blob: bd2e1a533bc182a113523bd16b1af09e81a3c8ea [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
Martin Schwidefsky08e7d9b2008-02-04 22:29:16 -080058#ifndef arch_rebalance_pgtables
59#define arch_rebalance_pgtables(addr, len) (addr)
60#endif
61
Daniel Cashmand07e2252016-01-14 15:19:53 -080062#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
63const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
64const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
65int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
66#endif
67#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
68const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
69const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
70int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
71#endif
72
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080073static bool ignore_rlimit_data = true;
74core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
Daniel Cashmand07e2252016-01-14 15:19:53 -080075
Hugh Dickinse0da3822005-04-19 13:29:15 -070076static void unmap_region(struct mm_struct *mm,
77 struct vm_area_struct *vma, struct vm_area_struct *prev,
78 unsigned long start, unsigned long end);
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080/* description of effects of mapping type and prot in current implementation.
81 * this is due to the limited x86 page protection hardware. The expected
82 * behavior is in parens:
83 *
84 * map_type prot
85 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
86 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 * w: (no) no w: (no) no w: (yes) yes w: (no) no
88 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070089 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
91 * w: (no) no w: (no) no w: (copy) copy w: (no) no
92 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
93 *
94 */
95pgprot_t protection_map[16] = {
96 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
97 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
98};
99
Hugh Dickins804af2c2006-07-26 21:39:49 +0100100pgprot_t vm_get_page_prot(unsigned long vm_flags)
101{
Dave Kleikampb845f312008-07-08 00:28:51 +1000102 return __pgprot(pgprot_val(protection_map[vm_flags &
103 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
104 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100105}
106EXPORT_SYMBOL(vm_get_page_prot);
107
Peter Feiner64e45502014-10-13 15:55:46 -0700108static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
109{
110 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
111}
112
113/* Update vma->vm_page_prot to reflect vma->vm_flags. */
114void vma_set_page_prot(struct vm_area_struct *vma)
115{
116 unsigned long vm_flags = vma->vm_flags;
117
118 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
119 if (vma_wants_writenotify(vma)) {
120 vm_flags &= ~VM_SHARED;
121 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
122 vm_flags);
123 }
124}
125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800127 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 */
129static void __remove_shared_vm_struct(struct vm_area_struct *vma,
130 struct file *file, struct address_space *mapping)
131{
132 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500133 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700135 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
137 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800138 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 flush_dcache_mmap_unlock(mapping);
140}
141
142/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700143 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700144 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700146void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147{
148 struct file *file = vma->vm_file;
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 if (file) {
151 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800152 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800154 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700156}
157
158/*
159 * Close a vm structure and free it, returning the next.
160 */
161static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
162{
163 struct vm_area_struct *next = vma->vm_next;
164
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700165 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 if (vma->vm_ops && vma->vm_ops->close)
167 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700168 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700169 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700170 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700172 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700175static unsigned long do_brk(unsigned long addr, unsigned long len);
176
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100177SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700179 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 unsigned long newbrk, oldbrk;
181 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700182 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800183 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
185 down_write(&mm->mmap_sem);
186
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700187#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800188 /*
189 * CONFIG_COMPAT_BRK can still be overridden by setting
190 * randomize_va_space to 2, which will still cause mm->start_brk
191 * to be arbitrarily shifted
192 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700193 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800194 min_brk = mm->start_brk;
195 else
196 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700197#else
198 min_brk = mm->start_brk;
199#endif
200 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700202
203 /*
204 * Check against rlimit here. If this check is done later after the test
205 * of oldbrk with newbrk then it can escape the test and let the data
206 * segment grow beyond its set limit the in case where the limit is
207 * not page aligned -Ram Gupta
208 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700209 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
210 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700211 goto out;
212
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 newbrk = PAGE_ALIGN(brk);
214 oldbrk = PAGE_ALIGN(mm->brk);
215 if (oldbrk == newbrk)
216 goto set_brk;
217
218 /* Always allow shrinking brk. */
219 if (brk <= mm->brk) {
220 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
221 goto set_brk;
222 goto out;
223 }
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 /* Check against existing mmap mappings. */
226 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
227 goto out;
228
229 /* Ok, looks good - let it rip. */
230 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
231 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233set_brk:
234 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800235 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
236 up_write(&mm->mmap_sem);
237 if (populate)
238 mm_populate(oldbrk, newbrk - oldbrk);
239 return brk;
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241out:
242 retval = mm->brk;
243 up_write(&mm->mmap_sem);
244 return retval;
245}
246
Michel Lespinassed3737182012-12-11 16:01:38 -0800247static long vma_compute_subtree_gap(struct vm_area_struct *vma)
248{
249 unsigned long max, subtree_gap;
250 max = vma->vm_start;
251 if (vma->vm_prev)
252 max -= vma->vm_prev->vm_end;
253 if (vma->vm_rb.rb_left) {
254 subtree_gap = rb_entry(vma->vm_rb.rb_left,
255 struct vm_area_struct, vm_rb)->rb_subtree_gap;
256 if (subtree_gap > max)
257 max = subtree_gap;
258 }
259 if (vma->vm_rb.rb_right) {
260 subtree_gap = rb_entry(vma->vm_rb.rb_right,
261 struct vm_area_struct, vm_rb)->rb_subtree_gap;
262 if (subtree_gap > max)
263 max = subtree_gap;
264 }
265 return max;
266}
267
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700268#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800269static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800271 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800272 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 struct rb_node *nd, *pn = NULL;
274 unsigned long prev = 0, pend = 0;
275
276 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
277 struct vm_area_struct *vma;
278 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800279 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700280 pr_emerg("vm_start %lx < prev %lx\n",
281 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800282 bug = 1;
283 }
284 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700285 pr_emerg("vm_start %lx < pend %lx\n",
286 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800287 bug = 1;
288 }
289 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700290 pr_emerg("vm_start %lx > vm_end %lx\n",
291 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800292 bug = 1;
293 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800294 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800295 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700296 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800297 vma->rb_subtree_gap,
298 vma_compute_subtree_gap(vma));
299 bug = 1;
300 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800301 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 i++;
303 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800304 prev = vma->vm_start;
305 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 }
307 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800308 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800310 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700311 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800312 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800314 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
316
Michel Lespinassed3737182012-12-11 16:01:38 -0800317static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
318{
319 struct rb_node *nd;
320
321 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
322 struct vm_area_struct *vma;
323 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700324 VM_BUG_ON_VMA(vma != ignore &&
325 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
326 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328}
329
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700330static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
332 int bug = 0;
333 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800334 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700335 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700336
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700337 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800338 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700339 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700340
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800341 if (anon_vma) {
342 anon_vma_lock_read(anon_vma);
343 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
344 anon_vma_interval_tree_verify(avc);
345 anon_vma_unlock_read(anon_vma);
346 }
347
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800348 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700349 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 i++;
351 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800352 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700353 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800354 bug = 1;
355 }
356 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700357 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700358 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800359 bug = 1;
360 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800361 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800362 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700363 if (i != -1)
364 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800365 bug = 1;
366 }
Sasha Levin96dad672014-10-09 15:28:39 -0700367 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
369#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800370#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371#define validate_mm(mm) do { } while (0)
372#endif
373
Michel Lespinassed3737182012-12-11 16:01:38 -0800374RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
375 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
376
377/*
378 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
379 * vma->vm_prev->vm_end values changed, without modifying the vma's position
380 * in the rbtree.
381 */
382static void vma_gap_update(struct vm_area_struct *vma)
383{
384 /*
385 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
386 * function that does exacltly what we want.
387 */
388 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
389}
390
391static inline void vma_rb_insert(struct vm_area_struct *vma,
392 struct rb_root *root)
393{
394 /* All rb_subtree_gap values must be consistent prior to insertion */
395 validate_mm_rb(root, NULL);
396
397 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
398}
399
400static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
401{
402 /*
403 * All rb_subtree_gap values must be consistent prior to erase,
404 * with the possible exception of the vma being erased.
405 */
406 validate_mm_rb(root, vma);
407
408 /*
409 * Note rb_erase_augmented is a fairly large inline function,
410 * so make sure we instantiate it only once with our desired
411 * augmented rbtree callbacks.
412 */
413 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
414}
415
Michel Lespinassebf181b92012-10-08 16:31:39 -0700416/*
417 * vma has some anon_vma assigned, and is already inserted on that
418 * anon_vma's interval trees.
419 *
420 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
421 * vma must be removed from the anon_vma's interval trees using
422 * anon_vma_interval_tree_pre_update_vma().
423 *
424 * After the update, the vma will be reinserted using
425 * anon_vma_interval_tree_post_update_vma().
426 *
427 * The entire update must be protected by exclusive mmap_sem and by
428 * the root anon_vma's mutex.
429 */
430static inline void
431anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
432{
433 struct anon_vma_chain *avc;
434
435 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
436 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
437}
438
439static inline void
440anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
441{
442 struct anon_vma_chain *avc;
443
444 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
445 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
446}
447
Hugh Dickins6597d782012-10-08 16:29:07 -0700448static int find_vma_links(struct mm_struct *mm, unsigned long addr,
449 unsigned long end, struct vm_area_struct **pprev,
450 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
Hugh Dickins6597d782012-10-08 16:29:07 -0700452 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 __rb_link = &mm->mm_rb.rb_node;
455 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 while (*__rb_link) {
458 struct vm_area_struct *vma_tmp;
459
460 __rb_parent = *__rb_link;
461 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
462
463 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700464 /* Fail if an existing vma overlaps the area */
465 if (vma_tmp->vm_start < end)
466 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 __rb_link = &__rb_parent->rb_left;
468 } else {
469 rb_prev = __rb_parent;
470 __rb_link = &__rb_parent->rb_right;
471 }
472 }
473
474 *pprev = NULL;
475 if (rb_prev)
476 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
477 *rb_link = __rb_link;
478 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700479 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480}
481
Cyril Hrubise8420a82013-04-29 15:08:33 -0700482static unsigned long count_vma_pages_range(struct mm_struct *mm,
483 unsigned long addr, unsigned long end)
484{
485 unsigned long nr_pages = 0;
486 struct vm_area_struct *vma;
487
488 /* Find first overlaping mapping */
489 vma = find_vma_intersection(mm, addr, end);
490 if (!vma)
491 return 0;
492
493 nr_pages = (min(end, vma->vm_end) -
494 max(addr, vma->vm_start)) >> PAGE_SHIFT;
495
496 /* Iterate over the rest of the overlaps */
497 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
498 unsigned long overlap_len;
499
500 if (vma->vm_start > end)
501 break;
502
503 overlap_len = min(end, vma->vm_end) - vma->vm_start;
504 nr_pages += overlap_len >> PAGE_SHIFT;
505 }
506
507 return nr_pages;
508}
509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
511 struct rb_node **rb_link, struct rb_node *rb_parent)
512{
Michel Lespinassed3737182012-12-11 16:01:38 -0800513 /* Update tracking information for the gap following the new vma. */
514 if (vma->vm_next)
515 vma_gap_update(vma->vm_next);
516 else
517 mm->highest_vm_end = vma->vm_end;
518
519 /*
520 * vma->vm_prev wasn't known when we followed the rbtree to find the
521 * correct insertion point for that vma. As a result, we could not
522 * update the vma vm_rb parents rb_subtree_gap values on the way down.
523 * So, we first insert the vma with a zero rb_subtree_gap value
524 * (to be consistent with what we did on the way down), and then
525 * immediately update the gap to the correct value. Finally we
526 * rebalance the rbtree after all augmented values have been set.
527 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800529 vma->rb_subtree_gap = 0;
530 vma_gap_update(vma);
531 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532}
533
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700534static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535{
ZhenwenXu48aae422009-01-06 14:40:21 -0800536 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 file = vma->vm_file;
539 if (file) {
540 struct address_space *mapping = file->f_mapping;
541
542 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500543 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700545 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800548 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 flush_dcache_mmap_unlock(mapping);
550 }
551}
552
553static void
554__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
555 struct vm_area_struct *prev, struct rb_node **rb_link,
556 struct rb_node *rb_parent)
557{
558 __vma_link_list(mm, vma, prev, rb_parent);
559 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
562static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
563 struct vm_area_struct *prev, struct rb_node **rb_link,
564 struct rb_node *rb_parent)
565{
566 struct address_space *mapping = NULL;
567
Huang Shijie64ac4942014-06-04 16:07:33 -0700568 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800570 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 __vma_link(mm, vma, prev, rb_link, rb_parent);
574 __vma_link_file(vma);
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800577 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579 mm->map_count++;
580 validate_mm(mm);
581}
582
583/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700584 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700585 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800587static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
Hugh Dickins6597d782012-10-08 16:29:07 -0700589 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800590 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Hugh Dickins6597d782012-10-08 16:29:07 -0700592 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
593 &prev, &rb_link, &rb_parent))
594 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 __vma_link(mm, vma, prev, rb_link, rb_parent);
596 mm->map_count++;
597}
598
599static inline void
600__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
601 struct vm_area_struct *prev)
602{
Michel Lespinassed3737182012-12-11 16:01:38 -0800603 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700604
Michel Lespinassed3737182012-12-11 16:01:38 -0800605 vma_rb_erase(vma, &mm->mm_rb);
606 prev->vm_next = next = vma->vm_next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700607 if (next)
608 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700609
610 /* Kill the cache */
611 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
614/*
615 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
616 * is already present in an i_mmap tree without adjusting the tree.
617 * The following helper function should be used when such adjustments
618 * are necessary. The "insert" vma (if any) is to be inserted
619 * before we drop the necessary locks.
620 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800621int vma_adjust(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
623{
624 struct mm_struct *mm = vma->vm_mm;
625 struct vm_area_struct *next = vma->vm_next;
626 struct vm_area_struct *importer = NULL;
627 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700628 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700629 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800631 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 long adjust_next = 0;
633 int remove_next = 0;
634
635 if (next && !insert) {
Linus Torvalds287d97a2010-04-10 15:22:30 -0700636 struct vm_area_struct *exporter = NULL;
637
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (end >= next->vm_end) {
639 /*
640 * vma expands, overlapping all the next, and
641 * perhaps the one after too (mprotect case 6).
642 */
643again: remove_next = 1 + (end > next->vm_end);
644 end = next->vm_end;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700645 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 importer = vma;
647 } else if (end > next->vm_start) {
648 /*
649 * vma expands, overlapping part of the next:
650 * mprotect case 5 shifting the boundary up.
651 */
652 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700653 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 importer = vma;
655 } else if (end < vma->vm_end) {
656 /*
657 * vma shrinks, and !insert tells it's not
658 * split_vma inserting another: so it must be
659 * mprotect case 4 shifting the boundary down.
660 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700661 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700662 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 importer = next;
664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Rik van Riel5beb4932010-03-05 13:42:07 -0800666 /*
667 * Easily overlooked: when mprotect shifts the boundary,
668 * make sure the expanding vma has anon_vma set if the
669 * shrinking vma had, to cover any anon pages imported.
670 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700671 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800672 int error;
673
Linus Torvalds287d97a2010-04-10 15:22:30 -0700674 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300675 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700676 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300677 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800678 }
679 }
680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 if (file) {
682 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800683 root = &mapping->i_mmap;
684 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530685
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800686 if (adjust_next)
687 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530688
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800689 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700692 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 * are visible to arm/parisc __flush_dcache_page
694 * throughout; but we cannot insert into address
695 * space until vma start or end is updated.
696 */
697 __vma_link_file(insert);
698 }
699 }
700
Andrea Arcangeli94fcc582011-01-13 15:47:08 -0800701 vma_adjust_trans_huge(vma, start, end, adjust_next);
702
Michel Lespinassebf181b92012-10-08 16:31:39 -0700703 anon_vma = vma->anon_vma;
704 if (!anon_vma && adjust_next)
705 anon_vma = next->anon_vma;
706 if (anon_vma) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700707 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
708 anon_vma != next->anon_vma, next);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000709 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700710 anon_vma_interval_tree_pre_update_vma(vma);
711 if (adjust_next)
712 anon_vma_interval_tree_pre_update_vma(next);
713 }
Rik van Riel012f18002010-08-09 17:18:40 -0700714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (root) {
716 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700717 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700719 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
721
Michel Lespinassed3737182012-12-11 16:01:38 -0800722 if (start != vma->vm_start) {
723 vma->vm_start = start;
724 start_changed = true;
725 }
726 if (end != vma->vm_end) {
727 vma->vm_end = end;
728 end_changed = true;
729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 vma->vm_pgoff = pgoff;
731 if (adjust_next) {
732 next->vm_start += adjust_next << PAGE_SHIFT;
733 next->vm_pgoff += adjust_next;
734 }
735
736 if (root) {
737 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700738 vma_interval_tree_insert(next, root);
739 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 flush_dcache_mmap_unlock(mapping);
741 }
742
743 if (remove_next) {
744 /*
745 * vma_merge has merged next into vma, and needs
746 * us to remove next before dropping the locks.
747 */
748 __vma_unlink(mm, next, vma);
749 if (file)
750 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 } else if (insert) {
752 /*
753 * split_vma has split insert from vma, and needs
754 * us to insert it before dropping the locks
755 * (it may either follow vma or precede it).
756 */
757 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800758 } else {
759 if (start_changed)
760 vma_gap_update(vma);
761 if (end_changed) {
762 if (!next)
763 mm->highest_vm_end = end;
764 else if (!adjust_next)
765 vma_gap_update(next);
766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
768
Michel Lespinassebf181b92012-10-08 16:31:39 -0700769 if (anon_vma) {
770 anon_vma_interval_tree_post_update_vma(vma);
771 if (adjust_next)
772 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800773 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800776 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530778 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100779 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530780
781 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100782 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530783 }
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700786 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530787 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700789 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800790 if (next->anon_vma)
791 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700793 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 kmem_cache_free(vm_area_cachep, next);
795 /*
796 * In mprotect's case 6 (see comments on vma_merge),
797 * we must remove another next too. It would clutter
798 * up the code too much to do both in one go.
799 */
Michel Lespinassed3737182012-12-11 16:01:38 -0800800 next = vma->vm_next;
801 if (remove_next == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 goto again;
Michel Lespinassed3737182012-12-11 16:01:38 -0800803 else if (next)
804 vma_gap_update(next);
805 else
806 mm->highest_vm_end = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530808 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100809 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
811 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800812
813 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
816/*
817 * If the vma has a ->close operation then the driver probably needs to release
818 * per-vma resources, so we don't attempt to merge those.
819 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700821 struct file *file, unsigned long vm_flags,
822 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800824 /*
825 * VM_SOFTDIRTY should not prevent from VMA merging, if we
826 * match the flags but dirty bit -- the caller should mark
827 * merged VMA as dirty. If dirty bit won't be excluded from
828 * comparison, we increase pressue on the memory system forcing
829 * the kernel to generate new VMAs when old one could be
830 * extended instead.
831 */
832 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return 0;
834 if (vma->vm_file != file)
835 return 0;
836 if (vma->vm_ops && vma->vm_ops->close)
837 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700838 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
839 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 return 1;
841}
842
843static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700844 struct anon_vma *anon_vma2,
845 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846{
Shaohua Li965f55d2011-05-24 17:11:20 -0700847 /*
848 * The list_is_singular() test is to avoid merging VMA cloned from
849 * parents. This can improve scalability caused by anon_vma lock.
850 */
851 if ((!anon_vma1 || !anon_vma2) && (!vma ||
852 list_is_singular(&vma->anon_vma_chain)))
853 return 1;
854 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855}
856
857/*
858 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
859 * in front of (at a lower virtual address and file offset than) the vma.
860 *
861 * We cannot merge two vmas if they have differently assigned (non-NULL)
862 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
863 *
864 * We don't check here for the merged mmap wrapping around the end of pagecache
865 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
866 * wrap, nor mmaps which cover the final page at index -1UL.
867 */
868static int
869can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700870 struct anon_vma *anon_vma, struct file *file,
871 pgoff_t vm_pgoff,
872 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700874 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700875 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 if (vma->vm_pgoff == vm_pgoff)
877 return 1;
878 }
879 return 0;
880}
881
882/*
883 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
884 * beyond (at a higher virtual address and file offset than) the vma.
885 *
886 * We cannot merge two vmas if they have differently assigned (non-NULL)
887 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
888 */
889static int
890can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700891 struct anon_vma *anon_vma, struct file *file,
892 pgoff_t vm_pgoff,
893 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700895 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -0700896 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -0700898 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
900 return 1;
901 }
902 return 0;
903}
904
905/*
906 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
907 * whether that can be merged with its predecessor or its successor.
908 * Or both (it neatly fills a hole).
909 *
910 * In most cases - when called for mmap, brk or mremap - [addr,end) is
911 * certain not to be mapped by the time vma_merge is called; but when
912 * called for mprotect, it is certain to be already mapped (either at
913 * an offset within prev, or at the start of next), and the flags of
914 * this area are about to be changed to vm_flags - and the no-change
915 * case has already been eliminated.
916 *
917 * The following mprotect cases have to be considered, where AAAA is
918 * the area passed down from mprotect_fixup, never extending beyond one
919 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
920 *
921 * AAAA AAAA AAAA AAAA
922 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
923 * cannot merge might become might become might become
924 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
925 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
926 * mremap move: PPPPNNNNNNNN 8
927 * AAAA
928 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
929 * might become case 1 below case 2 below case 3 below
930 *
931 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
932 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
933 */
934struct vm_area_struct *vma_merge(struct mm_struct *mm,
935 struct vm_area_struct *prev, unsigned long addr,
936 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -0700937 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700938 pgoff_t pgoff, struct mempolicy *policy,
939 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
941 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
942 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800943 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
945 /*
946 * We later require that vma->vm_flags == vm_flags,
947 * so this tests vma->vm_flags & VM_SPECIAL, too.
948 */
949 if (vm_flags & VM_SPECIAL)
950 return NULL;
951
952 if (prev)
953 next = prev->vm_next;
954 else
955 next = mm->mmap;
956 area = next;
957 if (next && next->vm_end == end) /* cases 6, 7, 8 */
958 next = next->vm_next;
959
960 /*
961 * Can it merge with the predecessor?
962 */
963 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -0700964 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700966 anon_vma, file, pgoff,
967 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 /*
969 * OK, it can. Can we now merge in the successor as well?
970 */
971 if (next && end == next->vm_start &&
972 mpol_equal(policy, vma_policy(next)) &&
973 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700974 anon_vma, file,
975 pgoff+pglen,
976 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -0700978 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 /* cases 1, 6 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800980 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 next->vm_end, prev->vm_pgoff, NULL);
982 } else /* cases 2, 5, 7 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800983 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 end, prev->vm_pgoff, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800985 if (err)
986 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -0700987 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 return prev;
989 }
990
991 /*
992 * Can this new request be merged in front of next?
993 */
994 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -0700995 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700997 anon_vma, file, pgoff+pglen,
998 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 if (prev && addr < prev->vm_end) /* case 4 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001000 err = vma_adjust(prev, prev->vm_start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 addr, prev->vm_pgoff, NULL);
1002 else /* cases 3, 8 */
Rik van Riel5beb4932010-03-05 13:42:07 -08001003 err = vma_adjust(area, addr, next->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 next->vm_pgoff - pglen, NULL);
Rik van Riel5beb4932010-03-05 13:42:07 -08001005 if (err)
1006 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001007 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 return area;
1009 }
1010
1011 return NULL;
1012}
1013
1014/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001015 * Rough compatbility check to quickly see if it's even worth looking
1016 * at sharing an anon_vma.
1017 *
1018 * They need to have the same vm_file, and the flags can only differ
1019 * in things that mprotect may change.
1020 *
1021 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1022 * we can merge the two vma's. For example, we refuse to merge a vma if
1023 * there is a vm_ops->close() function, because that indicates that the
1024 * driver is doing some kind of reference counting. But that doesn't
1025 * really matter for the anon_vma sharing case.
1026 */
1027static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1028{
1029 return a->vm_end == b->vm_start &&
1030 mpol_equal(vma_policy(a), vma_policy(b)) &&
1031 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001032 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001033 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1034}
1035
1036/*
1037 * Do some basic sanity checking to see if we can re-use the anon_vma
1038 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1039 * the same as 'old', the other will be the new one that is trying
1040 * to share the anon_vma.
1041 *
1042 * NOTE! This runs with mm_sem held for reading, so it is possible that
1043 * the anon_vma of 'old' is concurrently in the process of being set up
1044 * by another page fault trying to merge _that_. But that's ok: if it
1045 * is being set up, that automatically means that it will be a singleton
1046 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001047 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001048 *
1049 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1050 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1051 * is to return an anon_vma that is "complex" due to having gone through
1052 * a fork).
1053 *
1054 * We also make sure that the two vma's are compatible (adjacent,
1055 * and with the same memory policies). That's all stable, even with just
1056 * a read lock on the mm_sem.
1057 */
1058static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1059{
1060 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001061 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001062
1063 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1064 return anon_vma;
1065 }
1066 return NULL;
1067}
1068
1069/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1071 * neighbouring vmas for a suitable anon_vma, before it goes off
1072 * to allocate a new anon_vma. It checks because a repetitive
1073 * sequence of mprotects and faults may otherwise lead to distinct
1074 * anon_vmas being allocated, preventing vma merge in subsequent
1075 * mprotect.
1076 */
1077struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1078{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001079 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
1082 near = vma->vm_next;
1083 if (!near)
1084 goto try_prev;
1085
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001086 anon_vma = reusable_anon_vma(near, vma, near);
1087 if (anon_vma)
1088 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001090 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (!near)
1092 goto none;
1093
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001094 anon_vma = reusable_anon_vma(near, near, vma);
1095 if (anon_vma)
1096 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097none:
1098 /*
1099 * There's no absolute need to look only at touching neighbours:
1100 * we could search further afield for "compatible" anon_vmas.
1101 * But it would probably just be a waste of time searching,
1102 * or lead to too many vmas hanging off the same anon_vma.
1103 * We're trying to allow mprotect remerging later on,
1104 * not trying to minimize memory used for anon_vmas.
1105 */
1106 return NULL;
1107}
1108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109/*
Al Viro40401532012-02-13 03:58:52 +00001110 * If a hint addr is less than mmap_min_addr change hint to be as
1111 * low as possible but still greater than mmap_min_addr
1112 */
1113static inline unsigned long round_hint_to_min(unsigned long hint)
1114{
1115 hint &= PAGE_MASK;
1116 if (((void *)hint != NULL) &&
1117 (hint < mmap_min_addr))
1118 return PAGE_ALIGN(mmap_min_addr);
1119 return hint;
1120}
1121
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001122static inline int mlock_future_check(struct mm_struct *mm,
1123 unsigned long flags,
1124 unsigned long len)
1125{
1126 unsigned long locked, lock_limit;
1127
1128 /* mlock MCL_FUTURE? */
1129 if (flags & VM_LOCKED) {
1130 locked = len >> PAGE_SHIFT;
1131 locked += mm->locked_vm;
1132 lock_limit = rlimit(RLIMIT_MEMLOCK);
1133 lock_limit >>= PAGE_SHIFT;
1134 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1135 return -EAGAIN;
1136 }
1137 return 0;
1138}
1139
Al Viro40401532012-02-13 03:58:52 +00001140/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001141 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001143unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001145 unsigned long flags, vm_flags_t vm_flags,
1146 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147{
vishnu.pscc71aba2014-10-09 15:26:29 -07001148 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001149 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
Michel Lespinasse41badc12013-02-22 16:32:47 -08001151 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001152
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001153 if (!len)
1154 return -EINVAL;
1155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 /*
1157 * Does the application expect PROT_READ to imply PROT_EXEC?
1158 *
1159 * (the exception is when the underlying filesystem is noexec
1160 * mounted, in which case we dont add PROT_EXEC.)
1161 */
1162 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001163 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 prot |= PROT_EXEC;
1165
Eric Paris7cd94142007-11-26 18:47:40 -05001166 if (!(flags & MAP_FIXED))
1167 addr = round_hint_to_min(addr);
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 /* Careful about overflows.. */
1170 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001171 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 return -ENOMEM;
1173
1174 /* offset overflow? */
1175 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001176 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
1178 /* Too many mappings? */
1179 if (mm->map_count > sysctl_max_map_count)
1180 return -ENOMEM;
1181
1182 /* Obtain the address to map to. we verify (or select) it and ensure
1183 * that it represents a valid section of the address space.
1184 */
1185 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001186 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 return addr;
1188
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001189 if (prot == PROT_EXEC) {
1190 pkey = execute_only_pkey(mm);
1191 if (pkey < 0)
1192 pkey = 0;
1193 }
1194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 /* Do simple checking here so the lower-level routines won't have
1196 * to. we assume access permissions have been handled by the open
1197 * of the memory object, so we don't do any here.
1198 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001199 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1201
Huang Shijiecdf7b342009-09-21 17:03:36 -07001202 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 if (!can_do_mlock())
1204 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001205
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001206 if (mlock_future_check(mm, vm_flags, len))
1207 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001210 struct inode *inode = file_inode(file);
1211
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 switch (flags & MAP_TYPE) {
1213 case MAP_SHARED:
1214 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1215 return -EACCES;
1216
1217 /*
1218 * Make sure we don't allow writing to an append-only
1219 * file..
1220 */
1221 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1222 return -EACCES;
1223
1224 /*
1225 * Make sure there are no mandatory locks on the file.
1226 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001227 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 return -EAGAIN;
1229
1230 vm_flags |= VM_SHARED | VM_MAYSHARE;
1231 if (!(file->f_mode & FMODE_WRITE))
1232 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1233
1234 /* fall through */
1235 case MAP_PRIVATE:
1236 if (!(file->f_mode & FMODE_READ))
1237 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001238 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001239 if (vm_flags & VM_EXEC)
1240 return -EPERM;
1241 vm_flags &= ~VM_MAYEXEC;
1242 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001243
Al Viro72c2d532013-09-22 16:27:52 -04001244 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001245 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001246 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1247 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 break;
1249
1250 default:
1251 return -EINVAL;
1252 }
1253 } else {
1254 switch (flags & MAP_TYPE) {
1255 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001256 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1257 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001258 /*
1259 * Ignore pgoff.
1260 */
1261 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 vm_flags |= VM_SHARED | VM_MAYSHARE;
1263 break;
1264 case MAP_PRIVATE:
1265 /*
1266 * Set pgoff according to addr for anon_vma.
1267 */
1268 pgoff = addr >> PAGE_SHIFT;
1269 break;
1270 default:
1271 return -EINVAL;
1272 }
1273 }
1274
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001275 /*
1276 * Set 'VM_NORESERVE' if we should not account for the
1277 * memory use of this mapping.
1278 */
1279 if (flags & MAP_NORESERVE) {
1280 /* We honor MAP_NORESERVE if allowed to overcommit */
1281 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1282 vm_flags |= VM_NORESERVE;
1283
1284 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1285 if (file && is_file_hugepages(file))
1286 vm_flags |= VM_NORESERVE;
1287 }
1288
1289 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001290 if (!IS_ERR_VALUE(addr) &&
1291 ((vm_flags & VM_LOCKED) ||
1292 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001293 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001294 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001295}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001296
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001297SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1298 unsigned long, prot, unsigned long, flags,
1299 unsigned long, fd, unsigned long, pgoff)
1300{
1301 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001302 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001303
1304 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001305 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001306 file = fget(fd);
1307 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001308 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001309 if (is_file_hugepages(file))
1310 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001311 retval = -EINVAL;
1312 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1313 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001314 } else if (flags & MAP_HUGETLB) {
1315 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001316 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001317
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001318 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001319 if (!hs)
1320 return -EINVAL;
1321
1322 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001323 /*
1324 * VM_NORESERVE is used because the reservations will be
1325 * taken when vm_ops->mmap() is called
1326 * A dummy user value is used because we are not locking
1327 * memory so no accounting is necessary
1328 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001329 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001330 VM_NORESERVE,
1331 &user, HUGETLB_ANONHUGE_INODE,
1332 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001333 if (IS_ERR(file))
1334 return PTR_ERR(file);
1335 }
1336
1337 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1338
Al Viroeb36c582012-05-30 20:17:35 -04001339 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001340out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001341 if (file)
1342 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001343 return retval;
1344}
1345
Christoph Hellwiga4679372010-03-10 15:21:15 -08001346#ifdef __ARCH_WANT_SYS_OLD_MMAP
1347struct mmap_arg_struct {
1348 unsigned long addr;
1349 unsigned long len;
1350 unsigned long prot;
1351 unsigned long flags;
1352 unsigned long fd;
1353 unsigned long offset;
1354};
1355
1356SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1357{
1358 struct mmap_arg_struct a;
1359
1360 if (copy_from_user(&a, arg, sizeof(a)))
1361 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001362 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001363 return -EINVAL;
1364
1365 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1366 a.offset >> PAGE_SHIFT);
1367}
1368#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1369
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001370/*
1371 * Some shared mappigns will want the pages marked read-only
1372 * to track write events. If so, we'll downgrade vm_page_prot
1373 * to the private version (using protection_map[] without the
1374 * VM_SHARED bit).
1375 */
1376int vma_wants_writenotify(struct vm_area_struct *vma)
1377{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001378 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001379 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001380
1381 /* If it was private or non-writable, the write bit is already clear */
1382 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1383 return 0;
1384
1385 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001386 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001387 return 1;
1388
Peter Feiner64e45502014-10-13 15:55:46 -07001389 /* The open routine did something to the protections that pgprot_modify
1390 * won't preserve? */
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001391 if (pgprot_val(vma->vm_page_prot) !=
Peter Feiner64e45502014-10-13 15:55:46 -07001392 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001393 return 0;
1394
Peter Feiner64e45502014-10-13 15:55:46 -07001395 /* Do we need to track softdirty? */
1396 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1397 return 1;
1398
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001399 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001400 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001401 return 0;
1402
1403 /* Can the mapping track the dirty pages? */
1404 return vma->vm_file && vma->vm_file->f_mapping &&
1405 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1406}
1407
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001408/*
1409 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001410 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001411 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001412static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001413{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001414 /*
1415 * hugetlb has its own accounting separate from the core VM
1416 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1417 */
1418 if (file && is_file_hugepages(file))
1419 return 0;
1420
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001421 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1422}
1423
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001424unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001425 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001426{
1427 struct mm_struct *mm = current->mm;
1428 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001429 int error;
1430 struct rb_node **rb_link, *rb_parent;
1431 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001432
Cyril Hrubise8420a82013-04-29 15:08:33 -07001433 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001434 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001435 unsigned long nr_pages;
1436
1437 /*
1438 * MAP_FIXED may remove pages of mappings that intersects with
1439 * requested mapping. Account for the pages it would unmap.
1440 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001441 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1442
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001443 if (!may_expand_vm(mm, vm_flags,
1444 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001445 return -ENOMEM;
1446 }
1447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001449 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1450 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 if (do_munmap(mm, addr, len))
1452 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 }
1454
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001455 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001456 * Private writable mapping: check memory availability
1457 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001458 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001459 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001460 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001461 return -ENOMEM;
1462 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 }
1464
1465 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001466 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001468 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1469 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001470 if (vma)
1471 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
1473 /*
1474 * Determine the object being mapped and call the appropriate
1475 * specific mapper. the address has already been validated, but
1476 * not unmapped, but the maps are removed from the list.
1477 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001478 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 if (!vma) {
1480 error = -ENOMEM;
1481 goto unacct_error;
1482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483
1484 vma->vm_mm = mm;
1485 vma->vm_start = addr;
1486 vma->vm_end = addr + len;
1487 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001488 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001490 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
1492 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (vm_flags & VM_DENYWRITE) {
1494 error = deny_write_access(file);
1495 if (error)
1496 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001498 if (vm_flags & VM_SHARED) {
1499 error = mapping_map_writable(file->f_mapping);
1500 if (error)
1501 goto allow_write_and_free_vma;
1502 }
1503
1504 /* ->mmap() can change vma->vm_file, but must guarantee that
1505 * vma_link() below can deny write-access if VM_DENYWRITE is set
1506 * and map writably if VM_SHARED is set. This usually means the
1507 * new file must not have been exposed to user-space, yet.
1508 */
Al Virocb0942b2012-08-27 14:48:26 -04001509 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 error = file->f_op->mmap(file, vma);
1511 if (error)
1512 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001513
1514 /* Can addr have changed??
1515 *
1516 * Answer: Yes, several device drivers can do it in their
1517 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001518 * Bug: If addr is changed, prev, rb_link, rb_parent should
1519 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001520 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001521 WARN_ON_ONCE(addr != vma->vm_start);
1522
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001523 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001524 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 } else if (vm_flags & VM_SHARED) {
1526 error = shmem_zero_setup(vma);
1527 if (error)
1528 goto free_vma;
1529 }
1530
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001531 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001532 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001533 if (file) {
1534 if (vm_flags & VM_SHARED)
1535 mapping_unmap_writable(file->f_mapping);
1536 if (vm_flags & VM_DENYWRITE)
1537 allow_write_access(file);
1538 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001539 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001540out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001541 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001542
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001543 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001545 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1546 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001547 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001548 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001549 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001550 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301551
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001552 if (file)
1553 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301554
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001555 /*
1556 * New (or expanded) vma always get soft dirty status.
1557 * Otherwise user-space soft-dirty page tracker won't
1558 * be able to distinguish situation when vma area unmapped,
1559 * then new mapped in-place (which must be aimed as
1560 * a completely new data area).
1561 */
1562 vma->vm_flags |= VM_SOFTDIRTY;
1563
Peter Feiner64e45502014-10-13 15:55:46 -07001564 vma_set_page_prot(vma);
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 return addr;
1567
1568unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 vma->vm_file = NULL;
1570 fput(file);
1571
1572 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001573 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1574 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001575 if (vm_flags & VM_SHARED)
1576 mapping_unmap_writable(file->f_mapping);
1577allow_write_and_free_vma:
1578 if (vm_flags & VM_DENYWRITE)
1579 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580free_vma:
1581 kmem_cache_free(vm_area_cachep, vma);
1582unacct_error:
1583 if (charged)
1584 vm_unacct_memory(charged);
1585 return error;
1586}
1587
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001588unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1589{
1590 /*
1591 * We implement the search by looking for an rbtree node that
1592 * immediately follows a suitable gap. That is,
1593 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1594 * - gap_end = vma->vm_start >= info->low_limit + length;
1595 * - gap_end - gap_start >= length
1596 */
1597
1598 struct mm_struct *mm = current->mm;
1599 struct vm_area_struct *vma;
1600 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1601
1602 /* Adjust search length to account for worst case alignment overhead */
1603 length = info->length + info->align_mask;
1604 if (length < info->length)
1605 return -ENOMEM;
1606
1607 /* Adjust search limits by the desired length */
1608 if (info->high_limit < length)
1609 return -ENOMEM;
1610 high_limit = info->high_limit - length;
1611
1612 if (info->low_limit > high_limit)
1613 return -ENOMEM;
1614 low_limit = info->low_limit + length;
1615
1616 /* Check if rbtree root looks promising */
1617 if (RB_EMPTY_ROOT(&mm->mm_rb))
1618 goto check_highest;
1619 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1620 if (vma->rb_subtree_gap < length)
1621 goto check_highest;
1622
1623 while (true) {
1624 /* Visit left subtree if it looks promising */
1625 gap_end = vma->vm_start;
1626 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1627 struct vm_area_struct *left =
1628 rb_entry(vma->vm_rb.rb_left,
1629 struct vm_area_struct, vm_rb);
1630 if (left->rb_subtree_gap >= length) {
1631 vma = left;
1632 continue;
1633 }
1634 }
1635
1636 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1637check_current:
1638 /* Check if current node has a suitable gap */
1639 if (gap_start > high_limit)
1640 return -ENOMEM;
1641 if (gap_end >= low_limit && gap_end - gap_start >= length)
1642 goto found;
1643
1644 /* Visit right subtree if it looks promising */
1645 if (vma->vm_rb.rb_right) {
1646 struct vm_area_struct *right =
1647 rb_entry(vma->vm_rb.rb_right,
1648 struct vm_area_struct, vm_rb);
1649 if (right->rb_subtree_gap >= length) {
1650 vma = right;
1651 continue;
1652 }
1653 }
1654
1655 /* Go back up the rbtree to find next candidate node */
1656 while (true) {
1657 struct rb_node *prev = &vma->vm_rb;
1658 if (!rb_parent(prev))
1659 goto check_highest;
1660 vma = rb_entry(rb_parent(prev),
1661 struct vm_area_struct, vm_rb);
1662 if (prev == vma->vm_rb.rb_left) {
1663 gap_start = vma->vm_prev->vm_end;
1664 gap_end = vma->vm_start;
1665 goto check_current;
1666 }
1667 }
1668 }
1669
1670check_highest:
1671 /* Check highest gap, which does not precede any rbtree node */
1672 gap_start = mm->highest_vm_end;
1673 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1674 if (gap_start > high_limit)
1675 return -ENOMEM;
1676
1677found:
1678 /* We found a suitable gap. Clip it with the original low_limit. */
1679 if (gap_start < info->low_limit)
1680 gap_start = info->low_limit;
1681
1682 /* Adjust gap address to the desired alignment */
1683 gap_start += (info->align_offset - gap_start) & info->align_mask;
1684
1685 VM_BUG_ON(gap_start + info->length > info->high_limit);
1686 VM_BUG_ON(gap_start + info->length > gap_end);
1687 return gap_start;
1688}
1689
1690unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1691{
1692 struct mm_struct *mm = current->mm;
1693 struct vm_area_struct *vma;
1694 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1695
1696 /* Adjust search length to account for worst case alignment overhead */
1697 length = info->length + info->align_mask;
1698 if (length < info->length)
1699 return -ENOMEM;
1700
1701 /*
1702 * Adjust search limits by the desired length.
1703 * See implementation comment at top of unmapped_area().
1704 */
1705 gap_end = info->high_limit;
1706 if (gap_end < length)
1707 return -ENOMEM;
1708 high_limit = gap_end - length;
1709
1710 if (info->low_limit > high_limit)
1711 return -ENOMEM;
1712 low_limit = info->low_limit + length;
1713
1714 /* Check highest gap, which does not precede any rbtree node */
1715 gap_start = mm->highest_vm_end;
1716 if (gap_start <= high_limit)
1717 goto found_highest;
1718
1719 /* Check if rbtree root looks promising */
1720 if (RB_EMPTY_ROOT(&mm->mm_rb))
1721 return -ENOMEM;
1722 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1723 if (vma->rb_subtree_gap < length)
1724 return -ENOMEM;
1725
1726 while (true) {
1727 /* Visit right subtree if it looks promising */
1728 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1729 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1730 struct vm_area_struct *right =
1731 rb_entry(vma->vm_rb.rb_right,
1732 struct vm_area_struct, vm_rb);
1733 if (right->rb_subtree_gap >= length) {
1734 vma = right;
1735 continue;
1736 }
1737 }
1738
1739check_current:
1740 /* Check if current node has a suitable gap */
1741 gap_end = vma->vm_start;
1742 if (gap_end < low_limit)
1743 return -ENOMEM;
1744 if (gap_start <= high_limit && gap_end - gap_start >= length)
1745 goto found;
1746
1747 /* Visit left subtree if it looks promising */
1748 if (vma->vm_rb.rb_left) {
1749 struct vm_area_struct *left =
1750 rb_entry(vma->vm_rb.rb_left,
1751 struct vm_area_struct, vm_rb);
1752 if (left->rb_subtree_gap >= length) {
1753 vma = left;
1754 continue;
1755 }
1756 }
1757
1758 /* Go back up the rbtree to find next candidate node */
1759 while (true) {
1760 struct rb_node *prev = &vma->vm_rb;
1761 if (!rb_parent(prev))
1762 return -ENOMEM;
1763 vma = rb_entry(rb_parent(prev),
1764 struct vm_area_struct, vm_rb);
1765 if (prev == vma->vm_rb.rb_right) {
1766 gap_start = vma->vm_prev ?
1767 vma->vm_prev->vm_end : 0;
1768 goto check_current;
1769 }
1770 }
1771 }
1772
1773found:
1774 /* We found a suitable gap. Clip it with the original high_limit. */
1775 if (gap_end > info->high_limit)
1776 gap_end = info->high_limit;
1777
1778found_highest:
1779 /* Compute highest gap address at the desired alignment */
1780 gap_end -= info->length;
1781 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1782
1783 VM_BUG_ON(gap_end < info->low_limit);
1784 VM_BUG_ON(gap_end < gap_start);
1785 return gap_end;
1786}
1787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788/* Get an address range which is currently unmapped.
1789 * For shmat() with addr=0.
1790 *
1791 * Ugly calling convention alert:
1792 * Return value with the low bits set means error value,
1793 * ie
1794 * if (ret & ~PAGE_MASK)
1795 * error = ret;
1796 *
1797 * This function "knows" that -ENOMEM has the bits set.
1798 */
1799#ifndef HAVE_ARCH_UNMAPPED_AREA
1800unsigned long
1801arch_get_unmapped_area(struct file *filp, unsigned long addr,
1802 unsigned long len, unsigned long pgoff, unsigned long flags)
1803{
1804 struct mm_struct *mm = current->mm;
1805 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001806 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001808 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 return -ENOMEM;
1810
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001811 if (flags & MAP_FIXED)
1812 return addr;
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 if (addr) {
1815 addr = PAGE_ALIGN(addr);
1816 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001817 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 (!vma || addr + len <= vma->vm_start))
1819 return addr;
1820 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001822 info.flags = 0;
1823 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001824 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001825 info.high_limit = TASK_SIZE;
1826 info.align_mask = 0;
1827 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828}
vishnu.pscc71aba2014-10-09 15:26:29 -07001829#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831/*
1832 * This mmap-allocator allocates new areas top-down from below the
1833 * stack's low limit (the base):
1834 */
1835#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1836unsigned long
1837arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1838 const unsigned long len, const unsigned long pgoff,
1839 const unsigned long flags)
1840{
1841 struct vm_area_struct *vma;
1842 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001843 unsigned long addr = addr0;
1844 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
1846 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001847 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 return -ENOMEM;
1849
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001850 if (flags & MAP_FIXED)
1851 return addr;
1852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 /* requesting a specific address */
1854 if (addr) {
1855 addr = PAGE_ALIGN(addr);
1856 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001857 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 (!vma || addr + len <= vma->vm_start))
1859 return addr;
1860 }
1861
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001862 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1863 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001864 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001865 info.high_limit = mm->mmap_base;
1866 info.align_mask = 0;
1867 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07001868
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 /*
1870 * A failed mmap() very likely causes application failure,
1871 * so fall back to the bottom-up function here. This scenario
1872 * can happen with large stack limits and large mmap()
1873 * allocations.
1874 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001875 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001876 VM_BUG_ON(addr != -ENOMEM);
1877 info.flags = 0;
1878 info.low_limit = TASK_UNMAPPED_BASE;
1879 info.high_limit = TASK_SIZE;
1880 addr = vm_unmapped_area(&info);
1881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882
1883 return addr;
1884}
1885#endif
1886
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887unsigned long
1888get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1889 unsigned long pgoff, unsigned long flags)
1890{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001891 unsigned long (*get_area)(struct file *, unsigned long,
1892 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Al Viro9206de92009-12-03 15:23:11 -05001894 unsigned long error = arch_mmap_check(addr, len, flags);
1895 if (error)
1896 return error;
1897
1898 /* Careful about overflows.. */
1899 if (len > TASK_SIZE)
1900 return -ENOMEM;
1901
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001902 get_area = current->mm->get_unmapped_area;
Al Viro72c2d532013-09-22 16:27:52 -04001903 if (file && file->f_op->get_unmapped_area)
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001904 get_area = file->f_op->get_unmapped_area;
1905 addr = get_area(file, addr, len, pgoff, flags);
1906 if (IS_ERR_VALUE(addr))
1907 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001909 if (addr > TASK_SIZE - len)
1910 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001911 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07001912 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001913
Al Viro9ac4ed42012-05-30 17:13:15 -04001914 addr = arch_rebalance_pgtables(addr, len);
1915 error = security_mmap_addr(addr);
1916 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917}
1918
1919EXPORT_SYMBOL(get_unmapped_area);
1920
1921/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08001922struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001924 struct rb_node *rb_node;
1925 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001927 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001928 vma = vmacache_find(mm, addr);
1929 if (likely(vma))
1930 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001932 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001934 while (rb_node) {
1935 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001937 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07001938
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001939 if (tmp->vm_end > addr) {
1940 vma = tmp;
1941 if (tmp->vm_start <= addr)
1942 break;
1943 rb_node = rb_node->rb_left;
1944 } else
1945 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001947
1948 if (vma)
1949 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 return vma;
1951}
1952
1953EXPORT_SYMBOL(find_vma);
1954
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001955/*
1956 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001957 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958struct vm_area_struct *
1959find_vma_prev(struct mm_struct *mm, unsigned long addr,
1960 struct vm_area_struct **pprev)
1961{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001962 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001964 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05001965 if (vma) {
1966 *pprev = vma->vm_prev;
1967 } else {
1968 struct rb_node *rb_node = mm->mm_rb.rb_node;
1969 *pprev = NULL;
1970 while (rb_node) {
1971 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1972 rb_node = rb_node->rb_right;
1973 }
1974 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08001975 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976}
1977
1978/*
1979 * Verify that the stack growth is acceptable and
1980 * update accounting. This is shared with both the
1981 * grow-up and grow-down cases.
1982 */
ZhenwenXu48aae422009-01-06 14:40:21 -08001983static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984{
1985 struct mm_struct *mm = vma->vm_mm;
1986 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08001987 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
1989 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001990 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 return -ENOMEM;
1992
1993 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08001994 actual_size = size;
1995 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
1996 actual_size -= PAGE_SIZE;
Jason Low4db0c3c2015-04-15 16:14:08 -07001997 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 return -ENOMEM;
1999
2000 /* mlock limit tests */
2001 if (vma->vm_flags & VM_LOCKED) {
2002 unsigned long locked;
2003 unsigned long limit;
2004 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002005 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002006 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 if (locked > limit && !capable(CAP_IPC_LOCK))
2008 return -ENOMEM;
2009 }
2010
Adam Litke0d59a012007-01-30 14:35:39 -08002011 /* Check to ensure the stack will not grow into a hugetlb-only region */
2012 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2013 vma->vm_end - size;
2014 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2015 return -EFAULT;
2016
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 /*
2018 * Overcommit.. This must be the final test, as it will
2019 * update security statistics.
2020 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002021 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 return -ENOMEM;
2023
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 return 0;
2025}
2026
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002027#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002029 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2030 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002032int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033{
Oleg Nesterov09357812015-11-05 18:48:17 -08002034 struct mm_struct *mm = vma->vm_mm;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002035 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
2037 if (!(vma->vm_flags & VM_GROWSUP))
2038 return -EFAULT;
2039
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002040 /* Guard against wrapping around to address 0. */
2041 if (address < PAGE_ALIGN(address+4))
2042 address = PAGE_ALIGN(address+4);
2043 else
2044 return -ENOMEM;
2045
2046 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 if (unlikely(anon_vma_prepare(vma)))
2048 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
2050 /*
2051 * vma->vm_start/vm_end cannot change under us because the caller
2052 * is required to hold the mmap_sem in read mode. We need the
2053 * anon_vma lock to serialize against concurrent expand_stacks.
2054 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002055 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
2057 /* Somebody else might have raced and expanded it already */
2058 if (address > vma->vm_end) {
2059 unsigned long size, grow;
2060
2061 size = address - vma->vm_start;
2062 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2063
Hugh Dickins42c36f62011-05-09 17:44:42 -07002064 error = -ENOMEM;
2065 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2066 error = acct_stack_growth(vma, size, grow);
2067 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002068 /*
2069 * vma_gap_update() doesn't support concurrent
2070 * updates, but we only hold a shared mmap_sem
2071 * lock here, so we need to protect against
2072 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002073 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002074 * we don't guarantee that all growable vmas
2075 * in a mm share the same root anon vma.
2076 * So, we reuse mm->page_table_lock to guard
2077 * against concurrent vma expansions.
2078 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002079 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002080 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002081 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002082 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002083 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002084 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002085 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002086 if (vma->vm_next)
2087 vma_gap_update(vma->vm_next);
2088 else
Oleg Nesterov09357812015-11-05 18:48:17 -08002089 mm->highest_vm_end = address;
2090 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002091
Hugh Dickins42c36f62011-05-09 17:44:42 -07002092 perf_event_mmap(vma);
2093 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002094 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002096 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002097 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002098 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 return error;
2100}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002101#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2102
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103/*
2104 * vma is the first one with address < vma->vm_start. Have to extend vma.
2105 */
Michal Hockod05f3162011-05-24 17:11:44 -07002106int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002107 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108{
Oleg Nesterov09357812015-11-05 18:48:17 -08002109 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 int error;
2111
Eric Paris88694772007-11-26 18:47:26 -05002112 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002113 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002114 if (error)
2115 return error;
2116
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002117 /* We must make sure the anon_vma is allocated. */
2118 if (unlikely(anon_vma_prepare(vma)))
2119 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120
2121 /*
2122 * vma->vm_start/vm_end cannot change under us because the caller
2123 * is required to hold the mmap_sem in read mode. We need the
2124 * anon_vma lock to serialize against concurrent expand_stacks.
2125 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002126 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127
2128 /* Somebody else might have raced and expanded it already */
2129 if (address < vma->vm_start) {
2130 unsigned long size, grow;
2131
2132 size = vma->vm_end - address;
2133 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2134
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002135 error = -ENOMEM;
2136 if (grow <= vma->vm_pgoff) {
2137 error = acct_stack_growth(vma, size, grow);
2138 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002139 /*
2140 * vma_gap_update() doesn't support concurrent
2141 * updates, but we only hold a shared mmap_sem
2142 * lock here, so we need to protect against
2143 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002144 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002145 * we don't guarantee that all growable vmas
2146 * in a mm share the same root anon vma.
2147 * So, we reuse mm->page_table_lock to guard
2148 * against concurrent vma expansions.
2149 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002150 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002151 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002152 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002153 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002154 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002155 vma->vm_start = address;
2156 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002157 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002158 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002159 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002160
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002161 perf_event_mmap(vma);
2162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 }
2164 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002165 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002166 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002167 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 return error;
2169}
2170
Linus Torvalds09884962013-02-27 08:36:04 -08002171/*
2172 * Note how expand_stack() refuses to expand the stack all the way to
2173 * abut the next virtual mapping, *unless* that mapping itself is also
2174 * a stack mapping. We want to leave room for a guard page, after all
2175 * (the guard page itself is not added here, that is done by the
2176 * actual page faulting logic)
2177 *
2178 * This matches the behavior of the guard page logic (see mm/memory.c:
2179 * check_stack_guard_page()), which only allows the guard page to be
2180 * removed under these circumstances.
2181 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002182#ifdef CONFIG_STACK_GROWSUP
2183int expand_stack(struct vm_area_struct *vma, unsigned long address)
2184{
Linus Torvalds09884962013-02-27 08:36:04 -08002185 struct vm_area_struct *next;
2186
2187 address &= PAGE_MASK;
2188 next = vma->vm_next;
2189 if (next && next->vm_start == address + PAGE_SIZE) {
2190 if (!(next->vm_flags & VM_GROWSUP))
2191 return -ENOMEM;
2192 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002193 return expand_upwards(vma, address);
2194}
2195
2196struct vm_area_struct *
2197find_extend_vma(struct mm_struct *mm, unsigned long addr)
2198{
2199 struct vm_area_struct *vma, *prev;
2200
2201 addr &= PAGE_MASK;
2202 vma = find_vma_prev(mm, addr, &prev);
2203 if (vma && (vma->vm_start <= addr))
2204 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002205 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002206 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002207 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002208 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002209 return prev;
2210}
2211#else
2212int expand_stack(struct vm_area_struct *vma, unsigned long address)
2213{
Linus Torvalds09884962013-02-27 08:36:04 -08002214 struct vm_area_struct *prev;
2215
2216 address &= PAGE_MASK;
2217 prev = vma->vm_prev;
2218 if (prev && prev->vm_end == address) {
2219 if (!(prev->vm_flags & VM_GROWSDOWN))
2220 return -ENOMEM;
2221 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002222 return expand_downwards(vma, address);
2223}
2224
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002226find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227{
vishnu.pscc71aba2014-10-09 15:26:29 -07002228 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 unsigned long start;
2230
2231 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002232 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 if (!vma)
2234 return NULL;
2235 if (vma->vm_start <= addr)
2236 return vma;
2237 if (!(vma->vm_flags & VM_GROWSDOWN))
2238 return NULL;
2239 start = vma->vm_start;
2240 if (expand_stack(vma, addr))
2241 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002242 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002243 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 return vma;
2245}
2246#endif
2247
Jesse Barnese1d6d012014-12-12 16:55:27 -08002248EXPORT_SYMBOL_GPL(find_extend_vma);
2249
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002251 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002253 *
2254 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002256static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002258 unsigned long nr_accounted = 0;
2259
Hugh Dickins365e9c872005-10-29 18:16:18 -07002260 /* Update high watermark before we lower total_vm */
2261 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002263 long nrpages = vma_pages(vma);
2264
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002265 if (vma->vm_flags & VM_ACCOUNT)
2266 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002267 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002268 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002269 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002270 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 validate_mm(mm);
2272}
2273
2274/*
2275 * Get rid of page table information in the indicated region.
2276 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002277 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 */
2279static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002280 struct vm_area_struct *vma, struct vm_area_struct *prev,
2281 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282{
vishnu.pscc71aba2014-10-09 15:26:29 -07002283 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002284 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
2286 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002287 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002288 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002289 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002290 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002291 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002292 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293}
2294
2295/*
2296 * Create a list of vma's touched by the unmap, removing them from the mm's
2297 * vma list as we go..
2298 */
2299static void
2300detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2301 struct vm_area_struct *prev, unsigned long end)
2302{
2303 struct vm_area_struct **insertion_point;
2304 struct vm_area_struct *tail_vma = NULL;
2305
2306 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002307 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002309 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 mm->map_count--;
2311 tail_vma = vma;
2312 vma = vma->vm_next;
2313 } while (vma && vma->vm_start < end);
2314 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002315 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002316 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002317 vma_gap_update(vma);
2318 } else
2319 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002321
2322 /* Kill the cache */
2323 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324}
2325
2326/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002327 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2328 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002330static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 unsigned long addr, int new_below)
2332{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002334 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335
Andi Kleena5516432008-07-23 21:27:41 -07002336 if (is_vm_hugetlb_page(vma) && (addr &
2337 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 return -EINVAL;
2339
Christoph Lametere94b1762006-12-06 20:33:17 -08002340 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002342 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
2344 /* most fields are the same, copy all, and then fixup */
2345 *new = *vma;
2346
Rik van Riel5beb4932010-03-05 13:42:07 -08002347 INIT_LIST_HEAD(&new->anon_vma_chain);
2348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 if (new_below)
2350 new->vm_end = addr;
2351 else {
2352 new->vm_start = addr;
2353 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2354 }
2355
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002356 err = vma_dup_policy(vma, new);
2357 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002358 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002360 err = anon_vma_clone(new, vma);
2361 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002362 goto out_free_mpol;
2363
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002364 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 get_file(new->vm_file);
2366
2367 if (new->vm_ops && new->vm_ops->open)
2368 new->vm_ops->open(new);
2369
2370 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002371 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2373 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002374 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375
Rik van Riel5beb4932010-03-05 13:42:07 -08002376 /* Success. */
2377 if (!err)
2378 return 0;
2379
2380 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002381 if (new->vm_ops && new->vm_ops->close)
2382 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002383 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002384 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002385 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002386 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002387 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002388 out_free_vma:
2389 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002390 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391}
2392
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002393/*
2394 * Split a vma into two pieces at address 'addr', a new vma is allocated
2395 * either for the first part or the tail.
2396 */
2397int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2398 unsigned long addr, int new_below)
2399{
2400 if (mm->map_count >= sysctl_max_map_count)
2401 return -ENOMEM;
2402
2403 return __split_vma(mm, vma, addr, new_below);
2404}
2405
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406/* Munmap is split into 2 main parts -- this part which finds
2407 * what needs doing, and the areas themselves, which do the
2408 * work. This now handles partial unmappings.
2409 * Jeremy Fitzhardinge <jeremy@goop.org>
2410 */
2411int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2412{
2413 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002414 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002416 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 return -EINVAL;
2418
vishnu.pscc71aba2014-10-09 15:26:29 -07002419 len = PAGE_ALIGN(len);
2420 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 return -EINVAL;
2422
2423 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002424 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002425 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002427 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002428 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
2430 /* if it doesn't overlap, we have nothing.. */
2431 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002432 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 return 0;
2434
2435 /*
2436 * If we need to split any vma, do it now to save pain later.
2437 *
2438 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2439 * unmapped vm_area_struct will remain in use: so lower split_vma
2440 * places tmp vma above, and higher split_vma places tmp vma below.
2441 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002442 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002443 int error;
2444
2445 /*
2446 * Make sure that map_count on return from munmap() will
2447 * not exceed its limit; but let map_count go just above
2448 * its limit temporarily, to help free resources as expected.
2449 */
2450 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2451 return -ENOMEM;
2452
2453 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 if (error)
2455 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002456 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
2458
2459 /* Does it split the last one? */
2460 last = find_vma(mm, end);
2461 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002462 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 if (error)
2464 return error;
2465 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002466 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
2468 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002469 * unlock any mlock()ed ranges before detaching vmas
2470 */
2471 if (mm->locked_vm) {
2472 struct vm_area_struct *tmp = vma;
2473 while (tmp && tmp->vm_start < end) {
2474 if (tmp->vm_flags & VM_LOCKED) {
2475 mm->locked_vm -= vma_pages(tmp);
2476 munlock_vma_pages_all(tmp);
2477 }
2478 tmp = tmp->vm_next;
2479 }
2480 }
2481
2482 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 * Remove the vma's, and unmap the actual pages
2484 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002485 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2486 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
Dave Hansen1de4fa12014-11-14 07:18:31 -08002488 arch_unmap(mm, vma, start, end);
2489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002491 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
2493 return 0;
2494}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Al Virobfce2812012-04-20 21:57:04 -04002496int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002497{
2498 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002499 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002500
2501 down_write(&mm->mmap_sem);
2502 ret = do_munmap(mm, start, len);
2503 up_write(&mm->mmap_sem);
2504 return ret;
2505}
2506EXPORT_SYMBOL(vm_munmap);
2507
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002508SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 profile_munmap(addr);
Al Virobfce2812012-04-20 21:57:04 -04002511 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512}
2513
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002514
2515/*
2516 * Emulation of deprecated remap_file_pages() syscall.
2517 */
2518SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2519 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2520{
2521
2522 struct mm_struct *mm = current->mm;
2523 struct vm_area_struct *vma;
2524 unsigned long populate = 0;
2525 unsigned long ret = -EINVAL;
2526 struct file *file;
2527
Joe Perches756a025f02016-03-17 14:19:47 -07002528 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2529 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002530
2531 if (prot)
2532 return ret;
2533 start = start & PAGE_MASK;
2534 size = size & PAGE_MASK;
2535
2536 if (start + size <= start)
2537 return ret;
2538
2539 /* Does pgoff wrap? */
2540 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2541 return ret;
2542
2543 down_write(&mm->mmap_sem);
2544 vma = find_vma(mm, start);
2545
2546 if (!vma || !(vma->vm_flags & VM_SHARED))
2547 goto out;
2548
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002549 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002550 goto out;
2551
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002552 if (start + size > vma->vm_end) {
2553 struct vm_area_struct *next;
2554
2555 for (next = vma->vm_next; next; next = next->vm_next) {
2556 /* hole between vmas ? */
2557 if (next->vm_start != next->vm_prev->vm_end)
2558 goto out;
2559
2560 if (next->vm_file != vma->vm_file)
2561 goto out;
2562
2563 if (next->vm_flags != vma->vm_flags)
2564 goto out;
2565
2566 if (start + size <= next->vm_end)
2567 break;
2568 }
2569
2570 if (!next)
2571 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002572 }
2573
2574 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2575 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2576 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2577
2578 flags &= MAP_NONBLOCK;
2579 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2580 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002581 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002582 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002583
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002584 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002585 for (tmp = vma; tmp->vm_start >= start + size;
2586 tmp = tmp->vm_next) {
2587 munlock_vma_pages_range(tmp,
2588 max(tmp->vm_start, start),
2589 min(tmp->vm_end, start + size));
2590 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002591 }
2592
2593 file = get_file(vma->vm_file);
2594 ret = do_mmap_pgoff(vma->vm_file, start, size,
2595 prot, flags, pgoff, &populate);
2596 fput(file);
2597out:
2598 up_write(&mm->mmap_sem);
2599 if (populate)
2600 mm_populate(ret, populate);
2601 if (!IS_ERR_VALUE(ret))
2602 ret = 0;
2603 return ret;
2604}
2605
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606static inline void verify_mm_writelocked(struct mm_struct *mm)
2607{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002608#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2610 WARN_ON(1);
2611 up_read(&mm->mmap_sem);
2612 }
2613#endif
2614}
2615
2616/*
2617 * this is really a simplified "do_mmap". it only handles
2618 * anonymous maps. eventually we may be able to do some
2619 * brk-specific accounting here.
2620 */
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002621static unsigned long do_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622{
vishnu.pscc71aba2014-10-09 15:26:29 -07002623 struct mm_struct *mm = current->mm;
2624 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 unsigned long flags;
vishnu.pscc71aba2014-10-09 15:26:29 -07002626 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002628 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
2630 len = PAGE_ALIGN(len);
2631 if (!len)
2632 return addr;
2633
Kirill Korotaev3a459752006-09-07 14:17:04 +04002634 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2635
Al Viro2c6a1012009-12-03 19:40:46 -05002636 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002637 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002638 return error;
2639
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002640 error = mlock_future_check(mm, mm->def_flags, len);
2641 if (error)
2642 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
2644 /*
2645 * mm->mmap_sem is required to protect against another thread
2646 * changing the mappings in case we sleep.
2647 */
2648 verify_mm_writelocked(mm);
2649
2650 /*
2651 * Clear old maps. this also does some error checking for us
2652 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002653 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2654 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 if (do_munmap(mm, addr, len))
2656 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 }
2658
2659 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002660 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 return -ENOMEM;
2662
2663 if (mm->map_count > sysctl_max_map_count)
2664 return -ENOMEM;
2665
Al Viro191c5422012-02-13 03:58:52 +00002666 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 return -ENOMEM;
2668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002670 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002671 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002672 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 goto out;
2674
2675 /*
2676 * create a vma struct for an anonymous mapping
2677 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002678 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 if (!vma) {
2680 vm_unacct_memory(len >> PAGE_SHIFT);
2681 return -ENOMEM;
2682 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683
Rik van Riel5beb4932010-03-05 13:42:07 -08002684 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 vma->vm_mm = mm;
2686 vma->vm_start = addr;
2687 vma->vm_end = addr + len;
2688 vma->vm_pgoff = pgoff;
2689 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002690 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 vma_link(mm, vma, prev, rb_link, rb_parent);
2692out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002693 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002695 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002696 if (flags & VM_LOCKED)
2697 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002698 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 return addr;
2700}
2701
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002702unsigned long vm_brk(unsigned long addr, unsigned long len)
2703{
2704 struct mm_struct *mm = current->mm;
2705 unsigned long ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002706 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002707
2708 down_write(&mm->mmap_sem);
2709 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002710 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002711 up_write(&mm->mmap_sem);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002712 if (populate)
2713 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002714 return ret;
2715}
2716EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
2718/* Release all mmaps. */
2719void exit_mmap(struct mm_struct *mm)
2720{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002721 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002722 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 unsigned long nr_accounted = 0;
2724
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002725 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002726 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002727
Rik van Rielba470de2008-10-18 20:26:50 -07002728 if (mm->locked_vm) {
2729 vma = mm->mmap;
2730 while (vma) {
2731 if (vma->vm_flags & VM_LOCKED)
2732 munlock_vma_pages_all(vma);
2733 vma = vma->vm_next;
2734 }
2735 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002736
2737 arch_exit_mmap(mm);
2738
Rik van Rielba470de2008-10-18 20:26:50 -07002739 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002740 if (!vma) /* Can happen if dup_mmap() received an OOM */
2741 return;
2742
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002745 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002746 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002747 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002748 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002749
Hugh Dickins6ee86302013-04-29 15:07:44 -07002750 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002751 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002754 * Walk the list again, actually closing and freeing it,
2755 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002757 while (vma) {
2758 if (vma->vm_flags & VM_ACCOUNT)
2759 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002760 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002761 }
2762 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763}
2764
2765/* Insert vm structure into process list sorted by address
2766 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002767 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002769int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770{
Hugh Dickins6597d782012-10-08 16:29:07 -07002771 struct vm_area_struct *prev;
2772 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
Chen Gangc9d13f52015-09-08 15:04:08 -07002774 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2775 &prev, &rb_link, &rb_parent))
2776 return -ENOMEM;
2777 if ((vma->vm_flags & VM_ACCOUNT) &&
2778 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2779 return -ENOMEM;
2780
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 /*
2782 * The vm_pgoff of a purely anonymous vma should be irrelevant
2783 * until its first write fault, when page's anon_vma and index
2784 * are set. But now set the vm_pgoff it will almost certainly
2785 * end up with (unless mremap moves it elsewhere before that
2786 * first wfault), so /proc/pid/maps tells a consistent story.
2787 *
2788 * By setting it to reflect the virtual start address of the
2789 * vma, merges and splits can happen in a seamless way, just
2790 * using the existing file pgoff checks and manipulations.
2791 * Similarly in do_mmap_pgoff and in do_brk.
2792 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002793 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 BUG_ON(vma->anon_vma);
2795 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2796 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302797
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 vma_link(mm, vma, prev, rb_link, rb_parent);
2799 return 0;
2800}
2801
2802/*
2803 * Copy the vma structure to a new location in the same mm,
2804 * prior to moving page table entries, to effect an mremap move.
2805 */
2806struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002807 unsigned long addr, unsigned long len, pgoff_t pgoff,
2808 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809{
2810 struct vm_area_struct *vma = *vmap;
2811 unsigned long vma_start = vma->vm_start;
2812 struct mm_struct *mm = vma->vm_mm;
2813 struct vm_area_struct *new_vma, *prev;
2814 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002815 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816
2817 /*
2818 * If anonymous vma has not yet been faulted, update new pgoff
2819 * to match new location, to increase its chance of merging.
2820 */
Oleg Nesterovce757992015-09-08 14:58:34 -07002821 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002823 faulted_in_anon_vma = false;
2824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825
Hugh Dickins6597d782012-10-08 16:29:07 -07002826 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2827 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002829 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2830 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 if (new_vma) {
2832 /*
2833 * Source vma may have been merged into new_vma
2834 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08002835 if (unlikely(vma_start >= new_vma->vm_start &&
2836 vma_start < new_vma->vm_end)) {
2837 /*
2838 * The only way we can get a vma_merge with
2839 * self during an mremap is if the vma hasn't
2840 * been faulted in yet and we were allowed to
2841 * reset the dst vma->vm_pgoff to the
2842 * destination address of the mremap to allow
2843 * the merge to happen. mremap must change the
2844 * vm_pgoff linearity between src and dst vmas
2845 * (in turn preventing a vma_merge) to be
2846 * safe. It is only safe to keep the vm_pgoff
2847 * linear if there are no pages mapped yet.
2848 */
Sasha Levin81d1b092014-10-09 15:28:10 -07002849 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07002850 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07002851 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07002852 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08002854 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07002855 if (!new_vma)
2856 goto out;
2857 *new_vma = *vma;
2858 new_vma->vm_start = addr;
2859 new_vma->vm_end = addr + len;
2860 new_vma->vm_pgoff = pgoff;
2861 if (vma_dup_policy(vma, new_vma))
2862 goto out_free_vma;
2863 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2864 if (anon_vma_clone(new_vma, vma))
2865 goto out_free_mempol;
2866 if (new_vma->vm_file)
2867 get_file(new_vma->vm_file);
2868 if (new_vma->vm_ops && new_vma->vm_ops->open)
2869 new_vma->vm_ops->open(new_vma);
2870 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2871 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 }
2873 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08002874
Chen Gange3975892015-09-08 15:03:38 -07002875out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002876 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07002877out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08002878 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07002879out:
Rik van Riel5beb4932010-03-05 13:42:07 -08002880 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881}
akpm@osdl.org119f6572005-05-01 08:58:35 -07002882
2883/*
2884 * Return true if the calling process may expand its vm space by the passed
2885 * number of pages
2886 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002887bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07002888{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002889 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
2890 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002891
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002892 if (is_data_mapping(flags) &&
2893 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
2894 if (ignore_rlimit_data)
Joe Perches756a025f02016-03-17 14:19:47 -07002895 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Will be forbidden soon.\n",
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002896 current->comm, current->pid,
2897 (mm->data_vm + npages) << PAGE_SHIFT,
2898 rlimit(RLIMIT_DATA));
2899 else
2900 return false;
2901 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07002902
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002903 return true;
2904}
2905
2906void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
2907{
2908 mm->total_vm += npages;
2909
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002910 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002911 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002912 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002913 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08002914 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002915 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07002916}
Roland McGrathfa5dc222007-02-08 14:20:41 -08002917
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002918static int special_mapping_fault(struct vm_area_struct *vma,
2919 struct vm_fault *vmf);
2920
2921/*
2922 * Having a close hook prevents vma merging regardless of flags.
2923 */
2924static void special_mapping_close(struct vm_area_struct *vma)
2925{
2926}
2927
2928static const char *special_mapping_name(struct vm_area_struct *vma)
2929{
2930 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
2931}
2932
2933static const struct vm_operations_struct special_mapping_vmops = {
2934 .close = special_mapping_close,
2935 .fault = special_mapping_fault,
2936 .name = special_mapping_name,
2937};
2938
2939static const struct vm_operations_struct legacy_special_mapping_vmops = {
2940 .close = special_mapping_close,
2941 .fault = special_mapping_fault,
2942};
Roland McGrathfa5dc222007-02-08 14:20:41 -08002943
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002944static int special_mapping_fault(struct vm_area_struct *vma,
2945 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002946{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002947 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002948 struct page **pages;
2949
Andy Lutomirskif872f542015-12-29 20:12:19 -08002950 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002951 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08002952 } else {
2953 struct vm_special_mapping *sm = vma->vm_private_data;
2954
2955 if (sm->fault)
2956 return sm->fault(sm, vma, vmf);
2957
2958 pages = sm->pages;
2959 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002960
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002961 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002962 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002963
2964 if (*pages) {
2965 struct page *page = *pages;
2966 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002967 vmf->page = page;
2968 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002969 }
2970
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01002971 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002972}
2973
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002974static struct vm_area_struct *__install_special_mapping(
2975 struct mm_struct *mm,
2976 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08002977 unsigned long vm_flags, void *priv,
2978 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08002979{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002980 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002981 struct vm_area_struct *vma;
2982
2983 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2984 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01002985 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002986
Rik van Riel5beb4932010-03-05 13:42:07 -08002987 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002988 vma->vm_mm = mm;
2989 vma->vm_start = addr;
2990 vma->vm_end = addr + len;
2991
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002992 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07002993 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08002994
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07002995 vma->vm_ops = ops;
2996 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08002997
Tavis Ormandy462e635e2010-12-09 15:29:42 +01002998 ret = insert_vm_struct(mm, vma);
2999 if (ret)
3000 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003001
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003002 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003003
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003004 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003005
Stefani Seibold3935ed62014-03-17 23:22:02 +01003006 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003007
3008out:
3009 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003010 return ERR_PTR(ret);
3011}
3012
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003013/*
3014 * Called with mm->mmap_sem held for writing.
3015 * Insert a new vma covering the given region, with the given flags.
3016 * Its pages are supplied by the given array of struct page *.
3017 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3018 * The region past the last page supplied will always produce SIGBUS.
3019 * The array pointer and the pages it points to are assumed to stay alive
3020 * for as long as this mapping might exist.
3021 */
3022struct vm_area_struct *_install_special_mapping(
3023 struct mm_struct *mm,
3024 unsigned long addr, unsigned long len,
3025 unsigned long vm_flags, const struct vm_special_mapping *spec)
3026{
Chen Gang27f28b92015-11-05 18:48:41 -08003027 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3028 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003029}
3030
Stefani Seibold3935ed62014-03-17 23:22:02 +01003031int install_special_mapping(struct mm_struct *mm,
3032 unsigned long addr, unsigned long len,
3033 unsigned long vm_flags, struct page **pages)
3034{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003035 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003036 mm, addr, len, vm_flags, (void *)pages,
3037 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003038
Duan Jiong14bd5b42014-06-04 16:07:05 -07003039 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003040}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003041
3042static DEFINE_MUTEX(mm_all_locks_mutex);
3043
Peter Zijlstra454ed842008-08-11 09:30:25 +02003044static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003045{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003046 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003047 /*
3048 * The LSB of head.next can't change from under us
3049 * because we hold the mm_all_locks_mutex.
3050 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003051 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003052 /*
3053 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003054 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003055 * the same anon_vma we won't take it again.
3056 *
3057 * No need of atomic instructions here, head.next
3058 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003059 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003060 */
3061 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003062 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003063 BUG();
3064 }
3065}
3066
Peter Zijlstra454ed842008-08-11 09:30:25 +02003067static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003068{
3069 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3070 /*
3071 * AS_MM_ALL_LOCKS can't change from under us because
3072 * we hold the mm_all_locks_mutex.
3073 *
3074 * Operations on ->flags have to be atomic because
3075 * even if AS_MM_ALL_LOCKS is stable thanks to the
3076 * mm_all_locks_mutex, there may be other cpus
3077 * changing other bitflags in parallel to us.
3078 */
3079 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3080 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003081 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003082 }
3083}
3084
3085/*
3086 * This operation locks against the VM for all pte/vma/mm related
3087 * operations that could ever happen on a certain mm. This includes
3088 * vmtruncate, try_to_unmap, and all page faults.
3089 *
3090 * The caller must take the mmap_sem in write mode before calling
3091 * mm_take_all_locks(). The caller isn't allowed to release the
3092 * mmap_sem until mm_drop_all_locks() returns.
3093 *
3094 * mmap_sem in write mode is required in order to block all operations
3095 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003096 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003097 * anon_vmas to be associated with existing vmas.
3098 *
3099 * A single task can't take more than one mm_take_all_locks() in a row
3100 * or it would deadlock.
3101 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003102 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003103 * mapping->flags avoid to take the same lock twice, if more than one
3104 * vma in this mm is backed by the same anon_vma or address_space.
3105 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003106 * We take locks in following order, accordingly to comment at beginning
3107 * of mm/rmap.c:
3108 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3109 * hugetlb mapping);
3110 * - all i_mmap_rwsem locks;
3111 * - all anon_vma->rwseml
3112 *
3113 * We can take all locks within these types randomly because the VM code
3114 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3115 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003116 *
3117 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3118 * that may have to take thousand of locks.
3119 *
3120 * mm_take_all_locks() can fail if it's interrupted by signals.
3121 */
3122int mm_take_all_locks(struct mm_struct *mm)
3123{
3124 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003125 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003126
3127 BUG_ON(down_read_trylock(&mm->mmap_sem));
3128
3129 mutex_lock(&mm_all_locks_mutex);
3130
3131 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3132 if (signal_pending(current))
3133 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003134 if (vma->vm_file && vma->vm_file->f_mapping &&
3135 is_vm_hugetlb_page(vma))
3136 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3137 }
3138
3139 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3140 if (signal_pending(current))
3141 goto out_unlock;
3142 if (vma->vm_file && vma->vm_file->f_mapping &&
3143 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003144 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003145 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003146
3147 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3148 if (signal_pending(current))
3149 goto out_unlock;
3150 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003151 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3152 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003153 }
3154
Kautuk Consul584cff52011-10-31 17:08:59 -07003155 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003156
3157out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003158 mm_drop_all_locks(mm);
3159 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003160}
3161
3162static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3163{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003164 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003165 /*
3166 * The LSB of head.next can't change to 0 from under
3167 * us because we hold the mm_all_locks_mutex.
3168 *
3169 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003170 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003171 * never see our bitflag.
3172 *
3173 * No need of atomic instructions here, head.next
3174 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003175 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003176 */
3177 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003178 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003179 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003180 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003181 }
3182}
3183
3184static void vm_unlock_mapping(struct address_space *mapping)
3185{
3186 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3187 /*
3188 * AS_MM_ALL_LOCKS can't change to 0 from under us
3189 * because we hold the mm_all_locks_mutex.
3190 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003191 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003192 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3193 &mapping->flags))
3194 BUG();
3195 }
3196}
3197
3198/*
3199 * The mmap_sem cannot be released by the caller until
3200 * mm_drop_all_locks() returns.
3201 */
3202void mm_drop_all_locks(struct mm_struct *mm)
3203{
3204 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003205 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003206
3207 BUG_ON(down_read_trylock(&mm->mmap_sem));
3208 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3209
3210 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3211 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003212 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3213 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003214 if (vma->vm_file && vma->vm_file->f_mapping)
3215 vm_unlock_mapping(vma->vm_file->f_mapping);
3216 }
3217
3218 mutex_unlock(&mm_all_locks_mutex);
3219}
David Howells8feae132009-01-08 12:04:47 +00003220
3221/*
3222 * initialise the VMA slab
3223 */
3224void __init mmap_init(void)
3225{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003226 int ret;
3227
Tejun Heo908c7f12014-09-08 09:51:29 +09003228 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003229 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003230}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003231
3232/*
3233 * Initialise sysctl_user_reserve_kbytes.
3234 *
3235 * This is intended to prevent a user from starting a single memory hogging
3236 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3237 * mode.
3238 *
3239 * The default value is min(3% of free memory, 128MB)
3240 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3241 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003242static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003243{
3244 unsigned long free_kbytes;
3245
3246 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3247
3248 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3249 return 0;
3250}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003251subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003252
3253/*
3254 * Initialise sysctl_admin_reserve_kbytes.
3255 *
3256 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3257 * to log in and kill a memory hogging process.
3258 *
3259 * Systems with more than 256MB will reserve 8MB, enough to recover
3260 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3261 * only reserve 3% of free pages by default.
3262 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003263static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003264{
3265 unsigned long free_kbytes;
3266
3267 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3268
3269 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3270 return 0;
3271}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003272subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003273
3274/*
3275 * Reinititalise user and admin reserves if memory is added or removed.
3276 *
3277 * The default user reserve max is 128MB, and the default max for the
3278 * admin reserve is 8MB. These are usually, but not always, enough to
3279 * enable recovery from a memory hogging process using login/sshd, a shell,
3280 * and tools like top. It may make sense to increase or even disable the
3281 * reserve depending on the existence of swap or variations in the recovery
3282 * tools. So, the admin may have changed them.
3283 *
3284 * If memory is added and the reserves have been eliminated or increased above
3285 * the default max, then we'll trust the admin.
3286 *
3287 * If memory is removed and there isn't enough free memory, then we
3288 * need to reset the reserves.
3289 *
3290 * Otherwise keep the reserve set by the admin.
3291 */
3292static int reserve_mem_notifier(struct notifier_block *nb,
3293 unsigned long action, void *data)
3294{
3295 unsigned long tmp, free_kbytes;
3296
3297 switch (action) {
3298 case MEM_ONLINE:
3299 /* Default max is 128MB. Leave alone if modified by operator. */
3300 tmp = sysctl_user_reserve_kbytes;
3301 if (0 < tmp && tmp < (1UL << 17))
3302 init_user_reserve();
3303
3304 /* Default max is 8MB. Leave alone if modified by operator. */
3305 tmp = sysctl_admin_reserve_kbytes;
3306 if (0 < tmp && tmp < (1UL << 13))
3307 init_admin_reserve();
3308
3309 break;
3310 case MEM_OFFLINE:
3311 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3312
3313 if (sysctl_user_reserve_kbytes > free_kbytes) {
3314 init_user_reserve();
3315 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3316 sysctl_user_reserve_kbytes);
3317 }
3318
3319 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3320 init_admin_reserve();
3321 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3322 sysctl_admin_reserve_kbytes);
3323 }
3324 break;
3325 default:
3326 break;
3327 }
3328 return NOTIFY_OK;
3329}
3330
3331static struct notifier_block reserve_mem_nb = {
3332 .notifier_call = reserve_mem_notifier,
3333};
3334
3335static int __meminit init_reserve_notifier(void)
3336{
3337 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003338 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003339
3340 return 0;
3341}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003342subsys_initcall(init_reserve_notifier);