0)
1189 pkey = 0;
1190 }
1191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 /* Do simple checking here so the lower-level routines won't have
1193 * to. we assume access permissions have been handled by the open
1194 * of the memory object, so we don't do any here.
1195 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001196 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1198
Huang Shijiecdf7b342009-09-21 17:03:36 -07001199 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 if (!can_do_mlock())
1201 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001202
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001203 if (mlock_future_check(mm, vm_flags, len))
1204 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001207 struct inode *inode = file_inode(file);
1208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 switch (flags & MAP_TYPE) {
1210 case MAP_SHARED:
1211 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1212 return -EACCES;
1213
1214 /*
1215 * Make sure we don't allow writing to an append-only
1216 * file..
1217 */
1218 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1219 return -EACCES;
1220
1221 /*
1222 * Make sure there are no mandatory locks on the file.
1223 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001224 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 return -EAGAIN;
1226
1227 vm_flags |= VM_SHARED | VM_MAYSHARE;
1228 if (!(file->f_mode & FMODE_WRITE))
1229 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1230
1231 /* fall through */
1232 case MAP_PRIVATE:
1233 if (!(file->f_mode & FMODE_READ))
1234 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001235 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001236 if (vm_flags & VM_EXEC)
1237 return -EPERM;
1238 vm_flags &= ~VM_MAYEXEC;
1239 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001240
Al Viro72c2d532013-09-22 16:27:52 -04001241 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001242 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001243 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1244 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 break;
1246
1247 default:
1248 return -EINVAL;
1249 }
1250 } else {
1251 switch (flags & MAP_TYPE) {
1252 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001253 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1254 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001255 /*
1256 * Ignore pgoff.
1257 */
1258 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 vm_flags |= VM_SHARED | VM_MAYSHARE;
1260 break;
1261 case MAP_PRIVATE:
1262 /*
1263 * Set pgoff according to addr for anon_vma.
1264 */
1265 pgoff = addr >> PAGE_SHIFT;
1266 break;
1267 default:
1268 return -EINVAL;
1269 }
1270 }
1271
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001272 /*
1273 * Set 'VM_NORESERVE' if we should not account for the
1274 * memory use of this mapping.
1275 */
1276 if (flags & MAP_NORESERVE) {
1277 /* We honor MAP_NORESERVE if allowed to overcommit */
1278 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1279 vm_flags |= VM_NORESERVE;
1280
1281 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1282 if (file && is_file_hugepages(file))
1283 vm_flags |= VM_NORESERVE;
1284 }
1285
1286 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001287 if (!IS_ERR_VALUE(addr) &&
1288 ((vm_flags & VM_LOCKED) ||
1289 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001290 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001291 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001292}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001293
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001294SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1295 unsigned long, prot, unsigned long, flags,
1296 unsigned long, fd, unsigned long, pgoff)
1297{
1298 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001299 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001300
1301 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001302 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001303 file = fget(fd);
1304 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001305 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001306 if (is_file_hugepages(file))
1307 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001308 retval = -EINVAL;
1309 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1310 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001311 } else if (flags & MAP_HUGETLB) {
1312 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001313 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001314
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001315 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001316 if (!hs)
1317 return -EINVAL;
1318
1319 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001320 /*
1321 * VM_NORESERVE is used because the reservations will be
1322 * taken when vm_ops->mmap() is called
1323 * A dummy user value is used because we are not locking
1324 * memory so no accounting is necessary
1325 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001326 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001327 VM_NORESERVE,
1328 &user, HUGETLB_ANONHUGE_INODE,
1329 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001330 if (IS_ERR(file))
1331 return PTR_ERR(file);
1332 }
1333
1334 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1335
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001336 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001337out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001338 if (file)
1339 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001340 return retval;
1341}
1342
Christoph Hellwiga4679372010-03-10 15:21:15 -08001343#ifdef __ARCH_WANT_SYS_OLD_MMAP
1344struct mmap_arg_struct {
1345 unsigned long addr;
1346 unsigned long len;
1347 unsigned long prot;
1348 unsigned long flags;
1349 unsigned long fd;
1350 unsigned long offset;
1351};
1352
1353SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1354{
1355 struct mmap_arg_struct a;
1356
1357 if (copy_from_user(&a, arg, sizeof(a)))
1358 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001359 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001360 return -EINVAL;
1361
1362 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1363 a.offset >> PAGE_SHIFT);
1364}
1365#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1366
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001367/*
1368 * Some shared mappigns will want the pages marked read-only
1369 * to track write events. If so, we'll downgrade vm_page_prot
1370 * to the private version (using protection_map[] without the
1371 * VM_SHARED bit).
1372 */
1373int vma_wants_writenotify(struct vm_area_struct *vma)
1374{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001375 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001376 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001377
1378 /* If it was private or non-writable, the write bit is already clear */
1379 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1380 return 0;
1381
1382 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001383 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001384 return 1;
1385
Peter Feiner64e45502014-10-13 15:55:46 -07001386 /* The open routine did something to the protections that pgprot_modify
1387 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001388 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001389 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001390 return 0;
1391
Peter Feiner64e45502014-10-13 15:55:46 -07001392 /* Do we need to track softdirty? */
1393 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1394 return 1;
1395
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001396 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001397 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001398 return 0;
1399
1400 /* Can the mapping track the dirty pages? */
1401 return vma->vm_file && vma->vm_file->f_mapping &&
1402 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1403}
1404
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001405/*
1406 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001407 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001408 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001409static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001410{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001411 /*
1412 * hugetlb has its own accounting separate from the core VM
1413 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1414 */
1415 if (file && is_file_hugepages(file))
1416 return 0;
1417
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001418 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1419}
1420
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001421unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001422 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001423{
1424 struct mm_struct *mm = current->mm;
1425 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001426 int error;
1427 struct rb_node **rb_link, *rb_parent;
1428 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001429
Cyril Hrubise8420a82013-04-29 15:08:33 -07001430 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001431 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001432 unsigned long nr_pages;
1433
1434 /*
1435 * MAP_FIXED may remove pages of mappings that intersects with
1436 * requested mapping. Account for the pages it would unmap.
1437 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001438 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1439
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001440 if (!may_expand_vm(mm, vm_flags,
1441 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001442 return -ENOMEM;
1443 }
1444
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001446 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1447 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 if (do_munmap(mm, addr, len))
1449 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 }
1451
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001452 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001453 * Private writable mapping: check memory availability
1454 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001455 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001456 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001457 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001458 return -ENOMEM;
1459 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
1461
1462 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001463 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001465 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1466 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001467 if (vma)
1468 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
1470 /*
1471 * Determine the object being mapped and call the appropriate
1472 * specific mapper. the address has already been validated, but
1473 * not unmapped, but the maps are removed from the list.
1474 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001475 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (!vma) {
1477 error = -ENOMEM;
1478 goto unacct_error;
1479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
1481 vma->vm_mm = mm;
1482 vma->vm_start = addr;
1483 vma->vm_end = addr + len;
1484 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001485 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001487 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
1489 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 if (vm_flags & VM_DENYWRITE) {
1491 error = deny_write_access(file);
1492 if (error)
1493 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001495 if (vm_flags & VM_SHARED) {
1496 error = mapping_map_writable(file->f_mapping);
1497 if (error)
1498 goto allow_write_and_free_vma;
1499 }
1500
1501 /* ->mmap() can change vma->vm_file, but must guarantee that
1502 * vma_link() below can deny write-access if VM_DENYWRITE is set
1503 * and map writably if VM_SHARED is set. This usually means the
1504 * new file must not have been exposed to user-space, yet.
1505 */
Al Virocb0942b2012-08-27 14:48:26 -04001506 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 error = file->f_op->mmap(file, vma);
1508 if (error)
1509 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001510
1511 /* Can addr have changed??
1512 *
1513 * Answer: Yes, several device drivers can do it in their
1514 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001515 * Bug: If addr is changed, prev, rb_link, rb_parent should
1516 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001517 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001518 WARN_ON_ONCE(addr != vma->vm_start);
1519
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001520 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001521 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 } else if (vm_flags & VM_SHARED) {
1523 error = shmem_zero_setup(vma);
1524 if (error)
1525 goto free_vma;
1526 }
1527
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001528 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001529 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001530 if (file) {
1531 if (vm_flags & VM_SHARED)
1532 mapping_unmap_writable(file->f_mapping);
1533 if (vm_flags & VM_DENYWRITE)
1534 allow_write_access(file);
1535 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001536 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001537out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001538 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001539
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001540 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001542 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1543 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001544 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001545 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001546 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001547 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301548
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001549 if (file)
1550 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301551
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001552 /*
1553 * New (or expanded) vma always get soft dirty status.
1554 * Otherwise user-space soft-dirty page tracker won't
1555 * be able to distinguish situation when vma area unmapped,
1556 * then new mapped in-place (which must be aimed as
1557 * a completely new data area).
1558 */
1559 vma->vm_flags |= VM_SOFTDIRTY;
1560
Peter Feiner64e45502014-10-13 15:55:46 -07001561 vma_set_page_prot(vma);
1562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 return addr;
1564
1565unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 vma->vm_file = NULL;
1567 fput(file);
1568
1569 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001570 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1571 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001572 if (vm_flags & VM_SHARED)
1573 mapping_unmap_writable(file->f_mapping);
1574allow_write_and_free_vma:
1575 if (vm_flags & VM_DENYWRITE)
1576 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577free_vma:
1578 kmem_cache_free(vm_area_cachep, vma);
1579unacct_error:
1580 if (charged)
1581 vm_unacct_memory(charged);
1582 return error;
1583}
1584
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001585unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1586{
1587 /*
1588 * We implement the search by looking for an rbtree node that
1589 * immediately follows a suitable gap. That is,
1590 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1591 * - gap_end = vma->vm_start >= info->low_limit + length;
1592 * - gap_end - gap_start >= length
1593 */
1594
1595 struct mm_struct *mm = current->mm;
1596 struct vm_area_struct *vma;
1597 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1598
1599 /* Adjust search length to account for worst case alignment overhead */
1600 length = info->length + info->align_mask;
1601 if (length < info->length)
1602 return -ENOMEM;
1603
1604 /* Adjust search limits by the desired length */
1605 if (info->high_limit < length)
1606 return -ENOMEM;
1607 high_limit = info->high_limit - length;
1608
1609 if (info->low_limit > high_limit)
1610 return -ENOMEM;
1611 low_limit = info->low_limit + length;
1612
1613 /* Check if rbtree root looks promising */
1614 if (RB_EMPTY_ROOT(&mm->mm_rb))
1615 goto check_highest;
1616 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1617 if (vma->rb_subtree_gap < length)
1618 goto check_highest;
1619
1620 while (true) {
1621 /* Visit left subtree if it looks promising */
1622 gap_end = vma->vm_start;
1623 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1624 struct vm_area_struct *left =
1625 rb_entry(vma->vm_rb.rb_left,
1626 struct vm_area_struct, vm_rb);
1627 if (left->rb_subtree_gap >= length) {
1628 vma = left;
1629 continue;
1630 }
1631 }
1632
1633 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1634check_current:
1635 /* Check if current node has a suitable gap */
1636 if (gap_start > high_limit)
1637 return -ENOMEM;
1638 if (gap_end >= low_limit && gap_end - gap_start >= length)
1639 goto found;
1640
1641 /* Visit right subtree if it looks promising */
1642 if (vma->vm_rb.rb_right) {
1643 struct vm_area_struct *right =
1644 rb_entry(vma->vm_rb.rb_right,
1645 struct vm_area_struct, vm_rb);
1646 if (right->rb_subtree_gap >= length) {
1647 vma = right;
1648 continue;
1649 }
1650 }
1651
1652 /* Go back up the rbtree to find next candidate node */
1653 while (true) {
1654 struct rb_node *prev = &vma->vm_rb;
1655 if (!rb_parent(prev))
1656 goto check_highest;
1657 vma = rb_entry(rb_parent(prev),
1658 struct vm_area_struct, vm_rb);
1659 if (prev == vma->vm_rb.rb_left) {
1660 gap_start = vma->vm_prev->vm_end;
1661 gap_end = vma->vm_start;
1662 goto check_current;
1663 }
1664 }
1665 }
1666
1667check_highest:
1668 /* Check highest gap, which does not precede any rbtree node */
1669 gap_start = mm->highest_vm_end;
1670 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1671 if (gap_start > high_limit)
1672 return -ENOMEM;
1673
1674found:
1675 /* We found a suitable gap. Clip it with the original low_limit. */
1676 if (gap_start < info->low_limit)
1677 gap_start = info->low_limit;
1678
1679 /* Adjust gap address to the desired alignment */
1680 gap_start += (info->align_offset - gap_start) & info->align_mask;
1681
1682 VM_BUG_ON(gap_start + info->length > info->high_limit);
1683 VM_BUG_ON(gap_start + info->length > gap_end);
1684 return gap_start;
1685}
1686
1687unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1688{
1689 struct mm_struct *mm = current->mm;
1690 struct vm_area_struct *vma;
1691 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1692
1693 /* Adjust search length to account for worst case alignment overhead */
1694 length = info->length + info->align_mask;
1695 if (length < info->length)
1696 return -ENOMEM;
1697
1698 /*
1699 * Adjust search limits by the desired length.
1700 * See implementation comment at top of unmapped_area().
1701 */
1702 gap_end = info->high_limit;
1703 if (gap_end < length)
1704 return -ENOMEM;
1705 high_limit = gap_end - length;
1706
1707 if (info->low_limit > high_limit)
1708 return -ENOMEM;
1709 low_limit = info->low_limit + length;
1710
1711 /* Check highest gap, which does not precede any rbtree node */
1712 gap_start = mm->highest_vm_end;
1713 if (gap_start <= high_limit)
1714 goto found_highest;
1715
1716 /* Check if rbtree root looks promising */
1717 if (RB_EMPTY_ROOT(&mm->mm_rb))
1718 return -ENOMEM;
1719 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1720 if (vma->rb_subtree_gap < length)
1721 return -ENOMEM;
1722
1723 while (true) {
1724 /* Visit right subtree if it looks promising */
1725 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1726 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1727 struct vm_area_struct *right =
1728 rb_entry(vma->vm_rb.rb_right,
1729 struct vm_area_struct, vm_rb);
1730 if (right->rb_subtree_gap >= length) {
1731 vma = right;
1732 continue;
1733 }
1734 }
1735
1736check_current:
1737 /* Check if current node has a suitable gap */
1738 gap_end = vma->vm_start;
1739 if (gap_end < low_limit)
1740 return -ENOMEM;
1741 if (gap_start <= high_limit && gap_end - gap_start >= length)
1742 goto found;
1743
1744 /* Visit left subtree if it looks promising */
1745 if (vma->vm_rb.rb_left) {
1746 struct vm_area_struct *left =
1747 rb_entry(vma->vm_rb.rb_left,
1748 struct vm_area_struct, vm_rb);
1749 if (left->rb_subtree_gap >= length) {
1750 vma = left;
1751 continue;
1752 }
1753 }
1754
1755 /* Go back up the rbtree to find next candidate node */
1756 while (true) {
1757 struct rb_node *prev = &vma->vm_rb;
1758 if (!rb_parent(prev))
1759 return -ENOMEM;
1760 vma = rb_entry(rb_parent(prev),
1761 struct vm_area_struct, vm_rb);
1762 if (prev == vma->vm_rb.rb_right) {
1763 gap_start = vma->vm_prev ?
1764 vma->vm_prev->vm_end : 0;
1765 goto check_current;
1766 }
1767 }
1768 }
1769
1770found:
1771 /* We found a suitable gap. Clip it with the original high_limit. */
1772 if (gap_end > info->high_limit)
1773 gap_end = info->high_limit;
1774
1775found_highest:
1776 /* Compute highest gap address at the desired alignment */
1777 gap_end -= info->length;
1778 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1779
1780 VM_BUG_ON(gap_end < info->low_limit);
1781 VM_BUG_ON(gap_end < gap_start);
1782 return gap_end;
1783}
1784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785/* Get an address range which is currently unmapped.
1786 * For shmat() with addr=0.
1787 *
1788 * Ugly calling convention alert:
1789 * Return value with the low bits set means error value,
1790 * ie
1791 * if (ret & ~PAGE_MASK)
1792 * error = ret;
1793 *
1794 * This function "knows" that -ENOMEM has the bits set.
1795 */
1796#ifndef HAVE_ARCH_UNMAPPED_AREA
1797unsigned long
1798arch_get_unmapped_area(struct file *filp, unsigned long addr,
1799 unsigned long len, unsigned long pgoff, unsigned long flags)
1800{
1801 struct mm_struct *mm = current->mm;
1802 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001803 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001805 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 return -ENOMEM;
1807
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001808 if (flags & MAP_FIXED)
1809 return addr;
1810
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (addr) {
1812 addr = PAGE_ALIGN(addr);
1813 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001814 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 (!vma || addr + len <= vma->vm_start))
1816 return addr;
1817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001819 info.flags = 0;
1820 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001821 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001822 info.high_limit = TASK_SIZE;
1823 info.align_mask = 0;
1824 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825}
vishnu.pscc71aba2014-10-09 15:26:29 -07001826#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828/*
1829 * This mmap-allocator allocates new areas top-down from below the
1830 * stack's low limit (the base):
1831 */
1832#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1833unsigned long
1834arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1835 const unsigned long len, const unsigned long pgoff,
1836 const unsigned long flags)
1837{
1838 struct vm_area_struct *vma;
1839 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001840 unsigned long addr = addr0;
1841 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
1843 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001844 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 return -ENOMEM;
1846
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001847 if (flags & MAP_FIXED)
1848 return addr;
1849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 /* requesting a specific address */
1851 if (addr) {
1852 addr = PAGE_ALIGN(addr);
1853 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001854 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 (!vma || addr + len <= vma->vm_start))
1856 return addr;
1857 }
1858
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001859 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1860 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001861 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001862 info.high_limit = mm->mmap_base;
1863 info.align_mask = 0;
1864 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 /*
1867 * A failed mmap() very likely causes application failure,
1868 * so fall back to the bottom-up function here. This scenario
1869 * can happen with large stack limits and large mmap()
1870 * allocations.
1871 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001872 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001873 VM_BUG_ON(addr != -ENOMEM);
1874 info.flags = 0;
1875 info.low_limit = TASK_UNMAPPED_BASE;
1876 info.high_limit = TASK_SIZE;
1877 addr = vm_unmapped_area(&info);
1878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
1880 return addr;
1881}
1882#endif
1883
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884unsigned long
1885get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1886 unsigned long pgoff, unsigned long flags)
1887{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001888 unsigned long (*get_area)(struct file *, unsigned long,
1889 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Al Viro9206de92009-12-03 15:23:11 -05001891 unsigned long error = arch_mmap_check(addr, len, flags);
1892 if (error)
1893 return error;
1894
1895 /* Careful about overflows.. */
1896 if (len > TASK_SIZE)
1897 return -ENOMEM;
1898
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001899 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04001900 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001901 get_area = file->f_op->get_unmapped_area;
1902 addr = get_area(file, addr, len, pgoff, flags);
1903 if (IS_ERR_VALUE(addr))
1904 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001906 if (addr > TASK_SIZE - len)
1907 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001908 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001909 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001910
Al Viro9ac4ed42012-05-30 17:13:15 -04001911 error = security_mmap_addr(addr);
1912 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913}
1914
1915EXPORT_SYMBOL(get_unmapped_area);
1916
1917/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001918struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001920 struct rb_node *rb_node;
1921 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001923 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001924 vma = vmacache_find(mm, addr);
1925 if (likely(vma))
1926 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001928 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001930 while (rb_node) {
1931 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001933 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001934
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001935 if (tmp->vm_end > addr) {
1936 vma = tmp;
1937 if (tmp->vm_start <= addr)
1938 break;
1939 rb_node = rb_node->rb_left;
1940 } else
1941 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001943
1944 if (vma)
1945 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 return vma;
1947}
1948
1949EXPORT_SYMBOL(find_vma);
1950
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001951/*
1952 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001953 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954struct vm_area_struct *
1955find_vma_prev(struct mm_struct *mm, unsigned long addr,
1956 struct vm_area_struct **pprev)
1957{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001958 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001960 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05001961 if (vma) {
1962 *pprev = vma->vm_prev;
1963 } else {
1964 struct rb_node *rb_node = mm->mm_rb.rb_node;
1965 *pprev = NULL;
1966 while (rb_node) {
1967 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1968 rb_node = rb_node->rb_right;
1969 }
1970 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001971 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972}
1973
1974/*
1975 * Verify that the stack growth is acceptable and
1976 * update accounting. This is shared with both the
1977 * grow-up and grow-down cases.
1978 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001979static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980{
1981 struct mm_struct *mm = vma->vm_mm;
1982 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08001983 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
1985 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001986 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 return -ENOMEM;
1988
1989 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08001990 actual_size = size;
1991 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
1992 actual_size -= PAGE_SIZE;
Jason Low4db0c3c2015-04-15 16:14:08 -07001993 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return -ENOMEM;
1995
1996 /* mlock limit tests */
1997 if (vma->vm_flags & VM_LOCKED) {
1998 unsigned long locked;
1999 unsigned long limit;
2000 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002001 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002002 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 if (locked > limit && !capable(CAP_IPC_LOCK))
2004 return -ENOMEM;
2005 }
2006
Adam Litke0d59a012007-01-30 14:35:39 -08002007 /* Check to ensure the stack will not grow into a hugetlb-only region */
2008 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2009 vma->vm_end - size;
2010 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2011 return -EFAULT;
2012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 /*
2014 * Overcommit.. This must be the final test, as it will
2015 * update security statistics.
2016 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002017 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 return -ENOMEM;
2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 return 0;
2021}
2022
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002023#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002025 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2026 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002028int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
Oleg Nesterov09357812015-11-05 18:48:17 -08002030 struct mm_struct *mm = vma->vm_mm;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002031 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
2033 if (!(vma->vm_flags & VM_GROWSUP))
2034 return -EFAULT;
2035
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002036 /* Guard against wrapping around to address 0. */
2037 if (address < PAGE_ALIGN(address+4))
2038 address = PAGE_ALIGN(address+4);
2039 else
2040 return -ENOMEM;
2041
2042 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 if (unlikely(anon_vma_prepare(vma)))
2044 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
2046 /*
2047 * vma->vm_start/vm_end cannot change under us because the caller
2048 * is required to hold the mmap_sem in read mode. We need the
2049 * anon_vma lock to serialize against concurrent expand_stacks.
2050 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002051 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
2053 /* Somebody else might have raced and expanded it already */
2054 if (address > vma->vm_end) {
2055 unsigned long size, grow;
2056
2057 size = address - vma->vm_start;
2058 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2059
Hugh Dickins42c36f62011-05-09 17:44:42 -07002060 error = -ENOMEM;
2061 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2062 error = acct_stack_growth(vma, size, grow);
2063 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002064 /*
2065 * vma_gap_update() doesn't support concurrent
2066 * updates, but we only hold a shared mmap_sem
2067 * lock here, so we need to protect against
2068 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002069 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002070 * we don't guarantee that all growable vmas
2071 * in a mm share the same root anon vma.
2072 * So, we reuse mm->page_table_lock to guard
2073 * against concurrent vma expansions.
2074 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002075 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002076 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002077 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002078 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002079 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002080 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002081 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002082 if (vma->vm_next)
2083 vma_gap_update(vma->vm_next);
2084 else
Oleg Nesterov09357812015-11-05 18:48:17 -08002085 mm->highest_vm_end = address;
2086 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002087
Hugh Dickins42c36f62011-05-09 17:44:42 -07002088 perf_event_mmap(vma);
2089 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002092 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002093 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002094 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 return error;
2096}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002097#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2098
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099/*
2100 * vma is the first one with address < vma->vm_start. Have to extend vma.
2101 */
Michal Hockod05f3162011-05-24 17:11:44 -07002102int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002103 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
Oleg Nesterov09357812015-11-05 18:48:17 -08002105 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 int error;
2107
Eric Paris88694772007-11-26 18:47:26 -05002108 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002109 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002110 if (error)
2111 return error;
2112
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002113 /* We must make sure the anon_vma is allocated. */
2114 if (unlikely(anon_vma_prepare(vma)))
2115 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117 /*
2118 * vma->vm_start/vm_end cannot change under us because the caller
2119 * is required to hold the mmap_sem in read mode. We need the
2120 * anon_vma lock to serialize against concurrent expand_stacks.
2121 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002122 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
2124 /* Somebody else might have raced and expanded it already */
2125 if (address < vma->vm_start) {
2126 unsigned long size, grow;
2127
2128 size = vma->vm_end - address;
2129 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2130
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002131 error = -ENOMEM;
2132 if (grow <= vma->vm_pgoff) {
2133 error = acct_stack_growth(vma, size, grow);
2134 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002135 /*
2136 * vma_gap_update() doesn't support concurrent
2137 * updates, but we only hold a shared mmap_sem
2138 * lock here, so we need to protect against
2139 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002140 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002141 * we don't guarantee that all growable vmas
2142 * in a mm share the same root anon vma.
2143 * So, we reuse mm->page_table_lock to guard
2144 * against concurrent vma expansions.
2145 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002146 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002147 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002148 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002149 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002150 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002151 vma->vm_start = address;
2152 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002153 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002154 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002155 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002156
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002157 perf_event_mmap(vma);
2158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 }
2160 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002161 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002162 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002163 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 return error;
2165}
2166
Linus Torvalds09884962013-02-27 08:36:04 -08002167/*
2168 * Note how expand_stack() refuses to expand the stack all the way to
2169 * abut the next virtual mapping, *unless* that mapping itself is also
2170 * a stack mapping. We want to leave room for a guard page, after all
2171 * (the guard page itself is not added here, that is done by the
2172 * actual page faulting logic)
2173 *
2174 * This matches the behavior of the guard page logic (see mm/memory.c:
2175 * check_stack_guard_page()), which only allows the guard page to be
2176 * removed under these circumstances.
2177 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002178#ifdef CONFIG_STACK_GROWSUP
2179int expand_stack(struct vm_area_struct *vma, unsigned long address)
2180{
Linus Torvalds09884962013-02-27 08:36:04 -08002181 struct vm_area_struct *next;
2182
2183 address &= PAGE_MASK;
2184 next = vma->vm_next;
2185 if (next && next->vm_start == address + PAGE_SIZE) {
2186 if (!(next->vm_flags & VM_GROWSUP))
2187 return -ENOMEM;
2188 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002189 return expand_upwards(vma, address);
2190}
2191
2192struct vm_area_struct *
2193find_extend_vma(struct mm_struct *mm, unsigned long addr)
2194{
2195 struct vm_area_struct *vma, *prev;
2196
2197 addr &= PAGE_MASK;
2198 vma = find_vma_prev(mm, addr, &prev);
2199 if (vma && (vma->vm_start <= addr))
2200 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002201 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002202 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002203 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002204 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002205 return prev;
2206}
2207#else
2208int expand_stack(struct vm_area_struct *vma, unsigned long address)
2209{
Linus Torvalds09884962013-02-27 08:36:04 -08002210 struct vm_area_struct *prev;
2211
2212 address &= PAGE_MASK;
2213 prev = vma->vm_prev;
2214 if (prev && prev->vm_end == address) {
2215 if (!(prev->vm_flags & VM_GROWSDOWN))
2216 return -ENOMEM;
2217 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002218 return expand_downwards(vma, address);
2219}
2220
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002222find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223{
vishnu.pscc71aba2014-10-09 15:26:29 -07002224 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 unsigned long start;
2226
2227 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002228 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 if (!vma)
2230 return NULL;
2231 if (vma->vm_start <= addr)
2232 return vma;
2233 if (!(vma->vm_flags & VM_GROWSDOWN))
2234 return NULL;
2235 start = vma->vm_start;
2236 if (expand_stack(vma, addr))
2237 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002238 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002239 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 return vma;
2241}
2242#endif
2243
Jesse Barnese1d6d012014-12-12 16:55:27 -08002244EXPORT_SYMBOL_GPL(find_extend_vma);
2245
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002247 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002249 *
2250 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002252static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002254 unsigned long nr_accounted = 0;
2255
Hugh Dickins365e9c872005-10-29 18:16:18 -07002256 /* Update high watermark before we lower total_vm */
2257 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002259 long nrpages = vma_pages(vma);
2260
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002261 if (vma->vm_flags & VM_ACCOUNT)
2262 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002263 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002264 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002265 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002266 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 validate_mm(mm);
2268}
2269
2270/*
2271 * Get rid of page table information in the indicated region.
2272 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002273 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 */
2275static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002276 struct vm_area_struct *vma, struct vm_area_struct *prev,
2277 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278{
vishnu.pscc71aba2014-10-09 15:26:29 -07002279 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002280 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281
2282 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002283 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002284 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002285 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002286 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002287 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002288 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289}
2290
2291/*
2292 * Create a list of vma's touched by the unmap, removing them from the mm's
2293 * vma list as we go..
2294 */
2295static void
2296detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2297 struct vm_area_struct *prev, unsigned long end)
2298{
2299 struct vm_area_struct **insertion_point;
2300 struct vm_area_struct *tail_vma = NULL;
2301
2302 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002303 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002305 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 mm->map_count--;
2307 tail_vma = vma;
2308 vma = vma->vm_next;
2309 } while (vma && vma->vm_start < end);
2310 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002311 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002312 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002313 vma_gap_update(vma);
2314 } else
2315 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002317
2318 /* Kill the cache */
2319 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320}
2321
2322/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002323 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2324 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002326static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 unsigned long addr, int new_below)
2328{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002330 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Andi Kleena5516432008-07-23 21:27:41 -07002332 if (is_vm_hugetlb_page(vma) && (addr &
2333 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 return -EINVAL;
2335
Christoph Lametere94b1762006-12-06 20:33:17 -08002336 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002338 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
2340 /* most fields are the same, copy all, and then fixup */
2341 *new = *vma;
2342
Rik van Riel5beb4932010-03-05 13:42:07 -08002343 INIT_LIST_HEAD(&new->anon_vma_chain);
2344
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 if (new_below)
2346 new->vm_end = addr;
2347 else {
2348 new->vm_start = addr;
2349 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2350 }
2351
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002352 err = vma_dup_policy(vma, new);
2353 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002354 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002356 err = anon_vma_clone(new, vma);
2357 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002358 goto out_free_mpol;
2359
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002360 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 get_file(new->vm_file);
2362
2363 if (new->vm_ops && new->vm_ops->open)
2364 new->vm_ops->open(new);
2365
2366 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002367 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2369 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002370 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
Rik van Riel5beb4932010-03-05 13:42:07 -08002372 /* Success. */
2373 if (!err)
2374 return 0;
2375
2376 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002377 if (new->vm_ops && new->vm_ops->close)
2378 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002379 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002380 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002381 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002382 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002383 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002384 out_free_vma:
2385 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002386 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387}
2388
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002389/*
2390 * Split a vma into two pieces at address 'addr', a new vma is allocated
2391 * either for the first part or the tail.
2392 */
2393int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2394 unsigned long addr, int new_below)
2395{
2396 if (mm->map_count >= sysctl_max_map_count)
2397 return -ENOMEM;
2398
2399 return __split_vma(mm, vma, addr, new_below);
2400}
2401
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402/* Munmap is split into 2 main parts -- this part which finds
2403 * what needs doing, and the areas themselves, which do the
2404 * work. This now handles partial unmappings.
2405 * Jeremy Fitzhardinge <jeremy@goop.org>
2406 */
2407int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2408{
2409 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002410 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002412 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 return -EINVAL;
2414
vishnu.pscc71aba2014-10-09 15:26:29 -07002415 len = PAGE_ALIGN(len);
2416 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 return -EINVAL;
2418
2419 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002420 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002421 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002423 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002424 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
2426 /* if it doesn't overlap, we have nothing.. */
2427 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002428 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 return 0;
2430
2431 /*
2432 * If we need to split any vma, do it now to save pain later.
2433 *
2434 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2435 * unmapped vm_area_struct will remain in use: so lower split_vma
2436 * places tmp vma above, and higher split_vma places tmp vma below.
2437 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002438 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002439 int error;
2440
2441 /*
2442 * Make sure that map_count on return from munmap() will
2443 * not exceed its limit; but let map_count go just above
2444 * its limit temporarily, to help free resources as expected.
2445 */
2446 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2447 return -ENOMEM;
2448
2449 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (error)
2451 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002452 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 }
2454
2455 /* Does it split the last one? */
2456 last = find_vma(mm, end);
2457 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002458 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 if (error)
2460 return error;
2461 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002462 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
2464 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002465 * unlock any mlock()ed ranges before detaching vmas
2466 */
2467 if (mm->locked_vm) {
2468 struct vm_area_struct *tmp = vma;
2469 while (tmp && tmp->vm_start < end) {
2470 if (tmp->vm_flags & VM_LOCKED) {
2471 mm->locked_vm -= vma_pages(tmp);
2472 munlock_vma_pages_all(tmp);
2473 }
2474 tmp = tmp->vm_next;
2475 }
2476 }
2477
2478 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 * Remove the vma's, and unmap the actual pages
2480 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002481 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2482 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
Dave Hansen1de4fa12014-11-14 07:18:31 -08002484 arch_unmap(mm, vma, start, end);
2485
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002487 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
2489 return 0;
2490}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Al Virobfce2812012-04-20 21:57:04 -04002492int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002493{
2494 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002495 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002496
Michal Hockoae798782016-05-23 16:25:33 -07002497 if (down_write_killable(&mm->mmap_sem))
2498 return -EINTR;
2499
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002500 ret = do_munmap(mm, start, len);
2501 up_write(&mm->mmap_sem);
2502 return ret;
2503}
2504EXPORT_SYMBOL(vm_munmap);
2505
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002506SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507{
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002508 int ret;
2509 struct mm_struct *mm = current->mm;
2510
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 profile_munmap(addr);
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002512 if (down_write_killable(&mm->mmap_sem))
2513 return -EINTR;
2514 ret = do_munmap(mm, addr, len);
2515 up_write(&mm->mmap_sem);
2516 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517}
2518
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002519
2520/*
2521 * Emulation of deprecated remap_file_pages() syscall.
2522 */
2523SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2524 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2525{
2526
2527 struct mm_struct *mm = current->mm;
2528 struct vm_area_struct *vma;
2529 unsigned long populate = 0;
2530 unsigned long ret = -EINVAL;
2531 struct file *file;
2532
Joe Perches756a0252016-03-17 14:19:47 -07002533 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2534 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002535
2536 if (prot)
2537 return ret;
2538 start = start & PAGE_MASK;
2539 size = size & PAGE_MASK;
2540
2541 if (start + size <= start)
2542 return ret;
2543
2544 /* Does pgoff wrap? */
2545 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2546 return ret;
2547
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002548 if (down_write_killable(&mm->mmap_sem))
2549 return -EINTR;
2550
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002551 vma = find_vma(mm, start);
2552
2553 if (!vma || !(vma->vm_flags & VM_SHARED))
2554 goto out;
2555
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002556 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002557 goto out;
2558
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002559 if (start + size > vma->vm_end) {
2560 struct vm_area_struct *next;
2561
2562 for (next = vma->vm_next; next; next = next->vm_next) {
2563 /* hole between vmas ? */
2564 if (next->vm_start != next->vm_prev->vm_end)
2565 goto out;
2566
2567 if (next->vm_file != vma->vm_file)
2568 goto out;
2569
2570 if (next->vm_flags != vma->vm_flags)
2571 goto out;
2572
2573 if (start + size <= next->vm_end)
2574 break;
2575 }
2576
2577 if (!next)
2578 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002579 }
2580
2581 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2582 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2583 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2584
2585 flags &= MAP_NONBLOCK;
2586 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2587 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002588 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002589 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002590
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002591 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002592 for (tmp = vma; tmp->vm_start >= start + size;
2593 tmp = tmp->vm_next) {
2594 munlock_vma_pages_range(tmp,
2595 max(tmp->vm_start, start),
2596 min(tmp->vm_end, start + size));
2597 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002598 }
2599
2600 file = get_file(vma->vm_file);
2601 ret = do_mmap_pgoff(vma->vm_file, start, size,
2602 prot, flags, pgoff, &populate);
2603 fput(file);
2604out:
2605 up_write(&mm->mmap_sem);
2606 if (populate)
2607 mm_populate(ret, populate);
2608 if (!IS_ERR_VALUE(ret))
2609 ret = 0;
2610 return ret;
2611}
2612
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613static inline void verify_mm_writelocked(struct mm_struct *mm)
2614{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002615#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2617 WARN_ON(1);
2618 up_read(&mm->mmap_sem);
2619 }
2620#endif
2621}
2622
2623/*
2624 * this is really a simplified "do_mmap". it only handles
2625 * anonymous maps. eventually we may be able to do some
2626 * brk-specific accounting here.
2627 */
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002628static int do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629{
vishnu.pscc71aba2014-10-09 15:26:29 -07002630 struct mm_struct *mm = current->mm;
2631 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002633 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002635 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
2637 len = PAGE_ALIGN(len);
2638 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002639 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640
Kirill Korotaev3a459752006-09-07 14:17:04 +04002641 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2642
Al Viro2c6a1012009-12-03 19:40:46 -05002643 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002644 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002645 return error;
2646
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002647 error = mlock_future_check(mm, mm->def_flags, len);
2648 if (error)
2649 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
2651 /*
2652 * mm->mmap_sem is required to protect against another thread
2653 * changing the mappings in case we sleep.
2654 */
2655 verify_mm_writelocked(mm);
2656
2657 /*
2658 * Clear old maps. this also does some error checking for us
2659 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002660 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2661 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 if (do_munmap(mm, addr, len))
2663 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 }
2665
2666 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002667 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 return -ENOMEM;
2669
2670 if (mm->map_count > sysctl_max_map_count)
2671 return -ENOMEM;
2672
Al Viro191c5422012-02-13 03:58:52 +00002673 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return -ENOMEM;
2675
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002677 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002678 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002679 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 goto out;
2681
2682 /*
2683 * create a vma struct for an anonymous mapping
2684 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002685 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 if (!vma) {
2687 vm_unacct_memory(len >> PAGE_SHIFT);
2688 return -ENOMEM;
2689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
Rik van Riel5beb4932010-03-05 13:42:07 -08002691 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 vma->vm_mm = mm;
2693 vma->vm_start = addr;
2694 vma->vm_end = addr + len;
2695 vma->vm_pgoff = pgoff;
2696 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002697 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 vma_link(mm, vma, prev, rb_link, rb_parent);
2699out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002700 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002702 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002703 if (flags & VM_LOCKED)
2704 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002705 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002706 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707}
2708
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002709int vm_brk(unsigned long addr, unsigned long len)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002710{
2711 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002712 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002713 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002714
Michal Hocko2d6c9282016-05-23 16:25:42 -07002715 if (down_write_killable(&mm->mmap_sem))
2716 return -EINTR;
2717
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002718 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002719 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002720 up_write(&mm->mmap_sem);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002721 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002722 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002723 return ret;
2724}
2725EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726
2727/* Release all mmaps. */
2728void exit_mmap(struct mm_struct *mm)
2729{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002730 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002731 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 unsigned long nr_accounted = 0;
2733
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002734 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002735 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002736
Rik van Rielba470de2008-10-18 20:26:50 -07002737 if (mm->locked_vm) {
2738 vma = mm->mmap;
2739 while (vma) {
2740 if (vma->vm_flags & VM_LOCKED)
2741 munlock_vma_pages_all(vma);
2742 vma = vma->vm_next;
2743 }
2744 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002745
2746 arch_exit_mmap(mm);
2747
Rik van Rielba470de2008-10-18 20:26:50 -07002748 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002749 if (!vma) /* Can happen if dup_mmap() received an OOM */
2750 return;
2751
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002754 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002755 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002756 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002757 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002758
Hugh Dickins6ee86302013-04-29 15:07:44 -07002759 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002760 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002763 * Walk the list again, actually closing and freeing it,
2764 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002766 while (vma) {
2767 if (vma->vm_flags & VM_ACCOUNT)
2768 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002769 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002770 }
2771 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772}
2773
2774/* Insert vm structure into process list sorted by address
2775 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002776 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002778int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779{
Hugh Dickins6597d782012-10-08 16:29:07 -07002780 struct vm_area_struct *prev;
2781 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782
Chen Gangc9d13f52015-09-08 15:04:08 -07002783 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2784 &prev, &rb_link, &rb_parent))
2785 return -ENOMEM;
2786 if ((vma->vm_flags & VM_ACCOUNT) &&
2787 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2788 return -ENOMEM;
2789
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 /*
2791 * The vm_pgoff of a purely anonymous vma should be irrelevant
2792 * until its first write fault, when page's anon_vma and index
2793 * are set. But now set the vm_pgoff it will almost certainly
2794 * end up with (unless mremap moves it elsewhere before that
2795 * first wfault), so /proc/pid/maps tells a consistent story.
2796 *
2797 * By setting it to reflect the virtual start address of the
2798 * vma, merges and splits can happen in a seamless way, just
2799 * using the existing file pgoff checks and manipulations.
2800 * Similarly in do_mmap_pgoff and in do_brk.
2801 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002802 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 BUG_ON(vma->anon_vma);
2804 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2805 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302806
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 vma_link(mm, vma, prev, rb_link, rb_parent);
2808 return 0;
2809}
2810
2811/*
2812 * Copy the vma structure to a new location in the same mm,
2813 * prior to moving page table entries, to effect an mremap move.
2814 */
2815struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002816 unsigned long addr, unsigned long len, pgoff_t pgoff,
2817 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818{
2819 struct vm_area_struct *vma = *vmap;
2820 unsigned long vma_start = vma->vm_start;
2821 struct mm_struct *mm = vma->vm_mm;
2822 struct vm_area_struct *new_vma, *prev;
2823 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002824 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825
2826 /*
2827 * If anonymous vma has not yet been faulted, update new pgoff
2828 * to match new location, to increase its chance of merging.
2829 */
Oleg Nesterovce757992015-09-08 14:58:34 -07002830 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002832 faulted_in_anon_vma = false;
2833 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834
Hugh Dickins6597d782012-10-08 16:29:07 -07002835 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2836 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002838 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2839 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 if (new_vma) {
2841 /*
2842 * Source vma may have been merged into new_vma
2843 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002844 if (unlikely(vma_start >= new_vma->vm_start &&
2845 vma_start < new_vma->vm_end)) {
2846 /*
2847 * The only way we can get a vma_merge with
2848 * self during an mremap is if the vma hasn't
2849 * been faulted in yet and we were allowed to
2850 * reset the dst vma->vm_pgoff to the
2851 * destination address of the mremap to allow
2852 * the merge to happen. mremap must change the
2853 * vm_pgoff linearity between src and dst vmas
2854 * (in turn preventing a vma_merge) to be
2855 * safe. It is only safe to keep the vm_pgoff
2856 * linear if there are no pages mapped yet.
2857 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002858 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002859 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002860 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002861 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002863 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07002864 if (!new_vma)
2865 goto out;
2866 *new_vma = *vma;
2867 new_vma->vm_start = addr;
2868 new_vma->vm_end = addr + len;
2869 new_vma->vm_pgoff = pgoff;
2870 if (vma_dup_policy(vma, new_vma))
2871 goto out_free_vma;
2872 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2873 if (anon_vma_clone(new_vma, vma))
2874 goto out_free_mempol;
2875 if (new_vma->vm_file)
2876 get_file(new_vma->vm_file);
2877 if (new_vma->vm_ops && new_vma->vm_ops->open)
2878 new_vma->vm_ops->open(new_vma);
2879 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2880 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 }
2882 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002883
Chen Gange3975892015-09-08 15:03:38 -07002884out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002885 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07002886out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08002887 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07002888out:
Rik van Riel5beb4932010-03-05 13:42:07 -08002889 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002891
2892/*
2893 * Return true if the calling process may expand its vm space by the passed
2894 * number of pages
2895 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002896bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07002897{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002898 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
2899 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002900
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002901 if (is_data_mapping(flags) &&
2902 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07002903 /* Workaround for Valgrind */
2904 if (rlimit(RLIMIT_DATA) == 0 &&
2905 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
2906 return true;
2907 if (!ignore_rlimit_data) {
2908 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n",
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002909 current->comm, current->pid,
2910 (mm->data_vm + npages) << PAGE_SHIFT,
2911 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002912 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07002913 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002914 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07002915
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002916 return true;
2917}
2918
2919void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
2920{
2921 mm->total_vm += npages;
2922
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002923 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002924 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002925 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002926 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002927 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002928 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002929}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002930
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002931static int special_mapping_fault(struct vm_area_struct *vma,
2932 struct vm_fault *vmf);
2933
2934/*
2935 * Having a close hook prevents vma merging regardless of flags.
2936 */
2937static void special_mapping_close(struct vm_area_struct *vma)
2938{
2939}
2940
2941static const char *special_mapping_name(struct vm_area_struct *vma)
2942{
2943 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
2944}
2945
Dmitry Safonovb059a452016-06-28 14:35:38 +03002946static int special_mapping_mremap(struct vm_area_struct *new_vma)
2947{
2948 struct vm_special_mapping *sm = new_vma->vm_private_data;
2949
2950 if (sm->mremap)
2951 return sm->mremap(sm, new_vma);
2952 return 0;
2953}
2954
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002955static const struct vm_operations_struct special_mapping_vmops = {
2956 .close = special_mapping_close,
2957 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03002958 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002959 .name = special_mapping_name,
2960};
2961
2962static const struct vm_operations_struct legacy_special_mapping_vmops = {
2963 .close = special_mapping_close,
2964 .fault = special_mapping_fault,
2965};
Roland McGrathfa5dc222007-02-08 14:20:41 -08002966
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002967static int special_mapping_fault(struct vm_area_struct *vma,
2968 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002969{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002970 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002971 struct page **pages;
2972
Andy Lutomirskif872f542015-12-29 20:12:19 -08002973 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002974 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08002975 } else {
2976 struct vm_special_mapping *sm = vma->vm_private_data;
2977
2978 if (sm->fault)
2979 return sm->fault(sm, vma, vmf);
2980
2981 pages = sm->pages;
2982 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002983
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002984 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002985 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002986
2987 if (*pages) {
2988 struct page *page = *pages;
2989 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002990 vmf->page = page;
2991 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002992 }
2993
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002994 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002995}
2996
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002997static struct vm_area_struct *__install_special_mapping(
2998 struct mm_struct *mm,
2999 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003000 unsigned long vm_flags, void *priv,
3001 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003002{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003003 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003004 struct vm_area_struct *vma;
3005
3006 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3007 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003008 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003009
Rik van Riel5beb4932010-03-05 13:42:07 -08003010 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003011 vma->vm_mm = mm;
3012 vma->vm_start = addr;
3013 vma->vm_end = addr + len;
3014
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003015 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003016 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003017
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003018 vma->vm_ops = ops;
3019 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003020
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003021 ret = insert_vm_struct(mm, vma);
3022 if (ret)
3023 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003024
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003025 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003026
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003027 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003028
Stefani Seibold3935ed62014-03-17 23:22:02 +01003029 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003030
3031out:
3032 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003033 return ERR_PTR(ret);
3034}
3035
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003036/*
3037 * Called with mm->mmap_sem held for writing.
3038 * Insert a new vma covering the given region, with the given flags.
3039 * Its pages are supplied by the given array of struct page *.
3040 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3041 * The region past the last page supplied will always produce SIGBUS.
3042 * The array pointer and the pages it points to are assumed to stay alive
3043 * for as long as this mapping might exist.
3044 */
3045struct vm_area_struct *_install_special_mapping(
3046 struct mm_struct *mm,
3047 unsigned long addr, unsigned long len,
3048 unsigned long vm_flags, const struct vm_special_mapping *spec)
3049{
Chen Gang27f28b92015-11-05 18:48:41 -08003050 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3051 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003052}
3053
Stefani Seibold3935ed62014-03-17 23:22:02 +01003054int install_special_mapping(struct mm_struct *mm,
3055 unsigned long addr, unsigned long len,
3056 unsigned long vm_flags, struct page **pages)
3057{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003058 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003059 mm, addr, len, vm_flags, (void *)pages,
3060 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003061
Duan Jiong14bd5b42014-06-04 16:07:05 -07003062 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003063}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003064
3065static DEFINE_MUTEX(mm_all_locks_mutex);
3066
Peter Zijlstra454ed842008-08-11 09:30:25 +02003067static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003068{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003069 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003070 /*
3071 * The LSB of head.next can't change from under us
3072 * because we hold the mm_all_locks_mutex.
3073 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003074 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003075 /*
3076 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003077 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003078 * the same anon_vma we won't take it again.
3079 *
3080 * No need of atomic instructions here, head.next
3081 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003082 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003083 */
3084 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003085 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003086 BUG();
3087 }
3088}
3089
Peter Zijlstra454ed842008-08-11 09:30:25 +02003090static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003091{
3092 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3093 /*
3094 * AS_MM_ALL_LOCKS can't change from under us because
3095 * we hold the mm_all_locks_mutex.
3096 *
3097 * Operations on ->flags have to be atomic because
3098 * even if AS_MM_ALL_LOCKS is stable thanks to the
3099 * mm_all_locks_mutex, there may be other cpus
3100 * changing other bitflags in parallel to us.
3101 */
3102 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3103 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003104 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003105 }
3106}
3107
3108/*
3109 * This operation locks against the VM for all pte/vma/mm related
3110 * operations that could ever happen on a certain mm. This includes
3111 * vmtruncate, try_to_unmap, and all page faults.
3112 *
3113 * The caller must take the mmap_sem in write mode before calling
3114 * mm_take_all_locks(). The caller isn't allowed to release the
3115 * mmap_sem until mm_drop_all_locks() returns.
3116 *
3117 * mmap_sem in write mode is required in order to block all operations
3118 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003119 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003120 * anon_vmas to be associated with existing vmas.
3121 *
3122 * A single task can't take more than one mm_take_all_locks() in a row
3123 * or it would deadlock.
3124 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003125 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003126 * mapping->flags avoid to take the same lock twice, if more than one
3127 * vma in this mm is backed by the same anon_vma or address_space.
3128 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003129 * We take locks in following order, accordingly to comment at beginning
3130 * of mm/rmap.c:
3131 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3132 * hugetlb mapping);
3133 * - all i_mmap_rwsem locks;
3134 * - all anon_vma->rwseml
3135 *
3136 * We can take all locks within these types randomly because the VM code
3137 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3138 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003139 *
3140 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3141 * that may have to take thousand of locks.
3142 *
3143 * mm_take_all_locks() can fail if it's interrupted by signals.
3144 */
3145int mm_take_all_locks(struct mm_struct *mm)
3146{
3147 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003148 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003149
3150 BUG_ON(down_read_trylock(&mm->mmap_sem));
3151
3152 mutex_lock(&mm_all_locks_mutex);
3153
3154 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3155 if (signal_pending(current))
3156 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003157 if (vma->vm_file && vma->vm_file->f_mapping &&
3158 is_vm_hugetlb_page(vma))
3159 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3160 }
3161
3162 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3163 if (signal_pending(current))
3164 goto out_unlock;
3165 if (vma->vm_file && vma->vm_file->f_mapping &&
3166 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003167 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003168 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003169
3170 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3171 if (signal_pending(current))
3172 goto out_unlock;
3173 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003174 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3175 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003176 }
3177
Kautuk Consul584cff52011-10-31 17:08:59 -07003178 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003179
3180out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003181 mm_drop_all_locks(mm);
3182 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003183}
3184
3185static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3186{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003187 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003188 /*
3189 * The LSB of head.next can't change to 0 from under
3190 * us because we hold the mm_all_locks_mutex.
3191 *
3192 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003193 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003194 * never see our bitflag.
3195 *
3196 * No need of atomic instructions here, head.next
3197 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003198 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003199 */
3200 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003201 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003202 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003203 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003204 }
3205}
3206
3207static void vm_unlock_mapping(struct address_space *mapping)
3208{
3209 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3210 /*
3211 * AS_MM_ALL_LOCKS can't change to 0 from under us
3212 * because we hold the mm_all_locks_mutex.
3213 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003214 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003215 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3216 &mapping->flags))
3217 BUG();
3218 }
3219}
3220
3221/*
3222 * The mmap_sem cannot be released by the caller until
3223 * mm_drop_all_locks() returns.
3224 */
3225void mm_drop_all_locks(struct mm_struct *mm)
3226{
3227 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003228 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003229
3230 BUG_ON(down_read_trylock(&mm->mmap_sem));
3231 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3232
3233 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3234 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003235 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3236 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003237 if (vma->vm_file && vma->vm_file->f_mapping)
3238 vm_unlock_mapping(vma->vm_file->f_mapping);
3239 }
3240
3241 mutex_unlock(&mm_all_locks_mutex);
3242}
David Howells8feae132009-01-08 12:04:47 +00003243
3244/*
3245 * initialise the VMA slab
3246 */
3247void __init mmap_init(void)
3248{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003249 int ret;
3250
Tejun Heo908c7f12014-09-08 09:51:29 +09003251 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003252 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003253}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003254
3255/*
3256 * Initialise sysctl_user_reserve_kbytes.
3257 *
3258 * This is intended to prevent a user from starting a single memory hogging
3259 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3260 * mode.
3261 *
3262 * The default value is min(3% of free memory, 128MB)
3263 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3264 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003265static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003266{
3267 unsigned long free_kbytes;
3268
3269 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3270
3271 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3272 return 0;
3273}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003274subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003275
3276/*
3277 * Initialise sysctl_admin_reserve_kbytes.
3278 *
3279 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3280 * to log in and kill a memory hogging process.
3281 *
3282 * Systems with more than 256MB will reserve 8MB, enough to recover
3283 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3284 * only reserve 3% of free pages by default.
3285 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003286static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003287{
3288 unsigned long free_kbytes;
3289
3290 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3291
3292 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3293 return 0;
3294}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003295subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003296
3297/*
3298 * Reinititalise user and admin reserves if memory is added or removed.
3299 *
3300 * The default user reserve max is 128MB, and the default max for the
3301 * admin reserve is 8MB. These are usually, but not always, enough to
3302 * enable recovery from a memory hogging process using login/sshd, a shell,
3303 * and tools like top. It may make sense to increase or even disable the
3304 * reserve depending on the existence of swap or variations in the recovery
3305 * tools. So, the admin may have changed them.
3306 *
3307 * If memory is added and the reserves have been eliminated or increased above
3308 * the default max, then we'll trust the admin.
3309 *
3310 * If memory is removed and there isn't enough free memory, then we
3311 * need to reset the reserves.
3312 *
3313 * Otherwise keep the reserve set by the admin.
3314 */
3315static int reserve_mem_notifier(struct notifier_block *nb,
3316 unsigned long action, void *data)
3317{
3318 unsigned long tmp, free_kbytes;
3319
3320 switch (action) {
3321 case MEM_ONLINE:
3322 /* Default max is 128MB. Leave alone if modified by operator. */
3323 tmp = sysctl_user_reserve_kbytes;
3324 if (0 < tmp && tmp < (1UL << 17))
3325 init_user_reserve();
3326
3327 /* Default max is 8MB. Leave alone if modified by operator. */
3328 tmp = sysctl_admin_reserve_kbytes;
3329 if (0 < tmp && tmp < (1UL << 13))
3330 init_admin_reserve();
3331
3332 break;
3333 case MEM_OFFLINE:
3334 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3335
3336 if (sysctl_user_reserve_kbytes > free_kbytes) {
3337 init_user_reserve();
3338 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3339 sysctl_user_reserve_kbytes);
3340 }
3341
3342 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3343 init_admin_reserve();
3344 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3345 sysctl_admin_reserve_kbytes);
3346 }
3347 break;
3348 default:
3349 break;
3350 }
3351 return NOTIFY_OK;
3352}
3353
3354static struct notifier_block reserve_mem_nb = {
3355 .notifier_call = reserve_mem_notifier,
3356};
3357
3358static int __meminit init_reserve_notifier(void)
3359{
3360 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003361 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003362
3363 return 0;
3364}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003365subsys_initcall(init_reserve_notifier);