blob: 5df92da6e02e5258333f416569ca43040fb46790 [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>
Hugh Dickinsc01d5b32016-07-26 15:26:15 -070028#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/profile.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040030#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/mount.h>
32#include <linux/mempolicy.h>
33#include <linux/rmap.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070034#include <linux/mmu_notifier.h>
Konstantin Khlebnikov82f71ae2014-08-06 16:06:36 -070035#include <linux/mmdebug.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020036#include <linux/perf_event.h>
Al Viro120a7952010-10-30 02:54:44 -040037#include <linux/audit.h>
Andrea Arcangelib15d00b2011-01-13 15:46:59 -080038#include <linux/khugepaged.h>
Srikar Dronamraju2b144492012-02-09 14:56:42 +053039#include <linux/uprobes.h>
Michel Lespinassed3737182012-12-11 16:01:38 -080040#include <linux/rbtree_augmented.h>
Andrew Shewmaker16408792013-04-29 15:08:12 -070041#include <linux/notifier.h>
42#include <linux/memory.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070043#include <linux/printk.h>
Andrea Arcangeli19a809a2015-09-04 15:46:24 -070044#include <linux/userfaultfd_k.h>
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080045#include <linux/moduleparam.h>
Dave Hansen62b5f7d2016-02-12 13:02:40 -080046#include <linux/pkeys.h>
Andrea Arcangeli07b039e92017-09-06 16:25:00 -070047#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49#include <asm/uaccess.h>
50#include <asm/cacheflush.h>
51#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020052#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Jan Beulich42b77722008-07-23 21:27:10 -070054#include "internal.h"
55
Kirill Korotaev3a459752006-09-07 14:17:04 +040056#ifndef arch_mmap_check
57#define arch_mmap_check(addr, len, flags) (0)
58#endif
59
Daniel Cashmand07e2252016-01-14 15:19:53 -080060#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
61const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
62const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
63int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
64#endif
65#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
66const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
67const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
68int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
69#endif
70
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -070071static bool ignore_rlimit_data;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080072core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
Daniel Cashmand07e2252016-01-14 15:19:53 -080073
Hugh Dickinse0da3822005-04-19 13:29:15 -070074static void unmap_region(struct mm_struct *mm,
75 struct vm_area_struct *vma, struct vm_area_struct *prev,
76 unsigned long start, unsigned long end);
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/* description of effects of mapping type and prot in current implementation.
79 * this is due to the limited x86 page protection hardware. The expected
80 * behavior is in parens:
81 *
82 * map_type prot
83 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
84 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
85 * w: (no) no w: (no) no w: (yes) yes w: (no) no
86 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070087 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
89 * w: (no) no w: (no) no w: (copy) copy w: (no) no
90 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
91 *
Catalin Marinascab15ce2016-08-11 18:44:50 +010092 * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
93 * MAP_PRIVATE:
94 * r: (no) no
95 * w: (no) no
96 * x: (yes) yes
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 */
98pgprot_t protection_map[16] = {
99 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
100 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
101};
102
Hugh Dickins804af2c2006-07-26 21:39:49 +0100103pgprot_t vm_get_page_prot(unsigned long vm_flags)
104{
Dave Kleikampb845f312008-07-08 00:28:51 +1000105 return __pgprot(pgprot_val(protection_map[vm_flags &
106 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
107 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100108}
109EXPORT_SYMBOL(vm_get_page_prot);
110
Peter Feiner64e45502014-10-13 15:55:46 -0700111static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
112{
113 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
114}
115
116/* Update vma->vm_page_prot to reflect vma->vm_flags. */
117void vma_set_page_prot(struct vm_area_struct *vma)
118{
119 unsigned long vm_flags = vma->vm_flags;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700120 pgprot_t vm_page_prot;
Peter Feiner64e45502014-10-13 15:55:46 -0700121
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700122 vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
123 if (vma_wants_writenotify(vma, vm_page_prot)) {
Peter Feiner64e45502014-10-13 15:55:46 -0700124 vm_flags &= ~VM_SHARED;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700125 vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
Peter Feiner64e45502014-10-13 15:55:46 -0700126 }
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700127 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
128 WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
Peter Feiner64e45502014-10-13 15:55:46 -0700129}
130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800132 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 */
134static void __remove_shared_vm_struct(struct vm_area_struct *vma,
135 struct file *file, struct address_space *mapping)
136{
137 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500138 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700140 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800143 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 flush_dcache_mmap_unlock(mapping);
145}
146
147/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700148 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700149 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700151void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{
153 struct file *file = vma->vm_file;
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (file) {
156 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800157 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800159 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700161}
162
163/*
164 * Close a vm structure and free it, returning the next.
165 */
166static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
167{
168 struct vm_area_struct *next = vma->vm_next;
169
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700170 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 if (vma->vm_ops && vma->vm_ops->close)
172 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700173 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700174 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700175 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700177 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700180static int do_brk(unsigned long addr, unsigned long len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700181
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100182SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700184 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 unsigned long newbrk, oldbrk;
186 struct mm_struct *mm = current->mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700187 struct vm_area_struct *next;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700188 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800189 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700191 if (down_write_killable(&mm->mmap_sem))
192 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700194#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800195 /*
196 * CONFIG_COMPAT_BRK can still be overridden by setting
197 * randomize_va_space to 2, which will still cause mm->start_brk
198 * to be arbitrarily shifted
199 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700200 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800201 min_brk = mm->start_brk;
202 else
203 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700204#else
205 min_brk = mm->start_brk;
206#endif
207 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700209
210 /*
211 * Check against rlimit here. If this check is done later after the test
212 * of oldbrk with newbrk then it can escape the test and let the data
213 * segment grow beyond its set limit the in case where the limit is
214 * not page aligned -Ram Gupta
215 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700216 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
217 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700218 goto out;
219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 newbrk = PAGE_ALIGN(brk);
221 oldbrk = PAGE_ALIGN(mm->brk);
222 if (oldbrk == newbrk)
223 goto set_brk;
224
225 /* Always allow shrinking brk. */
226 if (brk <= mm->brk) {
227 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
228 goto set_brk;
229 goto out;
230 }
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 /* Check against existing mmap mappings. */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700233 next = find_vma(mm, oldbrk);
234 if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 goto out;
236
237 /* Ok, looks good - let it rip. */
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700238 if (do_brk(oldbrk, newbrk-oldbrk) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241set_brk:
242 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800243 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
244 up_write(&mm->mmap_sem);
245 if (populate)
246 mm_populate(oldbrk, newbrk - oldbrk);
247 return brk;
248
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249out:
250 retval = mm->brk;
251 up_write(&mm->mmap_sem);
252 return retval;
253}
254
Michel Lespinassed3737182012-12-11 16:01:38 -0800255static long vma_compute_subtree_gap(struct vm_area_struct *vma)
256{
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700257 unsigned long max, prev_end, subtree_gap;
258
259 /*
260 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
261 * allow two stack_guard_gaps between them here, and when choosing
262 * an unmapped area; whereas when expanding we only require one.
263 * That's a little inconsistent, but keeps the code here simpler.
264 */
265 max = vm_start_gap(vma);
266 if (vma->vm_prev) {
267 prev_end = vm_end_gap(vma->vm_prev);
268 if (max > prev_end)
269 max -= prev_end;
270 else
271 max = 0;
272 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800273 if (vma->vm_rb.rb_left) {
274 subtree_gap = rb_entry(vma->vm_rb.rb_left,
275 struct vm_area_struct, vm_rb)->rb_subtree_gap;
276 if (subtree_gap > max)
277 max = subtree_gap;
278 }
279 if (vma->vm_rb.rb_right) {
280 subtree_gap = rb_entry(vma->vm_rb.rb_right,
281 struct vm_area_struct, vm_rb)->rb_subtree_gap;
282 if (subtree_gap > max)
283 max = subtree_gap;
284 }
285 return max;
286}
287
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700288#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800289static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800291 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800292 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 struct rb_node *nd, *pn = NULL;
294 unsigned long prev = 0, pend = 0;
295
296 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
297 struct vm_area_struct *vma;
298 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800299 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700300 pr_emerg("vm_start %lx < prev %lx\n",
301 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800302 bug = 1;
303 }
304 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700305 pr_emerg("vm_start %lx < pend %lx\n",
306 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800307 bug = 1;
308 }
309 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700310 pr_emerg("vm_start %lx > vm_end %lx\n",
311 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800312 bug = 1;
313 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800314 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800315 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700316 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800317 vma->rb_subtree_gap,
318 vma_compute_subtree_gap(vma));
319 bug = 1;
320 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800321 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 i++;
323 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800324 prev = vma->vm_start;
325 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 }
327 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800328 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800330 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700331 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800332 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800334 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335}
336
Michel Lespinassed3737182012-12-11 16:01:38 -0800337static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
338{
339 struct rb_node *nd;
340
341 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
342 struct vm_area_struct *vma;
343 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700344 VM_BUG_ON_VMA(vma != ignore &&
345 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
346 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700350static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
352 int bug = 0;
353 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800354 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700355 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700356
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700357 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800358 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700359 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700360
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800361 if (anon_vma) {
362 anon_vma_lock_read(anon_vma);
363 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
364 anon_vma_interval_tree_verify(avc);
365 anon_vma_unlock_read(anon_vma);
366 }
367
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700368 highest_address = vm_end_gap(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700369 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 i++;
371 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800372 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700373 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800374 bug = 1;
375 }
376 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700377 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700378 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800379 bug = 1;
380 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800381 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800382 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700383 if (i != -1)
384 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800385 bug = 1;
386 }
Sasha Levin96dad672014-10-09 15:28:39 -0700387 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800390#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391#define validate_mm(mm) do { } while (0)
392#endif
393
Michel Lespinassed3737182012-12-11 16:01:38 -0800394RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
395 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
396
397/*
398 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
399 * vma->vm_prev->vm_end values changed, without modifying the vma's position
400 * in the rbtree.
401 */
402static void vma_gap_update(struct vm_area_struct *vma)
403{
404 /*
405 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
406 * function that does exacltly what we want.
407 */
408 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
409}
410
411static inline void vma_rb_insert(struct vm_area_struct *vma,
412 struct rb_root *root)
413{
414 /* All rb_subtree_gap values must be consistent prior to insertion */
415 validate_mm_rb(root, NULL);
416
417 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
418}
419
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700420static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
421{
422 /*
423 * Note rb_erase_augmented is a fairly large inline function,
424 * so make sure we instantiate it only once with our desired
425 * augmented rbtree callbacks.
426 */
427 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
428}
429
430static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
431 struct rb_root *root,
432 struct vm_area_struct *ignore)
433{
434 /*
435 * All rb_subtree_gap values must be consistent prior to erase,
436 * with the possible exception of the "next" vma being erased if
437 * next->vm_start was reduced.
438 */
439 validate_mm_rb(root, ignore);
440
441 __vma_rb_erase(vma, root);
442}
443
444static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
445 struct rb_root *root)
Michel Lespinassed3737182012-12-11 16:01:38 -0800446{
447 /*
448 * All rb_subtree_gap values must be consistent prior to erase,
449 * with the possible exception of the vma being erased.
450 */
451 validate_mm_rb(root, vma);
452
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700453 __vma_rb_erase(vma, root);
Michel Lespinassed3737182012-12-11 16:01:38 -0800454}
455
Michel Lespinassebf181b92012-10-08 16:31:39 -0700456/*
457 * vma has some anon_vma assigned, and is already inserted on that
458 * anon_vma's interval trees.
459 *
460 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
461 * vma must be removed from the anon_vma's interval trees using
462 * anon_vma_interval_tree_pre_update_vma().
463 *
464 * After the update, the vma will be reinserted using
465 * anon_vma_interval_tree_post_update_vma().
466 *
467 * The entire update must be protected by exclusive mmap_sem and by
468 * the root anon_vma's mutex.
469 */
470static inline void
471anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
472{
473 struct anon_vma_chain *avc;
474
475 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
476 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
477}
478
479static inline void
480anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
481{
482 struct anon_vma_chain *avc;
483
484 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
485 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
486}
487
Hugh Dickins6597d782012-10-08 16:29:07 -0700488static int find_vma_links(struct mm_struct *mm, unsigned long addr,
489 unsigned long end, struct vm_area_struct **pprev,
490 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
Hugh Dickins6597d782012-10-08 16:29:07 -0700492 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 __rb_link = &mm->mm_rb.rb_node;
495 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497 while (*__rb_link) {
498 struct vm_area_struct *vma_tmp;
499
500 __rb_parent = *__rb_link;
501 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
502
503 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700504 /* Fail if an existing vma overlaps the area */
505 if (vma_tmp->vm_start < end)
506 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 __rb_link = &__rb_parent->rb_left;
508 } else {
509 rb_prev = __rb_parent;
510 __rb_link = &__rb_parent->rb_right;
511 }
512 }
513
514 *pprev = NULL;
515 if (rb_prev)
516 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
517 *rb_link = __rb_link;
518 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700519 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520}
521
Cyril Hrubise8420a82013-04-29 15:08:33 -0700522static unsigned long count_vma_pages_range(struct mm_struct *mm,
523 unsigned long addr, unsigned long end)
524{
525 unsigned long nr_pages = 0;
526 struct vm_area_struct *vma;
527
528 /* Find first overlaping mapping */
529 vma = find_vma_intersection(mm, addr, end);
530 if (!vma)
531 return 0;
532
533 nr_pages = (min(end, vma->vm_end) -
534 max(addr, vma->vm_start)) >> PAGE_SHIFT;
535
536 /* Iterate over the rest of the overlaps */
537 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
538 unsigned long overlap_len;
539
540 if (vma->vm_start > end)
541 break;
542
543 overlap_len = min(end, vma->vm_end) - vma->vm_start;
544 nr_pages += overlap_len >> PAGE_SHIFT;
545 }
546
547 return nr_pages;
548}
549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
551 struct rb_node **rb_link, struct rb_node *rb_parent)
552{
Michel Lespinassed3737182012-12-11 16:01:38 -0800553 /* Update tracking information for the gap following the new vma. */
554 if (vma->vm_next)
555 vma_gap_update(vma->vm_next);
556 else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700557 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800558
559 /*
560 * vma->vm_prev wasn't known when we followed the rbtree to find the
561 * correct insertion point for that vma. As a result, we could not
562 * update the vma vm_rb parents rb_subtree_gap values on the way down.
563 * So, we first insert the vma with a zero rb_subtree_gap value
564 * (to be consistent with what we did on the way down), and then
565 * immediately update the gap to the correct value. Finally we
566 * rebalance the rbtree after all augmented values have been set.
567 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800569 vma->rb_subtree_gap = 0;
570 vma_gap_update(vma);
571 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572}
573
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700574static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575{
ZhenwenXu48aae422009-01-06 14:40:21 -0800576 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 file = vma->vm_file;
579 if (file) {
580 struct address_space *mapping = file->f_mapping;
581
582 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500583 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700585 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800588 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 flush_dcache_mmap_unlock(mapping);
590 }
591}
592
593static void
594__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
595 struct vm_area_struct *prev, struct rb_node **rb_link,
596 struct rb_node *rb_parent)
597{
598 __vma_link_list(mm, vma, prev, rb_parent);
599 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
602static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
603 struct vm_area_struct *prev, struct rb_node **rb_link,
604 struct rb_node *rb_parent)
605{
606 struct address_space *mapping = NULL;
607
Huang Shijie64ac4942014-06-04 16:07:33 -0700608 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800610 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
613 __vma_link(mm, vma, prev, rb_link, rb_parent);
614 __vma_link_file(vma);
615
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800617 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 mm->map_count++;
620 validate_mm(mm);
621}
622
623/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700624 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700625 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800627static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
Hugh Dickins6597d782012-10-08 16:29:07 -0700629 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800630 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Hugh Dickins6597d782012-10-08 16:29:07 -0700632 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
633 &prev, &rb_link, &rb_parent))
634 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 __vma_link(mm, vma, prev, rb_link, rb_parent);
636 mm->map_count++;
637}
638
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700639static __always_inline void __vma_unlink_common(struct mm_struct *mm,
640 struct vm_area_struct *vma,
641 struct vm_area_struct *prev,
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700642 bool has_prev,
643 struct vm_area_struct *ignore)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
Michel Lespinassed3737182012-12-11 16:01:38 -0800645 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700646
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700647 vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700648 next = vma->vm_next;
649 if (has_prev)
650 prev->vm_next = next;
651 else {
652 prev = vma->vm_prev;
653 if (prev)
654 prev->vm_next = next;
655 else
656 mm->mmap = next;
657 }
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700658 if (next)
659 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700660
661 /* Kill the cache */
662 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700665static inline void __vma_unlink_prev(struct mm_struct *mm,
666 struct vm_area_struct *vma,
667 struct vm_area_struct *prev)
668{
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700669 __vma_unlink_common(mm, vma, prev, true, vma);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700670}
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672/*
673 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
674 * is already present in an i_mmap tree without adjusting the tree.
675 * The following helper function should be used when such adjustments
676 * are necessary. The "insert" vma (if any) is to be inserted
677 * before we drop the necessary locks.
678 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700679int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
680 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
681 struct vm_area_struct *expand)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
683 struct mm_struct *mm = vma->vm_mm;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700684 struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700686 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700687 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800689 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 long adjust_next = 0;
691 int remove_next = 0;
692
693 if (next && !insert) {
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700694 struct vm_area_struct *exporter = NULL, *importer = NULL;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 if (end >= next->vm_end) {
697 /*
698 * vma expands, overlapping all the next, and
699 * perhaps the one after too (mprotect case 6).
Andrea Arcangeli86d12e42016-10-07 17:01:34 -0700700 * The only other cases that gets here are
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700701 * case 1, case 7 and case 8.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700703 if (next == expand) {
704 /*
705 * The only case where we don't expand "vma"
706 * and we expand "next" instead is case 8.
707 */
708 VM_WARN_ON(end != next->vm_end);
709 /*
710 * remove_next == 3 means we're
711 * removing "vma" and that to do so we
712 * swapped "vma" and "next".
713 */
714 remove_next = 3;
715 VM_WARN_ON(file != next->vm_file);
716 swap(vma, next);
717 } else {
718 VM_WARN_ON(expand != vma);
719 /*
720 * case 1, 6, 7, remove_next == 2 is case 6,
721 * remove_next == 1 is case 1 or 7.
722 */
723 remove_next = 1 + (end > next->vm_end);
724 VM_WARN_ON(remove_next == 2 &&
725 end != next->vm_next->vm_end);
726 VM_WARN_ON(remove_next == 1 &&
727 end != next->vm_end);
728 /* trim end to next, for case 6 first pass */
729 end = next->vm_end;
730 }
731
Linus Torvalds287d97a2010-04-10 15:22:30 -0700732 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 importer = vma;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700734
735 /*
736 * If next doesn't have anon_vma, import from vma after
737 * next, if the vma overlaps with it.
738 */
Andrea Arcangeli97a42cd2016-10-07 17:01:31 -0700739 if (remove_next == 2 && !next->anon_vma)
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700740 exporter = next->vm_next;
741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 } else if (end > next->vm_start) {
743 /*
744 * vma expands, overlapping part of the next:
745 * mprotect case 5 shifting the boundary up.
746 */
747 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700748 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 importer = vma;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700750 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 } else if (end < vma->vm_end) {
752 /*
753 * vma shrinks, and !insert tells it's not
754 * split_vma inserting another: so it must be
755 * mprotect case 4 shifting the boundary down.
756 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700757 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700758 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 importer = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700760 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Rik van Riel5beb4932010-03-05 13:42:07 -0800763 /*
764 * Easily overlooked: when mprotect shifts the boundary,
765 * make sure the expanding vma has anon_vma set if the
766 * shrinking vma had, to cover any anon pages imported.
767 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700768 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800769 int error;
770
Linus Torvalds287d97a2010-04-10 15:22:30 -0700771 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300772 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700773 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300774 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800775 }
776 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700777again:
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700778 vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
Kirill A. Shutemov37f9f552016-07-26 15:25:48 -0700779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 if (file) {
781 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800782 root = &mapping->i_mmap;
783 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530784
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800785 if (adjust_next)
786 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530787
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800788 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700791 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 * are visible to arm/parisc __flush_dcache_page
793 * throughout; but we cannot insert into address
794 * space until vma start or end is updated.
795 */
796 __vma_link_file(insert);
797 }
798 }
799
Michel Lespinassebf181b92012-10-08 16:31:39 -0700800 anon_vma = vma->anon_vma;
801 if (!anon_vma && adjust_next)
802 anon_vma = next->anon_vma;
803 if (anon_vma) {
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700804 VM_WARN_ON(adjust_next && next->anon_vma &&
805 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000806 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700807 anon_vma_interval_tree_pre_update_vma(vma);
808 if (adjust_next)
809 anon_vma_interval_tree_pre_update_vma(next);
810 }
Rik van Riel012f18002010-08-09 17:18:40 -0700811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 if (root) {
813 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700814 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700816 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 }
818
Michel Lespinassed3737182012-12-11 16:01:38 -0800819 if (start != vma->vm_start) {
820 vma->vm_start = start;
821 start_changed = true;
822 }
823 if (end != vma->vm_end) {
824 vma->vm_end = end;
825 end_changed = true;
826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 vma->vm_pgoff = pgoff;
828 if (adjust_next) {
829 next->vm_start += adjust_next << PAGE_SHIFT;
830 next->vm_pgoff += adjust_next;
831 }
832
833 if (root) {
834 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700835 vma_interval_tree_insert(next, root);
836 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 flush_dcache_mmap_unlock(mapping);
838 }
839
840 if (remove_next) {
841 /*
842 * vma_merge has merged next into vma, and needs
843 * us to remove next before dropping the locks.
844 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700845 if (remove_next != 3)
846 __vma_unlink_prev(mm, next, vma);
847 else
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700848 /*
849 * vma is not before next if they've been
850 * swapped.
851 *
852 * pre-swap() next->vm_start was reduced so
853 * tell validate_mm_rb to ignore pre-swap()
854 * "next" (which is stored in post-swap()
855 * "vma").
856 */
857 __vma_unlink_common(mm, next, NULL, false, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 if (file)
859 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 } else if (insert) {
861 /*
862 * split_vma has split insert from vma, and needs
863 * us to insert it before dropping the locks
864 * (it may either follow vma or precede it).
865 */
866 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800867 } else {
868 if (start_changed)
869 vma_gap_update(vma);
870 if (end_changed) {
871 if (!next)
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700872 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800873 else if (!adjust_next)
874 vma_gap_update(next);
875 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 }
877
Michel Lespinassebf181b92012-10-08 16:31:39 -0700878 if (anon_vma) {
879 anon_vma_interval_tree_post_update_vma(vma);
880 if (adjust_next)
881 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800882 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800885 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530887 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100888 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530889
890 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100891 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530892 }
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700895 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530896 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700898 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800899 if (next->anon_vma)
900 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700902 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 kmem_cache_free(vm_area_cachep, next);
904 /*
905 * In mprotect's case 6 (see comments on vma_merge),
906 * we must remove another next too. It would clutter
907 * up the code too much to do both in one go.
908 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700909 if (remove_next != 3) {
910 /*
911 * If "next" was removed and vma->vm_end was
912 * expanded (up) over it, in turn
913 * "next->vm_prev->vm_end" changed and the
914 * "vma->vm_next" gap must be updated.
915 */
916 next = vma->vm_next;
917 } else {
918 /*
919 * For the scope of the comment "next" and
920 * "vma" considered pre-swap(): if "vma" was
921 * removed, next->vm_start was expanded (down)
922 * over it and the "next" gap must be updated.
923 * Because of the swap() the post-swap() "vma"
924 * actually points to pre-swap() "next"
925 * (post-swap() "next" as opposed is now a
926 * dangling pointer).
927 */
928 next = vma;
929 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700930 if (remove_next == 2) {
931 remove_next = 1;
932 end = next->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 goto again;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700934 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800935 else if (next)
936 vma_gap_update(next);
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700937 else {
938 /*
939 * If remove_next == 2 we obviously can't
940 * reach this path.
941 *
942 * If remove_next == 3 we can't reach this
943 * path because pre-swap() next is always not
944 * NULL. pre-swap() "next" is not being
945 * removed and its next->vm_end is not altered
946 * (and furthermore "end" already matches
947 * next->vm_end in remove_next == 3).
948 *
949 * We reach this only in the remove_next == 1
950 * case if the "next" vma that was removed was
951 * the highest vma of the mm. However in such
952 * case next->vm_end == "end" and the extended
953 * "vma" has vma->vm_end == next->vm_end so
954 * mm->highest_vm_end doesn't need any update
955 * in remove_next == 1 case.
956 */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700957 VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530960 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100961 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800964
965 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966}
967
968/*
969 * If the vma has a ->close operation then the driver probably needs to release
970 * per-vma resources, so we don't attempt to merge those.
971 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700973 struct file *file, unsigned long vm_flags,
Colin Cross3e4578f2015-10-27 16:42:08 -0700974 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
975 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800977 /*
978 * VM_SOFTDIRTY should not prevent from VMA merging, if we
979 * match the flags but dirty bit -- the caller should mark
980 * merged VMA as dirty. If dirty bit won't be excluded from
981 * comparison, we increase pressue on the memory system forcing
982 * the kernel to generate new VMAs when old one could be
983 * extended instead.
984 */
985 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 return 0;
987 if (vma->vm_file != file)
988 return 0;
989 if (vma->vm_ops && vma->vm_ops->close)
990 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700991 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
992 return 0;
Colin Cross3e4578f2015-10-27 16:42:08 -0700993 if (vma_get_anon_name(vma) != anon_name)
994 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 return 1;
996}
997
998static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700999 struct anon_vma *anon_vma2,
1000 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001{
Shaohua Li965f55d2011-05-24 17:11:20 -07001002 /*
1003 * The list_is_singular() test is to avoid merging VMA cloned from
1004 * parents. This can improve scalability caused by anon_vma lock.
1005 */
1006 if ((!anon_vma1 || !anon_vma2) && (!vma ||
1007 list_is_singular(&vma->anon_vma_chain)))
1008 return 1;
1009 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010}
1011
1012/*
1013 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1014 * in front of (at a lower virtual address and file offset than) the vma.
1015 *
1016 * We cannot merge two vmas if they have differently assigned (non-NULL)
1017 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1018 *
1019 * We don't check here for the merged mmap wrapping around the end of pagecache
1020 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1021 * wrap, nor mmaps which cover the final page at index -1UL.
1022 */
1023static int
1024can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001025 struct anon_vma *anon_vma, struct file *file,
1026 pgoff_t vm_pgoff,
Colin Cross3e4578f2015-10-27 16:42:08 -07001027 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1028 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029{
Colin Cross3e4578f2015-10-27 16:42:08 -07001030 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001031 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (vma->vm_pgoff == vm_pgoff)
1033 return 1;
1034 }
1035 return 0;
1036}
1037
1038/*
1039 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1040 * beyond (at a higher virtual address and file offset than) the vma.
1041 *
1042 * We cannot merge two vmas if they have differently assigned (non-NULL)
1043 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1044 */
1045static int
1046can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001047 struct anon_vma *anon_vma, struct file *file,
1048 pgoff_t vm_pgoff,
Colin Cross3e4578f2015-10-27 16:42:08 -07001049 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1050 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051{
Colin Cross3e4578f2015-10-27 16:42:08 -07001052 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001053 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001055 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1057 return 1;
1058 }
1059 return 0;
1060}
1061
1062/*
Colin Cross3e4578f2015-10-27 16:42:08 -07001063 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
1064 * figure out whether that can be merged with its predecessor or its
1065 * successor. Or both (it neatly fills a hole).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 *
1067 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1068 * certain not to be mapped by the time vma_merge is called; but when
1069 * called for mprotect, it is certain to be already mapped (either at
1070 * an offset within prev, or at the start of next), and the flags of
1071 * this area are about to be changed to vm_flags - and the no-change
1072 * case has already been eliminated.
1073 *
1074 * The following mprotect cases have to be considered, where AAAA is
1075 * the area passed down from mprotect_fixup, never extending beyond one
1076 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1077 *
1078 * AAAA AAAA AAAA AAAA
1079 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1080 * cannot merge might become might become might become
1081 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1082 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001083 * mremap move: PPPPXXXXXXXX 8
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 * AAAA
1085 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1086 * might become case 1 below case 2 below case 3 below
1087 *
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001088 * It is important for case 8 that the the vma NNNN overlapping the
1089 * region AAAA is never going to extended over XXXX. Instead XXXX must
1090 * be extended in region AAAA and NNNN must be removed. This way in
1091 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1092 * rmap_locks, the properties of the merged vma will be already
1093 * correct for the whole merged range. Some of those properties like
1094 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1095 * be correct for the whole merged range immediately after the
1096 * rmap_locks are released. Otherwise if XXXX would be removed and
1097 * NNNN would be extended over the XXXX range, remove_migration_ptes
1098 * or other rmap walkers (if working on addresses beyond the "end"
1099 * parameter) may establish ptes with the wrong permissions of NNNN
1100 * instead of the right permissions of XXXX.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 */
1102struct vm_area_struct *vma_merge(struct mm_struct *mm,
1103 struct vm_area_struct *prev, unsigned long addr,
1104 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001105 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001106 pgoff_t pgoff, struct mempolicy *policy,
Colin Cross3e4578f2015-10-27 16:42:08 -07001107 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1108 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
1110 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1111 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001112 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 /*
1115 * We later require that vma->vm_flags == vm_flags,
1116 * so this tests vma->vm_flags & VM_SPECIAL, too.
1117 */
1118 if (vm_flags & VM_SPECIAL)
1119 return NULL;
1120
1121 if (prev)
1122 next = prev->vm_next;
1123 else
1124 next = mm->mmap;
1125 area = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001126 if (area && area->vm_end == end) /* cases 6, 7, 8 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 next = next->vm_next;
1128
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001129 /* verify some invariant that must be enforced by the caller */
1130 VM_WARN_ON(prev && addr <= prev->vm_start);
1131 VM_WARN_ON(area && end > area->vm_end);
1132 VM_WARN_ON(addr >= end);
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 /*
1135 * Can it merge with the predecessor?
1136 */
1137 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001138 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001140 anon_vma, file, pgoff,
Colin Cross3e4578f2015-10-27 16:42:08 -07001141 vm_userfaultfd_ctx,
1142 anon_name)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 /*
1144 * OK, it can. Can we now merge in the successor as well?
1145 */
1146 if (next && end == next->vm_start &&
1147 mpol_equal(policy, vma_policy(next)) &&
1148 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001149 anon_vma, file,
1150 pgoff+pglen,
Colin Cross3e4578f2015-10-27 16:42:08 -07001151 vm_userfaultfd_ctx,
1152 anon_name) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001154 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 /* cases 1, 6 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001156 err = __vma_adjust(prev, prev->vm_start,
1157 next->vm_end, prev->vm_pgoff, NULL,
1158 prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 } else /* cases 2, 5, 7 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001160 err = __vma_adjust(prev, prev->vm_start,
1161 end, prev->vm_pgoff, NULL, prev);
Rik van Riel5beb4932010-03-05 13:42:07 -08001162 if (err)
1163 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001164 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 return prev;
1166 }
1167
1168 /*
1169 * Can this new request be merged in front of next?
1170 */
1171 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001172 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001174 anon_vma, file, pgoff+pglen,
Colin Cross3e4578f2015-10-27 16:42:08 -07001175 vm_userfaultfd_ctx,
1176 anon_name)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (prev && addr < prev->vm_end) /* case 4 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001178 err = __vma_adjust(prev, prev->vm_start,
1179 addr, prev->vm_pgoff, NULL, next);
1180 else { /* cases 3, 8 */
1181 err = __vma_adjust(area, addr, next->vm_end,
1182 next->vm_pgoff - pglen, NULL, next);
1183 /*
1184 * In case 3 area is already equal to next and
1185 * this is a noop, but in case 8 "area" has
1186 * been removed and next was expanded over it.
1187 */
1188 area = next;
1189 }
Rik van Riel5beb4932010-03-05 13:42:07 -08001190 if (err)
1191 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001192 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 return area;
1194 }
1195
1196 return NULL;
1197}
1198
1199/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001200 * Rough compatbility check to quickly see if it's even worth looking
1201 * at sharing an anon_vma.
1202 *
1203 * They need to have the same vm_file, and the flags can only differ
1204 * in things that mprotect may change.
1205 *
1206 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1207 * we can merge the two vma's. For example, we refuse to merge a vma if
1208 * there is a vm_ops->close() function, because that indicates that the
1209 * driver is doing some kind of reference counting. But that doesn't
1210 * really matter for the anon_vma sharing case.
1211 */
1212static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1213{
1214 return a->vm_end == b->vm_start &&
1215 mpol_equal(vma_policy(a), vma_policy(b)) &&
1216 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001217 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001218 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1219}
1220
1221/*
1222 * Do some basic sanity checking to see if we can re-use the anon_vma
1223 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1224 * the same as 'old', the other will be the new one that is trying
1225 * to share the anon_vma.
1226 *
1227 * NOTE! This runs with mm_sem held for reading, so it is possible that
1228 * the anon_vma of 'old' is concurrently in the process of being set up
1229 * by another page fault trying to merge _that_. But that's ok: if it
1230 * is being set up, that automatically means that it will be a singleton
1231 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001232 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001233 *
1234 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1235 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1236 * is to return an anon_vma that is "complex" due to having gone through
1237 * a fork).
1238 *
1239 * We also make sure that the two vma's are compatible (adjacent,
1240 * and with the same memory policies). That's all stable, even with just
1241 * a read lock on the mm_sem.
1242 */
1243static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1244{
1245 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001246 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001247
1248 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1249 return anon_vma;
1250 }
1251 return NULL;
1252}
1253
1254/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1256 * neighbouring vmas for a suitable anon_vma, before it goes off
1257 * to allocate a new anon_vma. It checks because a repetitive
1258 * sequence of mprotects and faults may otherwise lead to distinct
1259 * anon_vmas being allocated, preventing vma merge in subsequent
1260 * mprotect.
1261 */
1262struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1263{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001264 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
1267 near = vma->vm_next;
1268 if (!near)
1269 goto try_prev;
1270
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001271 anon_vma = reusable_anon_vma(near, vma, near);
1272 if (anon_vma)
1273 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001275 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (!near)
1277 goto none;
1278
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001279 anon_vma = reusable_anon_vma(near, near, vma);
1280 if (anon_vma)
1281 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282none:
1283 /*
1284 * There's no absolute need to look only at touching neighbours:
1285 * we could search further afield for "compatible" anon_vmas.
1286 * But it would probably just be a waste of time searching,
1287 * or lead to too many vmas hanging off the same anon_vma.
1288 * We're trying to allow mprotect remerging later on,
1289 * not trying to minimize memory used for anon_vmas.
1290 */
1291 return NULL;
1292}
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294/*
Al Viro40401532012-02-13 03:58:52 +00001295 * If a hint addr is less than mmap_min_addr change hint to be as
1296 * low as possible but still greater than mmap_min_addr
1297 */
1298static inline unsigned long round_hint_to_min(unsigned long hint)
1299{
1300 hint &= PAGE_MASK;
1301 if (((void *)hint != NULL) &&
1302 (hint < mmap_min_addr))
1303 return PAGE_ALIGN(mmap_min_addr);
1304 return hint;
1305}
1306
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001307static inline int mlock_future_check(struct mm_struct *mm,
1308 unsigned long flags,
1309 unsigned long len)
1310{
1311 unsigned long locked, lock_limit;
1312
1313 /* mlock MCL_FUTURE? */
1314 if (flags & VM_LOCKED) {
1315 locked = len >> PAGE_SHIFT;
1316 locked += mm->locked_vm;
1317 lock_limit = rlimit(RLIMIT_MEMLOCK);
1318 lock_limit >>= PAGE_SHIFT;
1319 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1320 return -EAGAIN;
1321 }
1322 return 0;
1323}
1324
Al Viro40401532012-02-13 03:58:52 +00001325/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001326 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001328unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001330 unsigned long flags, vm_flags_t vm_flags,
1331 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332{
vishnu.pscc71aba2014-10-09 15:26:29 -07001333 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001334 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Michel Lespinasse41badc12013-02-22 16:32:47 -08001336 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001337
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001338 if (!len)
1339 return -EINVAL;
1340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 /*
1342 * Does the application expect PROT_READ to imply PROT_EXEC?
1343 *
1344 * (the exception is when the underlying filesystem is noexec
1345 * mounted, in which case we dont add PROT_EXEC.)
1346 */
1347 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001348 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 prot |= PROT_EXEC;
1350
Eric Paris7cd94142007-11-26 18:47:40 -05001351 if (!(flags & MAP_FIXED))
1352 addr = round_hint_to_min(addr);
1353
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 /* Careful about overflows.. */
1355 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001356 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 return -ENOMEM;
1358
1359 /* offset overflow? */
1360 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001361 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
1363 /* Too many mappings? */
1364 if (mm->map_count > sysctl_max_map_count)
1365 return -ENOMEM;
1366
1367 /* Obtain the address to map to. we verify (or select) it and ensure
1368 * that it represents a valid section of the address space.
1369 */
1370 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001371 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return addr;
1373
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001374 if (prot == PROT_EXEC) {
1375 pkey = execute_only_pkey(mm);
1376 if (pkey < 0)
1377 pkey = 0;
1378 }
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 /* Do simple checking here so the lower-level routines won't have
1381 * to. we assume access permissions have been handled by the open
1382 * of the memory object, so we don't do any here.
1383 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001384 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1386
Huang Shijiecdf7b342009-09-21 17:03:36 -07001387 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (!can_do_mlock())
1389 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001390
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001391 if (mlock_future_check(mm, vm_flags, len))
1392 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001395 struct inode *inode = file_inode(file);
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 switch (flags & MAP_TYPE) {
1398 case MAP_SHARED:
1399 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1400 return -EACCES;
1401
1402 /*
1403 * Make sure we don't allow writing to an append-only
1404 * file..
1405 */
1406 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1407 return -EACCES;
1408
1409 /*
1410 * Make sure there are no mandatory locks on the file.
1411 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001412 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return -EAGAIN;
1414
1415 vm_flags |= VM_SHARED | VM_MAYSHARE;
1416 if (!(file->f_mode & FMODE_WRITE))
1417 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1418
1419 /* fall through */
1420 case MAP_PRIVATE:
1421 if (!(file->f_mode & FMODE_READ))
1422 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001423 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001424 if (vm_flags & VM_EXEC)
1425 return -EPERM;
1426 vm_flags &= ~VM_MAYEXEC;
1427 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001428
Al Viro72c2d532013-09-22 16:27:52 -04001429 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001430 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001431 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1432 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 break;
1434
1435 default:
1436 return -EINVAL;
1437 }
1438 } else {
1439 switch (flags & MAP_TYPE) {
1440 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001441 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1442 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001443 /*
1444 * Ignore pgoff.
1445 */
1446 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 vm_flags |= VM_SHARED | VM_MAYSHARE;
1448 break;
1449 case MAP_PRIVATE:
1450 /*
1451 * Set pgoff according to addr for anon_vma.
1452 */
1453 pgoff = addr >> PAGE_SHIFT;
1454 break;
1455 default:
1456 return -EINVAL;
1457 }
1458 }
1459
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001460 /*
1461 * Set 'VM_NORESERVE' if we should not account for the
1462 * memory use of this mapping.
1463 */
1464 if (flags & MAP_NORESERVE) {
1465 /* We honor MAP_NORESERVE if allowed to overcommit */
1466 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1467 vm_flags |= VM_NORESERVE;
1468
1469 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1470 if (file && is_file_hugepages(file))
1471 vm_flags |= VM_NORESERVE;
1472 }
1473
1474 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001475 if (!IS_ERR_VALUE(addr) &&
1476 ((vm_flags & VM_LOCKED) ||
1477 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001478 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001479 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001480}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001481
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001482SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1483 unsigned long, prot, unsigned long, flags,
1484 unsigned long, fd, unsigned long, pgoff)
1485{
1486 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001487 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001488
1489 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001490 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001491 file = fget(fd);
1492 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001493 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001494 if (is_file_hugepages(file))
1495 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001496 retval = -EINVAL;
1497 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1498 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001499 } else if (flags & MAP_HUGETLB) {
1500 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001501 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001502
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001503 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001504 if (!hs)
1505 return -EINVAL;
1506
1507 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001508 /*
1509 * VM_NORESERVE is used because the reservations will be
1510 * taken when vm_ops->mmap() is called
1511 * A dummy user value is used because we are not locking
1512 * memory so no accounting is necessary
1513 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001514 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001515 VM_NORESERVE,
1516 &user, HUGETLB_ANONHUGE_INODE,
1517 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001518 if (IS_ERR(file))
1519 return PTR_ERR(file);
1520 }
1521
1522 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1523
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001524 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001525out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001526 if (file)
1527 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001528 return retval;
1529}
1530
Christoph Hellwiga4679372010-03-10 15:21:15 -08001531#ifdef __ARCH_WANT_SYS_OLD_MMAP
1532struct mmap_arg_struct {
1533 unsigned long addr;
1534 unsigned long len;
1535 unsigned long prot;
1536 unsigned long flags;
1537 unsigned long fd;
1538 unsigned long offset;
1539};
1540
1541SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1542{
1543 struct mmap_arg_struct a;
1544
1545 if (copy_from_user(&a, arg, sizeof(a)))
1546 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001547 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001548 return -EINVAL;
1549
1550 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1551 a.offset >> PAGE_SHIFT);
1552}
1553#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1554
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001555/*
1556 * Some shared mappigns will want the pages marked read-only
1557 * to track write events. If so, we'll downgrade vm_page_prot
1558 * to the private version (using protection_map[] without the
1559 * VM_SHARED bit).
1560 */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001561int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001562{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001563 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001564 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001565
1566 /* If it was private or non-writable, the write bit is already clear */
1567 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1568 return 0;
1569
1570 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001571 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001572 return 1;
1573
Peter Feiner64e45502014-10-13 15:55:46 -07001574 /* The open routine did something to the protections that pgprot_modify
1575 * won't preserve? */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001576 if (pgprot_val(vm_page_prot) !=
1577 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001578 return 0;
1579
Peter Feiner64e45502014-10-13 15:55:46 -07001580 /* Do we need to track softdirty? */
1581 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1582 return 1;
1583
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001584 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001585 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001586 return 0;
1587
1588 /* Can the mapping track the dirty pages? */
1589 return vma->vm_file && vma->vm_file->f_mapping &&
1590 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1591}
1592
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001593/*
1594 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001595 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001596 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001597static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001598{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001599 /*
1600 * hugetlb has its own accounting separate from the core VM
1601 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1602 */
1603 if (file && is_file_hugepages(file))
1604 return 0;
1605
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001606 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1607}
1608
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001609unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001610 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001611{
1612 struct mm_struct *mm = current->mm;
1613 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001614 int error;
1615 struct rb_node **rb_link, *rb_parent;
1616 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001617
Cyril Hrubise8420a82013-04-29 15:08:33 -07001618 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001619 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001620 unsigned long nr_pages;
1621
1622 /*
1623 * MAP_FIXED may remove pages of mappings that intersects with
1624 * requested mapping. Account for the pages it would unmap.
1625 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001626 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1627
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001628 if (!may_expand_vm(mm, vm_flags,
1629 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001630 return -ENOMEM;
1631 }
1632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001634 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1635 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (do_munmap(mm, addr, len))
1637 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001640 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001641 * Private writable mapping: check memory availability
1642 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001643 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001644 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001645 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001646 return -ENOMEM;
1647 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
1649
1650 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001651 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001653 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
Colin Cross3e4578f2015-10-27 16:42:08 -07001654 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001655 if (vma)
1656 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
1658 /*
1659 * Determine the object being mapped and call the appropriate
1660 * specific mapper. the address has already been validated, but
1661 * not unmapped, but the maps are removed from the list.
1662 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001663 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 if (!vma) {
1665 error = -ENOMEM;
1666 goto unacct_error;
1667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 vma->vm_mm = mm;
1670 vma->vm_start = addr;
1671 vma->vm_end = addr + len;
1672 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001673 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001675 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
1677 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 if (vm_flags & VM_DENYWRITE) {
1679 error = deny_write_access(file);
1680 if (error)
1681 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001683 if (vm_flags & VM_SHARED) {
1684 error = mapping_map_writable(file->f_mapping);
1685 if (error)
1686 goto allow_write_and_free_vma;
1687 }
1688
1689 /* ->mmap() can change vma->vm_file, but must guarantee that
1690 * vma_link() below can deny write-access if VM_DENYWRITE is set
1691 * and map writably if VM_SHARED is set. This usually means the
1692 * new file must not have been exposed to user-space, yet.
1693 */
Al Virocb0942b2012-08-27 14:48:26 -04001694 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 error = file->f_op->mmap(file, vma);
1696 if (error)
1697 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001698
1699 /* Can addr have changed??
1700 *
1701 * Answer: Yes, several device drivers can do it in their
1702 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001703 * Bug: If addr is changed, prev, rb_link, rb_parent should
1704 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001705 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001706 WARN_ON_ONCE(addr != vma->vm_start);
1707
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001708 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001709 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 } else if (vm_flags & VM_SHARED) {
1711 error = shmem_zero_setup(vma);
1712 if (error)
1713 goto free_vma;
1714 }
1715
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001716 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001717 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001718 if (file) {
1719 if (vm_flags & VM_SHARED)
1720 mapping_unmap_writable(file->f_mapping);
1721 if (vm_flags & VM_DENYWRITE)
1722 allow_write_access(file);
1723 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001724 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001725out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001726 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001727
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001728 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001730 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1731 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001732 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001733 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001734 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001735 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301736
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001737 if (file)
1738 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301739
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001740 /*
1741 * New (or expanded) vma always get soft dirty status.
1742 * Otherwise user-space soft-dirty page tracker won't
1743 * be able to distinguish situation when vma area unmapped,
1744 * then new mapped in-place (which must be aimed as
1745 * a completely new data area).
1746 */
1747 vma->vm_flags |= VM_SOFTDIRTY;
1748
Peter Feiner64e45502014-10-13 15:55:46 -07001749 vma_set_page_prot(vma);
1750
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 return addr;
1752
1753unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 vma->vm_file = NULL;
1755 fput(file);
1756
1757 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001758 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1759 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001760 if (vm_flags & VM_SHARED)
1761 mapping_unmap_writable(file->f_mapping);
1762allow_write_and_free_vma:
1763 if (vm_flags & VM_DENYWRITE)
1764 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765free_vma:
1766 kmem_cache_free(vm_area_cachep, vma);
1767unacct_error:
1768 if (charged)
1769 vm_unacct_memory(charged);
1770 return error;
1771}
1772
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001773unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1774{
1775 /*
1776 * We implement the search by looking for an rbtree node that
1777 * immediately follows a suitable gap. That is,
1778 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1779 * - gap_end = vma->vm_start >= info->low_limit + length;
1780 * - gap_end - gap_start >= length
1781 */
1782
1783 struct mm_struct *mm = current->mm;
1784 struct vm_area_struct *vma;
1785 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1786
1787 /* Adjust search length to account for worst case alignment overhead */
1788 length = info->length + info->align_mask;
1789 if (length < info->length)
1790 return -ENOMEM;
1791
1792 /* Adjust search limits by the desired length */
1793 if (info->high_limit < length)
1794 return -ENOMEM;
1795 high_limit = info->high_limit - length;
1796
1797 if (info->low_limit > high_limit)
1798 return -ENOMEM;
1799 low_limit = info->low_limit + length;
1800
1801 /* Check if rbtree root looks promising */
1802 if (RB_EMPTY_ROOT(&mm->mm_rb))
1803 goto check_highest;
1804 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1805 if (vma->rb_subtree_gap < length)
1806 goto check_highest;
1807
1808 while (true) {
1809 /* Visit left subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001810 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001811 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1812 struct vm_area_struct *left =
1813 rb_entry(vma->vm_rb.rb_left,
1814 struct vm_area_struct, vm_rb);
1815 if (left->rb_subtree_gap >= length) {
1816 vma = left;
1817 continue;
1818 }
1819 }
1820
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001821 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001822check_current:
1823 /* Check if current node has a suitable gap */
1824 if (gap_start > high_limit)
1825 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001826 if (gap_end >= low_limit &&
1827 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001828 goto found;
1829
1830 /* Visit right subtree if it looks promising */
1831 if (vma->vm_rb.rb_right) {
1832 struct vm_area_struct *right =
1833 rb_entry(vma->vm_rb.rb_right,
1834 struct vm_area_struct, vm_rb);
1835 if (right->rb_subtree_gap >= length) {
1836 vma = right;
1837 continue;
1838 }
1839 }
1840
1841 /* Go back up the rbtree to find next candidate node */
1842 while (true) {
1843 struct rb_node *prev = &vma->vm_rb;
1844 if (!rb_parent(prev))
1845 goto check_highest;
1846 vma = rb_entry(rb_parent(prev),
1847 struct vm_area_struct, vm_rb);
1848 if (prev == vma->vm_rb.rb_left) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001849 gap_start = vm_end_gap(vma->vm_prev);
1850 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001851 goto check_current;
1852 }
1853 }
1854 }
1855
1856check_highest:
1857 /* Check highest gap, which does not precede any rbtree node */
1858 gap_start = mm->highest_vm_end;
1859 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1860 if (gap_start > high_limit)
1861 return -ENOMEM;
1862
1863found:
1864 /* We found a suitable gap. Clip it with the original low_limit. */
1865 if (gap_start < info->low_limit)
1866 gap_start = info->low_limit;
1867
1868 /* Adjust gap address to the desired alignment */
1869 gap_start += (info->align_offset - gap_start) & info->align_mask;
1870
1871 VM_BUG_ON(gap_start + info->length > info->high_limit);
1872 VM_BUG_ON(gap_start + info->length > gap_end);
1873 return gap_start;
1874}
1875
1876unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1877{
1878 struct mm_struct *mm = current->mm;
1879 struct vm_area_struct *vma;
1880 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1881
1882 /* Adjust search length to account for worst case alignment overhead */
1883 length = info->length + info->align_mask;
1884 if (length < info->length)
1885 return -ENOMEM;
1886
1887 /*
1888 * Adjust search limits by the desired length.
1889 * See implementation comment at top of unmapped_area().
1890 */
1891 gap_end = info->high_limit;
1892 if (gap_end < length)
1893 return -ENOMEM;
1894 high_limit = gap_end - length;
1895
1896 if (info->low_limit > high_limit)
1897 return -ENOMEM;
1898 low_limit = info->low_limit + length;
1899
1900 /* Check highest gap, which does not precede any rbtree node */
1901 gap_start = mm->highest_vm_end;
1902 if (gap_start <= high_limit)
1903 goto found_highest;
1904
1905 /* Check if rbtree root looks promising */
1906 if (RB_EMPTY_ROOT(&mm->mm_rb))
1907 return -ENOMEM;
1908 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1909 if (vma->rb_subtree_gap < length)
1910 return -ENOMEM;
1911
1912 while (true) {
1913 /* Visit right subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001914 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001915 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1916 struct vm_area_struct *right =
1917 rb_entry(vma->vm_rb.rb_right,
1918 struct vm_area_struct, vm_rb);
1919 if (right->rb_subtree_gap >= length) {
1920 vma = right;
1921 continue;
1922 }
1923 }
1924
1925check_current:
1926 /* Check if current node has a suitable gap */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001927 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001928 if (gap_end < low_limit)
1929 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001930 if (gap_start <= high_limit &&
1931 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001932 goto found;
1933
1934 /* Visit left subtree if it looks promising */
1935 if (vma->vm_rb.rb_left) {
1936 struct vm_area_struct *left =
1937 rb_entry(vma->vm_rb.rb_left,
1938 struct vm_area_struct, vm_rb);
1939 if (left->rb_subtree_gap >= length) {
1940 vma = left;
1941 continue;
1942 }
1943 }
1944
1945 /* Go back up the rbtree to find next candidate node */
1946 while (true) {
1947 struct rb_node *prev = &vma->vm_rb;
1948 if (!rb_parent(prev))
1949 return -ENOMEM;
1950 vma = rb_entry(rb_parent(prev),
1951 struct vm_area_struct, vm_rb);
1952 if (prev == vma->vm_rb.rb_right) {
1953 gap_start = vma->vm_prev ?
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001954 vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001955 goto check_current;
1956 }
1957 }
1958 }
1959
1960found:
1961 /* We found a suitable gap. Clip it with the original high_limit. */
1962 if (gap_end > info->high_limit)
1963 gap_end = info->high_limit;
1964
1965found_highest:
1966 /* Compute highest gap address at the desired alignment */
1967 gap_end -= info->length;
1968 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1969
1970 VM_BUG_ON(gap_end < info->low_limit);
1971 VM_BUG_ON(gap_end < gap_start);
1972 return gap_end;
1973}
1974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975/* Get an address range which is currently unmapped.
1976 * For shmat() with addr=0.
1977 *
1978 * Ugly calling convention alert:
1979 * Return value with the low bits set means error value,
1980 * ie
1981 * if (ret & ~PAGE_MASK)
1982 * error = ret;
1983 *
1984 * This function "knows" that -ENOMEM has the bits set.
1985 */
1986#ifndef HAVE_ARCH_UNMAPPED_AREA
1987unsigned long
1988arch_get_unmapped_area(struct file *filp, unsigned long addr,
1989 unsigned long len, unsigned long pgoff, unsigned long flags)
1990{
1991 struct mm_struct *mm = current->mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001992 struct vm_area_struct *vma, *prev;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001993 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001995 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return -ENOMEM;
1997
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001998 if (flags & MAP_FIXED)
1999 return addr;
2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (addr) {
2002 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002003 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002004 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002005 (!vma || addr + len <= vm_start_gap(vma)) &&
2006 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 return addr;
2008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002010 info.flags = 0;
2011 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08002012 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002013 info.high_limit = TASK_SIZE;
2014 info.align_mask = 0;
2015 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016}
vishnu.pscc71aba2014-10-09 15:26:29 -07002017#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019/*
2020 * This mmap-allocator allocates new areas top-down from below the
2021 * stack's low limit (the base):
2022 */
2023#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2024unsigned long
2025arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
2026 const unsigned long len, const unsigned long pgoff,
2027 const unsigned long flags)
2028{
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002029 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002031 unsigned long addr = addr0;
2032 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
2034 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002035 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 return -ENOMEM;
2037
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002038 if (flags & MAP_FIXED)
2039 return addr;
2040
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 /* requesting a specific address */
2042 if (addr) {
2043 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002044 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002045 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002046 (!vma || addr + len <= vm_start_gap(vma)) &&
2047 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return addr;
2049 }
2050
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002051 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2052 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002053 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002054 info.high_limit = mm->mmap_base;
2055 info.align_mask = 0;
2056 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07002057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 /*
2059 * A failed mmap() very likely causes application failure,
2060 * so fall back to the bottom-up function here. This scenario
2061 * can happen with large stack limits and large mmap()
2062 * allocations.
2063 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002064 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002065 VM_BUG_ON(addr != -ENOMEM);
2066 info.flags = 0;
2067 info.low_limit = TASK_UNMAPPED_BASE;
2068 info.high_limit = TASK_SIZE;
2069 addr = vm_unmapped_area(&info);
2070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
2072 return addr;
2073}
2074#endif
2075
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076unsigned long
2077get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2078 unsigned long pgoff, unsigned long flags)
2079{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002080 unsigned long (*get_area)(struct file *, unsigned long,
2081 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
Al Viro9206de92009-12-03 15:23:11 -05002083 unsigned long error = arch_mmap_check(addr, len, flags);
2084 if (error)
2085 return error;
2086
2087 /* Careful about overflows.. */
2088 if (len > TASK_SIZE)
2089 return -ENOMEM;
2090
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002091 get_area = current->mm->get_unmapped_area;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002092 if (file) {
2093 if (file->f_op->get_unmapped_area)
2094 get_area = file->f_op->get_unmapped_area;
2095 } else if (flags & MAP_SHARED) {
2096 /*
2097 * mmap_region() will call shmem_zero_setup() to create a file,
2098 * so use shmem's get_unmapped_area in case it can be huge.
2099 * do_mmap_pgoff() will clear pgoff, so match alignment.
2100 */
2101 pgoff = 0;
2102 get_area = shmem_get_unmapped_area;
2103 }
2104
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002105 addr = get_area(file, addr, len, pgoff, flags);
2106 if (IS_ERR_VALUE(addr))
2107 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002109 if (addr > TASK_SIZE - len)
2110 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002111 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002112 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002113
Al Viro9ac4ed42012-05-30 17:13:15 -04002114 error = security_mmap_addr(addr);
2115 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116}
2117
2118EXPORT_SYMBOL(get_unmapped_area);
2119
2120/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002121struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002123 struct rb_node *rb_node;
2124 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002126 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002127 vma = vmacache_find(mm, addr);
2128 if (likely(vma))
2129 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002131 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002133 while (rb_node) {
2134 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002136 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002137
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002138 if (tmp->vm_end > addr) {
2139 vma = tmp;
2140 if (tmp->vm_start <= addr)
2141 break;
2142 rb_node = rb_node->rb_left;
2143 } else
2144 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002146
2147 if (vma)
2148 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 return vma;
2150}
2151
2152EXPORT_SYMBOL(find_vma);
2153
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002154/*
2155 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002156 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157struct vm_area_struct *
2158find_vma_prev(struct mm_struct *mm, unsigned long addr,
2159 struct vm_area_struct **pprev)
2160{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002161 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002163 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002164 if (vma) {
2165 *pprev = vma->vm_prev;
2166 } else {
2167 struct rb_node *rb_node = mm->mm_rb.rb_node;
2168 *pprev = NULL;
2169 while (rb_node) {
2170 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2171 rb_node = rb_node->rb_right;
2172 }
2173 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002174 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175}
2176
2177/*
2178 * Verify that the stack growth is acceptable and
2179 * update accounting. This is shared with both the
2180 * grow-up and grow-down cases.
2181 */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002182static int acct_stack_growth(struct vm_area_struct *vma,
2183 unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184{
2185 struct mm_struct *mm = vma->vm_mm;
2186 struct rlimit *rlim = current->signal->rlim;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002187 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188
2189 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002190 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 return -ENOMEM;
2192
2193 /* Stack limit test */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002194 if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 return -ENOMEM;
2196
2197 /* mlock limit tests */
2198 if (vma->vm_flags & VM_LOCKED) {
2199 unsigned long locked;
2200 unsigned long limit;
2201 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002202 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002203 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 if (locked > limit && !capable(CAP_IPC_LOCK))
2205 return -ENOMEM;
2206 }
2207
Adam Litke0d59a012007-01-30 14:35:39 -08002208 /* Check to ensure the stack will not grow into a hugetlb-only region */
2209 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2210 vma->vm_end - size;
2211 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2212 return -EFAULT;
2213
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 /*
2215 * Overcommit.. This must be the final test, as it will
2216 * update security statistics.
2217 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002218 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 return -ENOMEM;
2220
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 return 0;
2222}
2223
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002224#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002226 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2227 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002229int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230{
Oleg Nesterov09357812015-11-05 18:48:17 -08002231 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002232 struct vm_area_struct *next;
2233 unsigned long gap_addr;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002234 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235
2236 if (!(vma->vm_flags & VM_GROWSUP))
2237 return -EFAULT;
2238
Helge Deller5d10ad622017-06-19 17:34:05 +02002239 /* Guard against exceeding limits of the address space. */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002240 address &= PAGE_MASK;
Helge Deller38dfd2e2017-07-14 14:49:38 -07002241 if (address >= (TASK_SIZE & PAGE_MASK))
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002242 return -ENOMEM;
Helge Deller5d10ad622017-06-19 17:34:05 +02002243 address += PAGE_SIZE;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002244
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002245 /* Enforce stack_guard_gap */
2246 gap_addr = address + stack_guard_gap;
Helge Deller5d10ad622017-06-19 17:34:05 +02002247
2248 /* Guard against overflow */
2249 if (gap_addr < address || gap_addr > TASK_SIZE)
2250 gap_addr = TASK_SIZE;
2251
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002252 next = vma->vm_next;
Michal Hockoc57664b2017-07-10 15:49:51 -07002253 if (next && next->vm_start < gap_addr &&
2254 (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002255 if (!(next->vm_flags & VM_GROWSUP))
2256 return -ENOMEM;
2257 /* Check that both stack segments have the same anon_vma? */
2258 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002259
2260 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 if (unlikely(anon_vma_prepare(vma)))
2262 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263
2264 /*
2265 * vma->vm_start/vm_end cannot change under us because the caller
2266 * is required to hold the mmap_sem in read mode. We need the
2267 * anon_vma lock to serialize against concurrent expand_stacks.
2268 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002269 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
2271 /* Somebody else might have raced and expanded it already */
2272 if (address > vma->vm_end) {
2273 unsigned long size, grow;
2274
2275 size = address - vma->vm_start;
2276 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2277
Hugh Dickins42c36f62011-05-09 17:44:42 -07002278 error = -ENOMEM;
2279 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2280 error = acct_stack_growth(vma, size, grow);
2281 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002282 /*
2283 * vma_gap_update() doesn't support concurrent
2284 * updates, but we only hold a shared mmap_sem
2285 * lock here, so we need to protect against
2286 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002287 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002288 * we don't guarantee that all growable vmas
2289 * in a mm share the same root anon vma.
2290 * So, we reuse mm->page_table_lock to guard
2291 * against concurrent vma expansions.
2292 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002293 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002294 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002295 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002296 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002297 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002298 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002299 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002300 if (vma->vm_next)
2301 vma_gap_update(vma->vm_next);
2302 else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002303 mm->highest_vm_end = vm_end_gap(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002304 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002305
Hugh Dickins42c36f62011-05-09 17:44:42 -07002306 perf_event_mmap(vma);
2307 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002310 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002311 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002312 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 return error;
2314}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002315#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2316
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317/*
2318 * vma is the first one with address < vma->vm_start. Have to extend vma.
2319 */
Michal Hockod05f3162011-05-24 17:11:44 -07002320int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002321 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322{
Oleg Nesterov09357812015-11-05 18:48:17 -08002323 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002324 struct vm_area_struct *prev;
2325 unsigned long gap_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 int error;
2327
Eric Paris88694772007-11-26 18:47:26 -05002328 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002329 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002330 if (error)
2331 return error;
2332
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002333 /* Enforce stack_guard_gap */
2334 gap_addr = address - stack_guard_gap;
2335 if (gap_addr > address)
2336 return -ENOMEM;
2337 prev = vma->vm_prev;
Michal Hockoc57664b2017-07-10 15:49:51 -07002338 if (prev && prev->vm_end > gap_addr &&
2339 (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002340 if (!(prev->vm_flags & VM_GROWSDOWN))
2341 return -ENOMEM;
2342 /* Check that both stack segments have the same anon_vma? */
2343 }
2344
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002345 /* We must make sure the anon_vma is allocated. */
2346 if (unlikely(anon_vma_prepare(vma)))
2347 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
2349 /*
2350 * vma->vm_start/vm_end cannot change under us because the caller
2351 * is required to hold the mmap_sem in read mode. We need the
2352 * anon_vma lock to serialize against concurrent expand_stacks.
2353 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002354 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355
2356 /* Somebody else might have raced and expanded it already */
2357 if (address < vma->vm_start) {
2358 unsigned long size, grow;
2359
2360 size = vma->vm_end - address;
2361 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2362
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002363 error = -ENOMEM;
2364 if (grow <= vma->vm_pgoff) {
2365 error = acct_stack_growth(vma, size, grow);
2366 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002367 /*
2368 * vma_gap_update() doesn't support concurrent
2369 * updates, but we only hold a shared mmap_sem
2370 * lock here, so we need to protect against
2371 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002372 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002373 * we don't guarantee that all growable vmas
2374 * in a mm share the same root anon vma.
2375 * So, we reuse mm->page_table_lock to guard
2376 * against concurrent vma expansions.
2377 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002378 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002379 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002380 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002381 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002382 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002383 vma->vm_start = address;
2384 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002385 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002386 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002387 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002388
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002389 perf_event_mmap(vma);
2390 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 }
2392 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002393 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002394 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002395 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 return error;
2397}
2398
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002399/* enforced gap between the expanding stack and other mappings. */
2400unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2401
2402static int __init cmdline_parse_stack_guard_gap(char *p)
2403{
2404 unsigned long val;
2405 char *endptr;
2406
2407 val = simple_strtoul(p, &endptr, 10);
2408 if (!*endptr)
2409 stack_guard_gap = val << PAGE_SHIFT;
2410
2411 return 0;
2412}
2413__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2414
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002415#ifdef CONFIG_STACK_GROWSUP
2416int expand_stack(struct vm_area_struct *vma, unsigned long address)
2417{
2418 return expand_upwards(vma, address);
2419}
2420
2421struct vm_area_struct *
2422find_extend_vma(struct mm_struct *mm, unsigned long addr)
2423{
2424 struct vm_area_struct *vma, *prev;
2425
2426 addr &= PAGE_MASK;
2427 vma = find_vma_prev(mm, addr, &prev);
2428 if (vma && (vma->vm_start <= addr))
2429 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002430 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002431 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002432 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002433 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002434 return prev;
2435}
2436#else
2437int expand_stack(struct vm_area_struct *vma, unsigned long address)
2438{
2439 return expand_downwards(vma, address);
2440}
2441
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002443find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444{
vishnu.pscc71aba2014-10-09 15:26:29 -07002445 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 unsigned long start;
2447
2448 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002449 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (!vma)
2451 return NULL;
2452 if (vma->vm_start <= addr)
2453 return vma;
2454 if (!(vma->vm_flags & VM_GROWSDOWN))
2455 return NULL;
2456 start = vma->vm_start;
2457 if (expand_stack(vma, addr))
2458 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002459 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002460 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 return vma;
2462}
2463#endif
2464
Jesse Barnese1d6d012014-12-12 16:55:27 -08002465EXPORT_SYMBOL_GPL(find_extend_vma);
2466
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002468 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002470 *
2471 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002473static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002475 unsigned long nr_accounted = 0;
2476
Hugh Dickins365e9c872005-10-29 18:16:18 -07002477 /* Update high watermark before we lower total_vm */
2478 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002480 long nrpages = vma_pages(vma);
2481
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002482 if (vma->vm_flags & VM_ACCOUNT)
2483 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002484 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002485 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002486 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002487 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 validate_mm(mm);
2489}
2490
2491/*
2492 * Get rid of page table information in the indicated region.
2493 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002494 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 */
2496static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002497 struct vm_area_struct *vma, struct vm_area_struct *prev,
2498 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499{
vishnu.pscc71aba2014-10-09 15:26:29 -07002500 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002501 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
2503 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002504 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002505 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002506 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002507 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002508 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002509 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510}
2511
2512/*
2513 * Create a list of vma's touched by the unmap, removing them from the mm's
2514 * vma list as we go..
2515 */
2516static void
2517detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2518 struct vm_area_struct *prev, unsigned long end)
2519{
2520 struct vm_area_struct **insertion_point;
2521 struct vm_area_struct *tail_vma = NULL;
2522
2523 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002524 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002526 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 mm->map_count--;
2528 tail_vma = vma;
2529 vma = vma->vm_next;
2530 } while (vma && vma->vm_start < end);
2531 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002532 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002533 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002534 vma_gap_update(vma);
2535 } else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002536 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002538
2539 /* Kill the cache */
2540 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541}
2542
2543/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002544 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2545 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002547static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 unsigned long addr, int new_below)
2549{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002551 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552
Dan Williamscebe1392017-11-29 16:10:28 -08002553 if (vma->vm_ops && vma->vm_ops->split) {
2554 err = vma->vm_ops->split(vma, addr);
2555 if (err)
2556 return err;
2557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558
Christoph Lametere94b1762006-12-06 20:33:17 -08002559 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002561 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
2563 /* most fields are the same, copy all, and then fixup */
2564 *new = *vma;
2565
Rik van Riel5beb4932010-03-05 13:42:07 -08002566 INIT_LIST_HEAD(&new->anon_vma_chain);
2567
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (new_below)
2569 new->vm_end = addr;
2570 else {
2571 new->vm_start = addr;
2572 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2573 }
2574
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002575 err = vma_dup_policy(vma, new);
2576 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002577 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002579 err = anon_vma_clone(new, vma);
2580 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002581 goto out_free_mpol;
2582
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002583 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 get_file(new->vm_file);
2585
2586 if (new->vm_ops && new->vm_ops->open)
2587 new->vm_ops->open(new);
2588
2589 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002590 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2592 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002593 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
Rik van Riel5beb4932010-03-05 13:42:07 -08002595 /* Success. */
2596 if (!err)
2597 return 0;
2598
2599 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002600 if (new->vm_ops && new->vm_ops->close)
2601 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002602 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002603 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002604 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002605 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002606 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002607 out_free_vma:
2608 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002609 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610}
2611
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002612/*
2613 * Split a vma into two pieces at address 'addr', a new vma is allocated
2614 * either for the first part or the tail.
2615 */
2616int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2617 unsigned long addr, int new_below)
2618{
2619 if (mm->map_count >= sysctl_max_map_count)
2620 return -ENOMEM;
2621
2622 return __split_vma(mm, vma, addr, new_below);
2623}
2624
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625/* Munmap is split into 2 main parts -- this part which finds
2626 * what needs doing, and the areas themselves, which do the
2627 * work. This now handles partial unmappings.
2628 * Jeremy Fitzhardinge <jeremy@goop.org>
2629 */
2630int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2631{
2632 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002633 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002635 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 return -EINVAL;
2637
vishnu.pscc71aba2014-10-09 15:26:29 -07002638 len = PAGE_ALIGN(len);
2639 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 return -EINVAL;
2641
2642 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002643 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002644 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002646 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002647 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
2649 /* if it doesn't overlap, we have nothing.. */
2650 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002651 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 return 0;
2653
2654 /*
2655 * If we need to split any vma, do it now to save pain later.
2656 *
2657 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2658 * unmapped vm_area_struct will remain in use: so lower split_vma
2659 * places tmp vma above, and higher split_vma places tmp vma below.
2660 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002661 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002662 int error;
2663
2664 /*
2665 * Make sure that map_count on return from munmap() will
2666 * not exceed its limit; but let map_count go just above
2667 * its limit temporarily, to help free resources as expected.
2668 */
2669 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2670 return -ENOMEM;
2671
2672 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 if (error)
2674 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002675 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 }
2677
2678 /* Does it split the last one? */
2679 last = find_vma(mm, end);
2680 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002681 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 if (error)
2683 return error;
2684 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002685 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686
2687 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002688 * unlock any mlock()ed ranges before detaching vmas
2689 */
2690 if (mm->locked_vm) {
2691 struct vm_area_struct *tmp = vma;
2692 while (tmp && tmp->vm_start < end) {
2693 if (tmp->vm_flags & VM_LOCKED) {
2694 mm->locked_vm -= vma_pages(tmp);
2695 munlock_vma_pages_all(tmp);
2696 }
2697 tmp = tmp->vm_next;
2698 }
2699 }
2700
2701 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 * Remove the vma's, and unmap the actual pages
2703 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002704 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2705 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Dave Hansen1de4fa12014-11-14 07:18:31 -08002707 arch_unmap(mm, vma, start, end);
2708
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002710 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
2712 return 0;
2713}
Guenter Roeckabfb4862016-03-24 10:39:14 -07002714EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715
Al Virobfce2812012-04-20 21:57:04 -04002716int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002717{
2718 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002719 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002720
Michal Hockoae798782016-05-23 16:25:33 -07002721 if (down_write_killable(&mm->mmap_sem))
2722 return -EINTR;
2723
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002724 ret = do_munmap(mm, start, len);
2725 up_write(&mm->mmap_sem);
2726 return ret;
2727}
2728EXPORT_SYMBOL(vm_munmap);
2729
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002730SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731{
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002732 int ret;
2733 struct mm_struct *mm = current->mm;
2734
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 profile_munmap(addr);
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002736 if (down_write_killable(&mm->mmap_sem))
2737 return -EINTR;
2738 ret = do_munmap(mm, addr, len);
2739 up_write(&mm->mmap_sem);
2740 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741}
2742
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002743
2744/*
2745 * Emulation of deprecated remap_file_pages() syscall.
2746 */
2747SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2748 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2749{
2750
2751 struct mm_struct *mm = current->mm;
2752 struct vm_area_struct *vma;
2753 unsigned long populate = 0;
2754 unsigned long ret = -EINVAL;
2755 struct file *file;
2756
Joe Perches756a025f02016-03-17 14:19:47 -07002757 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2758 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002759
2760 if (prot)
2761 return ret;
2762 start = start & PAGE_MASK;
2763 size = size & PAGE_MASK;
2764
2765 if (start + size <= start)
2766 return ret;
2767
2768 /* Does pgoff wrap? */
2769 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2770 return ret;
2771
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002772 if (down_write_killable(&mm->mmap_sem))
2773 return -EINTR;
2774
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002775 vma = find_vma(mm, start);
2776
2777 if (!vma || !(vma->vm_flags & VM_SHARED))
2778 goto out;
2779
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002780 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002781 goto out;
2782
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002783 if (start + size > vma->vm_end) {
2784 struct vm_area_struct *next;
2785
2786 for (next = vma->vm_next; next; next = next->vm_next) {
2787 /* hole between vmas ? */
2788 if (next->vm_start != next->vm_prev->vm_end)
2789 goto out;
2790
2791 if (next->vm_file != vma->vm_file)
2792 goto out;
2793
2794 if (next->vm_flags != vma->vm_flags)
2795 goto out;
2796
2797 if (start + size <= next->vm_end)
2798 break;
2799 }
2800
2801 if (!next)
2802 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002803 }
2804
2805 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2806 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2807 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2808
2809 flags &= MAP_NONBLOCK;
2810 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2811 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002812 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002813 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002814
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002815 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002816 for (tmp = vma; tmp->vm_start >= start + size;
2817 tmp = tmp->vm_next) {
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07002818 /*
2819 * Split pmd and munlock page on the border
2820 * of the range.
2821 */
2822 vma_adjust_trans_huge(tmp, start, start + size, 0);
2823
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002824 munlock_vma_pages_range(tmp,
2825 max(tmp->vm_start, start),
2826 min(tmp->vm_end, start + size));
2827 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002828 }
2829
2830 file = get_file(vma->vm_file);
2831 ret = do_mmap_pgoff(vma->vm_file, start, size,
2832 prot, flags, pgoff, &populate);
2833 fput(file);
2834out:
2835 up_write(&mm->mmap_sem);
2836 if (populate)
2837 mm_populate(ret, populate);
2838 if (!IS_ERR_VALUE(ret))
2839 ret = 0;
2840 return ret;
2841}
2842
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843static inline void verify_mm_writelocked(struct mm_struct *mm)
2844{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002845#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2847 WARN_ON(1);
2848 up_read(&mm->mmap_sem);
2849 }
2850#endif
2851}
2852
2853/*
2854 * this is really a simplified "do_mmap". it only handles
2855 * anonymous maps. eventually we may be able to do some
2856 * brk-specific accounting here.
2857 */
Kees Cookba093a62016-08-02 14:04:54 -07002858static int do_brk(unsigned long addr, unsigned long request)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859{
vishnu.pscc71aba2014-10-09 15:26:29 -07002860 struct mm_struct *mm = current->mm;
2861 struct vm_area_struct *vma, *prev;
Kees Cookba093a62016-08-02 14:04:54 -07002862 unsigned long flags, len;
vishnu.pscc71aba2014-10-09 15:26:29 -07002863 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002865 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866
Kees Cookba093a62016-08-02 14:04:54 -07002867 len = PAGE_ALIGN(request);
2868 if (len < request)
2869 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002871 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872
Kirill Korotaev3a459752006-09-07 14:17:04 +04002873 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2874
Al Viro2c6a1012009-12-03 19:40:46 -05002875 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002876 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002877 return error;
2878
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002879 error = mlock_future_check(mm, mm->def_flags, len);
2880 if (error)
2881 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882
2883 /*
2884 * mm->mmap_sem is required to protect against another thread
2885 * changing the mappings in case we sleep.
2886 */
2887 verify_mm_writelocked(mm);
2888
2889 /*
2890 * Clear old maps. this also does some error checking for us
2891 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002892 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2893 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 if (do_munmap(mm, addr, len))
2895 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 }
2897
2898 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002899 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 return -ENOMEM;
2901
2902 if (mm->map_count > sysctl_max_map_count)
2903 return -ENOMEM;
2904
Al Viro191c5422012-02-13 03:58:52 +00002905 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 return -ENOMEM;
2907
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002909 vma = vma_merge(mm, prev, addr, addr + len, flags,
Colin Cross3e4578f2015-10-27 16:42:08 -07002910 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
Rik van Rielba470de2008-10-18 20:26:50 -07002911 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 goto out;
2913
2914 /*
2915 * create a vma struct for an anonymous mapping
2916 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002917 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 if (!vma) {
2919 vm_unacct_memory(len >> PAGE_SHIFT);
2920 return -ENOMEM;
2921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
Rik van Riel5beb4932010-03-05 13:42:07 -08002923 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 vma->vm_mm = mm;
2925 vma->vm_start = addr;
2926 vma->vm_end = addr + len;
2927 vma->vm_pgoff = pgoff;
2928 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002929 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 vma_link(mm, vma, prev, rb_link, rb_parent);
2931out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002932 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002934 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002935 if (flags & VM_LOCKED)
2936 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002937 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002938 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939}
2940
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002941int vm_brk(unsigned long addr, unsigned long len)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002942{
2943 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002944 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002945 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002946
Michal Hocko2d6c9282016-05-23 16:25:42 -07002947 if (down_write_killable(&mm->mmap_sem))
2948 return -EINTR;
2949
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002950 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002951 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002952 up_write(&mm->mmap_sem);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002953 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002954 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002955 return ret;
2956}
2957EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958
2959/* Release all mmaps. */
2960void exit_mmap(struct mm_struct *mm)
2961{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002962 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002963 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 unsigned long nr_accounted = 0;
2965
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002966 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002967 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002968
Rik van Rielba470de2008-10-18 20:26:50 -07002969 if (mm->locked_vm) {
2970 vma = mm->mmap;
2971 while (vma) {
2972 if (vma->vm_flags & VM_LOCKED)
2973 munlock_vma_pages_all(vma);
2974 vma = vma->vm_next;
2975 }
2976 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002977
2978 arch_exit_mmap(mm);
2979
Rik van Rielba470de2008-10-18 20:26:50 -07002980 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002981 if (!vma) /* Can happen if dup_mmap() received an OOM */
2982 return;
2983
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002986 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002987 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002988 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002989 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002990
Michal Hocko065e76d2017-12-14 15:33:15 -08002991 if (unlikely(mm_is_oom_victim(mm))) {
Andrea Arcangeli07b039e92017-09-06 16:25:00 -07002992 /*
2993 * Wait for oom_reap_task() to stop working on this
2994 * mm. Because MMF_OOM_SKIP is already set before
2995 * calling down_read(), oom_reap_task() will not run
2996 * on this "mm" post up_write().
2997 *
Michal Hocko065e76d2017-12-14 15:33:15 -08002998 * mm_is_oom_victim() cannot be set from under us
2999 * either because victim->mm is already set to NULL
Andrea Arcangeli07b039e92017-09-06 16:25:00 -07003000 * under task_lock before calling mmput and oom_mm is
Michal Hocko065e76d2017-12-14 15:33:15 -08003001 * set not NULL by the OOM killer only if victim->mm
Andrea Arcangeli07b039e92017-09-06 16:25:00 -07003002 * is found not NULL while holding the task_lock.
3003 */
Michal Hocko065e76d2017-12-14 15:33:15 -08003004 set_bit(MMF_OOM_SKIP, &mm->flags);
Andrea Arcangeli07b039e92017-09-06 16:25:00 -07003005 down_write(&mm->mmap_sem);
3006 up_write(&mm->mmap_sem);
3007 }
Hugh Dickins6ee86302013-04-29 15:07:44 -07003008 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05003009 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07003012 * Walk the list again, actually closing and freeing it,
3013 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003015 while (vma) {
3016 if (vma->vm_flags & VM_ACCOUNT)
3017 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07003018 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003019 }
3020 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021}
3022
3023/* Insert vm structure into process list sorted by address
3024 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003025 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 */
Hugh Dickins6597d782012-10-08 16:29:07 -07003027int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028{
Hugh Dickins6597d782012-10-08 16:29:07 -07003029 struct vm_area_struct *prev;
3030 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031
Chen Gangc9d13f52015-09-08 15:04:08 -07003032 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3033 &prev, &rb_link, &rb_parent))
3034 return -ENOMEM;
3035 if ((vma->vm_flags & VM_ACCOUNT) &&
3036 security_vm_enough_memory_mm(mm, vma_pages(vma)))
3037 return -ENOMEM;
3038
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 /*
3040 * The vm_pgoff of a purely anonymous vma should be irrelevant
3041 * until its first write fault, when page's anon_vma and index
3042 * are set. But now set the vm_pgoff it will almost certainly
3043 * end up with (unless mremap moves it elsewhere before that
3044 * first wfault), so /proc/pid/maps tells a consistent story.
3045 *
3046 * By setting it to reflect the virtual start address of the
3047 * vma, merges and splits can happen in a seamless way, just
3048 * using the existing file pgoff checks and manipulations.
3049 * Similarly in do_mmap_pgoff and in do_brk.
3050 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003051 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 BUG_ON(vma->anon_vma);
3053 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3054 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05303055
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 vma_link(mm, vma, prev, rb_link, rb_parent);
3057 return 0;
3058}
3059
3060/*
3061 * Copy the vma structure to a new location in the same mm,
3062 * prior to moving page table entries, to effect an mremap move.
3063 */
3064struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07003065 unsigned long addr, unsigned long len, pgoff_t pgoff,
3066 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067{
3068 struct vm_area_struct *vma = *vmap;
3069 unsigned long vma_start = vma->vm_start;
3070 struct mm_struct *mm = vma->vm_mm;
3071 struct vm_area_struct *new_vma, *prev;
3072 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003073 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074
3075 /*
3076 * If anonymous vma has not yet been faulted, update new pgoff
3077 * to match new location, to increase its chance of merging.
3078 */
Oleg Nesterovce757992015-09-08 14:58:34 -07003079 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003081 faulted_in_anon_vma = false;
3082 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
Hugh Dickins6597d782012-10-08 16:29:07 -07003084 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3085 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07003087 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
Colin Cross3e4578f2015-10-27 16:42:08 -07003088 vma->vm_userfaultfd_ctx, vma_get_anon_name(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 if (new_vma) {
3090 /*
3091 * Source vma may have been merged into new_vma
3092 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003093 if (unlikely(vma_start >= new_vma->vm_start &&
3094 vma_start < new_vma->vm_end)) {
3095 /*
3096 * The only way we can get a vma_merge with
3097 * self during an mremap is if the vma hasn't
3098 * been faulted in yet and we were allowed to
3099 * reset the dst vma->vm_pgoff to the
3100 * destination address of the mremap to allow
3101 * the merge to happen. mremap must change the
3102 * vm_pgoff linearity between src and dst vmas
3103 * (in turn preventing a vma_merge) to be
3104 * safe. It is only safe to keep the vm_pgoff
3105 * linear if there are no pages mapped yet.
3106 */
Sasha Levin81d1b092014-10-09 15:28:10 -07003107 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07003108 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07003109 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07003110 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08003112 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07003113 if (!new_vma)
3114 goto out;
3115 *new_vma = *vma;
3116 new_vma->vm_start = addr;
3117 new_vma->vm_end = addr + len;
3118 new_vma->vm_pgoff = pgoff;
3119 if (vma_dup_policy(vma, new_vma))
3120 goto out_free_vma;
3121 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3122 if (anon_vma_clone(new_vma, vma))
3123 goto out_free_mempol;
3124 if (new_vma->vm_file)
3125 get_file(new_vma->vm_file);
3126 if (new_vma->vm_ops && new_vma->vm_ops->open)
3127 new_vma->vm_ops->open(new_vma);
3128 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3129 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 }
3131 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003132
Chen Gange3975892015-09-08 15:03:38 -07003133out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07003134 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07003135out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08003136 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07003137out:
Rik van Riel5beb4932010-03-05 13:42:07 -08003138 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003140
3141/*
3142 * Return true if the calling process may expand its vm space by the passed
3143 * number of pages
3144 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003145bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003146{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003147 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3148 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003149
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003150 if (is_data_mapping(flags) &&
3151 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003152 /* Workaround for Valgrind */
3153 if (rlimit(RLIMIT_DATA) == 0 &&
3154 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3155 return true;
3156 if (!ignore_rlimit_data) {
3157 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n",
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003158 current->comm, current->pid,
3159 (mm->data_vm + npages) << PAGE_SHIFT,
3160 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003161 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003162 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003163 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07003164
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003165 return true;
3166}
3167
3168void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3169{
3170 mm->total_vm += npages;
3171
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003172 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003173 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003174 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003175 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003176 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003177 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003178}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003179
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003180static int special_mapping_fault(struct vm_area_struct *vma,
3181 struct vm_fault *vmf);
3182
3183/*
3184 * Having a close hook prevents vma merging regardless of flags.
3185 */
3186static void special_mapping_close(struct vm_area_struct *vma)
3187{
3188}
3189
3190static const char *special_mapping_name(struct vm_area_struct *vma)
3191{
3192 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3193}
3194
Dmitry Safonovb059a452016-06-28 14:35:38 +03003195static int special_mapping_mremap(struct vm_area_struct *new_vma)
3196{
3197 struct vm_special_mapping *sm = new_vma->vm_private_data;
3198
3199 if (sm->mremap)
3200 return sm->mremap(sm, new_vma);
3201 return 0;
3202}
3203
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003204static const struct vm_operations_struct special_mapping_vmops = {
3205 .close = special_mapping_close,
3206 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03003207 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003208 .name = special_mapping_name,
3209};
3210
3211static const struct vm_operations_struct legacy_special_mapping_vmops = {
3212 .close = special_mapping_close,
3213 .fault = special_mapping_fault,
3214};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003215
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003216static int special_mapping_fault(struct vm_area_struct *vma,
3217 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003218{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003219 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003220 struct page **pages;
3221
Andy Lutomirskif872f542015-12-29 20:12:19 -08003222 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003223 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08003224 } else {
3225 struct vm_special_mapping *sm = vma->vm_private_data;
3226
3227 if (sm->fault)
3228 return sm->fault(sm, vma, vmf);
3229
3230 pages = sm->pages;
3231 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003232
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003233 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003234 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003235
3236 if (*pages) {
3237 struct page *page = *pages;
3238 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003239 vmf->page = page;
3240 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003241 }
3242
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003243 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003244}
3245
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003246static struct vm_area_struct *__install_special_mapping(
3247 struct mm_struct *mm,
3248 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003249 unsigned long vm_flags, void *priv,
3250 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003251{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003252 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003253 struct vm_area_struct *vma;
3254
3255 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3256 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003257 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003258
Rik van Riel5beb4932010-03-05 13:42:07 -08003259 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003260 vma->vm_mm = mm;
3261 vma->vm_start = addr;
3262 vma->vm_end = addr + len;
3263
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003264 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003265 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003266
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003267 vma->vm_ops = ops;
3268 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003269
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003270 ret = insert_vm_struct(mm, vma);
3271 if (ret)
3272 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003273
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003274 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003275
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003276 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003277
Stefani Seibold3935ed62014-03-17 23:22:02 +01003278 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003279
3280out:
3281 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003282 return ERR_PTR(ret);
3283}
3284
Dmitry Safonov2eefd872016-09-05 16:33:05 +03003285bool vma_is_special_mapping(const struct vm_area_struct *vma,
3286 const struct vm_special_mapping *sm)
3287{
3288 return vma->vm_private_data == sm &&
3289 (vma->vm_ops == &special_mapping_vmops ||
3290 vma->vm_ops == &legacy_special_mapping_vmops);
3291}
3292
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003293/*
3294 * Called with mm->mmap_sem held for writing.
3295 * Insert a new vma covering the given region, with the given flags.
3296 * Its pages are supplied by the given array of struct page *.
3297 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3298 * The region past the last page supplied will always produce SIGBUS.
3299 * The array pointer and the pages it points to are assumed to stay alive
3300 * for as long as this mapping might exist.
3301 */
3302struct vm_area_struct *_install_special_mapping(
3303 struct mm_struct *mm,
3304 unsigned long addr, unsigned long len,
3305 unsigned long vm_flags, const struct vm_special_mapping *spec)
3306{
Chen Gang27f28b92015-11-05 18:48:41 -08003307 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3308 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003309}
3310
Stefani Seibold3935ed62014-03-17 23:22:02 +01003311int install_special_mapping(struct mm_struct *mm,
3312 unsigned long addr, unsigned long len,
3313 unsigned long vm_flags, struct page **pages)
3314{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003315 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003316 mm, addr, len, vm_flags, (void *)pages,
3317 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003318
Duan Jiong14bd5b42014-06-04 16:07:05 -07003319 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003320}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003321
3322static DEFINE_MUTEX(mm_all_locks_mutex);
3323
Peter Zijlstra454ed842008-08-11 09:30:25 +02003324static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003325{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003326 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003327 /*
3328 * The LSB of head.next can't change from under us
3329 * because we hold the mm_all_locks_mutex.
3330 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003331 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003332 /*
3333 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003334 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003335 * the same anon_vma we won't take it again.
3336 *
3337 * No need of atomic instructions here, head.next
3338 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003339 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003340 */
3341 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003342 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003343 BUG();
3344 }
3345}
3346
Peter Zijlstra454ed842008-08-11 09:30:25 +02003347static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003348{
3349 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3350 /*
3351 * AS_MM_ALL_LOCKS can't change from under us because
3352 * we hold the mm_all_locks_mutex.
3353 *
3354 * Operations on ->flags have to be atomic because
3355 * even if AS_MM_ALL_LOCKS is stable thanks to the
3356 * mm_all_locks_mutex, there may be other cpus
3357 * changing other bitflags in parallel to us.
3358 */
3359 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3360 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003361 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003362 }
3363}
3364
3365/*
3366 * This operation locks against the VM for all pte/vma/mm related
3367 * operations that could ever happen on a certain mm. This includes
3368 * vmtruncate, try_to_unmap, and all page faults.
3369 *
3370 * The caller must take the mmap_sem in write mode before calling
3371 * mm_take_all_locks(). The caller isn't allowed to release the
3372 * mmap_sem until mm_drop_all_locks() returns.
3373 *
3374 * mmap_sem in write mode is required in order to block all operations
3375 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003376 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003377 * anon_vmas to be associated with existing vmas.
3378 *
3379 * A single task can't take more than one mm_take_all_locks() in a row
3380 * or it would deadlock.
3381 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003382 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003383 * mapping->flags avoid to take the same lock twice, if more than one
3384 * vma in this mm is backed by the same anon_vma or address_space.
3385 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003386 * We take locks in following order, accordingly to comment at beginning
3387 * of mm/rmap.c:
3388 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3389 * hugetlb mapping);
3390 * - all i_mmap_rwsem locks;
3391 * - all anon_vma->rwseml
3392 *
3393 * We can take all locks within these types randomly because the VM code
3394 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3395 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003396 *
3397 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3398 * that may have to take thousand of locks.
3399 *
3400 * mm_take_all_locks() can fail if it's interrupted by signals.
3401 */
3402int mm_take_all_locks(struct mm_struct *mm)
3403{
3404 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003405 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003406
3407 BUG_ON(down_read_trylock(&mm->mmap_sem));
3408
3409 mutex_lock(&mm_all_locks_mutex);
3410
3411 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3412 if (signal_pending(current))
3413 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003414 if (vma->vm_file && vma->vm_file->f_mapping &&
3415 is_vm_hugetlb_page(vma))
3416 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3417 }
3418
3419 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3420 if (signal_pending(current))
3421 goto out_unlock;
3422 if (vma->vm_file && vma->vm_file->f_mapping &&
3423 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003424 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003425 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003426
3427 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3428 if (signal_pending(current))
3429 goto out_unlock;
3430 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003431 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3432 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003433 }
3434
Kautuk Consul584cff52011-10-31 17:08:59 -07003435 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003436
3437out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003438 mm_drop_all_locks(mm);
3439 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003440}
3441
3442static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3443{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003444 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003445 /*
3446 * The LSB of head.next can't change to 0 from under
3447 * us because we hold the mm_all_locks_mutex.
3448 *
3449 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003450 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003451 * never see our bitflag.
3452 *
3453 * No need of atomic instructions here, head.next
3454 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003455 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003456 */
3457 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003458 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003459 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003460 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003461 }
3462}
3463
3464static void vm_unlock_mapping(struct address_space *mapping)
3465{
3466 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3467 /*
3468 * AS_MM_ALL_LOCKS can't change to 0 from under us
3469 * because we hold the mm_all_locks_mutex.
3470 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003471 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003472 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3473 &mapping->flags))
3474 BUG();
3475 }
3476}
3477
3478/*
3479 * The mmap_sem cannot be released by the caller until
3480 * mm_drop_all_locks() returns.
3481 */
3482void mm_drop_all_locks(struct mm_struct *mm)
3483{
3484 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003485 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003486
3487 BUG_ON(down_read_trylock(&mm->mmap_sem));
3488 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3489
3490 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3491 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003492 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3493 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003494 if (vma->vm_file && vma->vm_file->f_mapping)
3495 vm_unlock_mapping(vma->vm_file->f_mapping);
3496 }
3497
3498 mutex_unlock(&mm_all_locks_mutex);
3499}
David Howells8feae132009-01-08 12:04:47 +00003500
3501/*
3502 * initialise the VMA slab
3503 */
3504void __init mmap_init(void)
3505{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003506 int ret;
3507
Tejun Heo908c7f12014-09-08 09:51:29 +09003508 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003509 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003510}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003511
3512/*
3513 * Initialise sysctl_user_reserve_kbytes.
3514 *
3515 * This is intended to prevent a user from starting a single memory hogging
3516 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3517 * mode.
3518 *
3519 * The default value is min(3% of free memory, 128MB)
3520 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3521 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003522static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003523{
3524 unsigned long free_kbytes;
3525
3526 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3527
3528 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3529 return 0;
3530}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003531subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003532
3533/*
3534 * Initialise sysctl_admin_reserve_kbytes.
3535 *
3536 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3537 * to log in and kill a memory hogging process.
3538 *
3539 * Systems with more than 256MB will reserve 8MB, enough to recover
3540 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3541 * only reserve 3% of free pages by default.
3542 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003543static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003544{
3545 unsigned long free_kbytes;
3546
3547 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3548
3549 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3550 return 0;
3551}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003552subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003553
3554/*
3555 * Reinititalise user and admin reserves if memory is added or removed.
3556 *
3557 * The default user reserve max is 128MB, and the default max for the
3558 * admin reserve is 8MB. These are usually, but not always, enough to
3559 * enable recovery from a memory hogging process using login/sshd, a shell,
3560 * and tools like top. It may make sense to increase or even disable the
3561 * reserve depending on the existence of swap or variations in the recovery
3562 * tools. So, the admin may have changed them.
3563 *
3564 * If memory is added and the reserves have been eliminated or increased above
3565 * the default max, then we'll trust the admin.
3566 *
3567 * If memory is removed and there isn't enough free memory, then we
3568 * need to reset the reserves.
3569 *
3570 * Otherwise keep the reserve set by the admin.
3571 */
3572static int reserve_mem_notifier(struct notifier_block *nb,
3573 unsigned long action, void *data)
3574{
3575 unsigned long tmp, free_kbytes;
3576
3577 switch (action) {
3578 case MEM_ONLINE:
3579 /* Default max is 128MB. Leave alone if modified by operator. */
3580 tmp = sysctl_user_reserve_kbytes;
3581 if (0 < tmp && tmp < (1UL << 17))
3582 init_user_reserve();
3583
3584 /* Default max is 8MB. Leave alone if modified by operator. */
3585 tmp = sysctl_admin_reserve_kbytes;
3586 if (0 < tmp && tmp < (1UL << 13))
3587 init_admin_reserve();
3588
3589 break;
3590 case MEM_OFFLINE:
3591 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3592
3593 if (sysctl_user_reserve_kbytes > free_kbytes) {
3594 init_user_reserve();
3595 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3596 sysctl_user_reserve_kbytes);
3597 }
3598
3599 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3600 init_admin_reserve();
3601 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3602 sysctl_admin_reserve_kbytes);
3603 }
3604 break;
3605 default:
3606 break;
3607 }
3608 return NOTIFY_OK;
3609}
3610
3611static struct notifier_block reserve_mem_nb = {
3612 .notifier_call = reserve_mem_notifier,
3613};
3614
3615static int __meminit init_reserve_notifier(void)
3616{
3617 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003618 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003619
3620 return 0;
3621}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003622subsys_initcall(init_reserve_notifier);