blob: 6f90f07f12b8f623d21b0bf282e2ed0c5c5a2084 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#include <asm/uaccess.h>
49#include <asm/cacheflush.h>
50#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020051#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Jan Beulich42b77722008-07-23 21:27:10 -070053#include "internal.h"
54
Kirill Korotaev3a459752006-09-07 14:17:04 +040055#ifndef arch_mmap_check
56#define arch_mmap_check(addr, len, flags) (0)
57#endif
58
Daniel Cashmand07e2252016-01-14 15:19:53 -080059#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
60const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
61const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
62int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
63#endif
64#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
65const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
66const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
67int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
68#endif
69
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -070070static bool ignore_rlimit_data;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080071core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
Daniel Cashmand07e2252016-01-14 15:19:53 -080072
Hugh Dickinse0da3822005-04-19 13:29:15 -070073static void unmap_region(struct mm_struct *mm,
74 struct vm_area_struct *vma, struct vm_area_struct *prev,
75 unsigned long start, unsigned long end);
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/* description of effects of mapping type and prot in current implementation.
78 * this is due to the limited x86 page protection hardware. The expected
79 * behavior is in parens:
80 *
81 * map_type prot
82 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
83 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
84 * w: (no) no w: (no) no w: (yes) yes w: (no) no
85 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070086 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
88 * w: (no) no w: (no) no w: (copy) copy w: (no) no
89 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
90 *
Catalin Marinascab15ce2016-08-11 18:44:50 +010091 * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
92 * MAP_PRIVATE:
93 * r: (no) no
94 * w: (no) no
95 * x: (yes) yes
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 */
97pgprot_t protection_map[16] = {
98 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
99 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
100};
101
Hugh Dickins804af2c2006-07-26 21:39:49 +0100102pgprot_t vm_get_page_prot(unsigned long vm_flags)
103{
Dave Kleikampb845f312008-07-08 00:28:51 +1000104 return __pgprot(pgprot_val(protection_map[vm_flags &
105 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
106 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100107}
108EXPORT_SYMBOL(vm_get_page_prot);
109
Peter Feiner64e45502014-10-13 15:55:46 -0700110static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
111{
112 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
113}
114
115/* Update vma->vm_page_prot to reflect vma->vm_flags. */
116void vma_set_page_prot(struct vm_area_struct *vma)
117{
118 unsigned long vm_flags = vma->vm_flags;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700119 pgprot_t vm_page_prot;
Peter Feiner64e45502014-10-13 15:55:46 -0700120
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700121 vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
122 if (vma_wants_writenotify(vma, vm_page_prot)) {
Peter Feiner64e45502014-10-13 15:55:46 -0700123 vm_flags &= ~VM_SHARED;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700124 vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
Peter Feiner64e45502014-10-13 15:55:46 -0700125 }
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700126 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
127 WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
Peter Feiner64e45502014-10-13 15:55:46 -0700128}
129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800131 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 */
133static void __remove_shared_vm_struct(struct vm_area_struct *vma,
134 struct file *file, struct address_space *mapping)
135{
136 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500137 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700139 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
141 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800142 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 flush_dcache_mmap_unlock(mapping);
144}
145
146/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700147 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700148 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700150void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
152 struct file *file = vma->vm_file;
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 if (file) {
155 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800156 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800158 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700160}
161
162/*
163 * Close a vm structure and free it, returning the next.
164 */
165static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
166{
167 struct vm_area_struct *next = vma->vm_next;
168
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700169 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 if (vma->vm_ops && vma->vm_ops->close)
171 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700172 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700173 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700174 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700176 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700179static int do_brk(unsigned long addr, unsigned long len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700180
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100181SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700183 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 unsigned long newbrk, oldbrk;
185 struct mm_struct *mm = current->mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700186 struct vm_area_struct *next;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700187 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800188 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700190 if (down_write_killable(&mm->mmap_sem))
191 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700193#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800194 /*
195 * CONFIG_COMPAT_BRK can still be overridden by setting
196 * randomize_va_space to 2, which will still cause mm->start_brk
197 * to be arbitrarily shifted
198 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700199 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800200 min_brk = mm->start_brk;
201 else
202 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700203#else
204 min_brk = mm->start_brk;
205#endif
206 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700208
209 /*
210 * Check against rlimit here. If this check is done later after the test
211 * of oldbrk with newbrk then it can escape the test and let the data
212 * segment grow beyond its set limit the in case where the limit is
213 * not page aligned -Ram Gupta
214 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700215 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
216 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700217 goto out;
218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 newbrk = PAGE_ALIGN(brk);
220 oldbrk = PAGE_ALIGN(mm->brk);
221 if (oldbrk == newbrk)
222 goto set_brk;
223
224 /* Always allow shrinking brk. */
225 if (brk <= mm->brk) {
226 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
227 goto set_brk;
228 goto out;
229 }
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 /* Check against existing mmap mappings. */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700232 next = find_vma(mm, oldbrk);
233 if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 goto out;
235
236 /* Ok, looks good - let it rip. */
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700237 if (do_brk(oldbrk, newbrk-oldbrk) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240set_brk:
241 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800242 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
243 up_write(&mm->mmap_sem);
244 if (populate)
245 mm_populate(oldbrk, newbrk - oldbrk);
246 return brk;
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248out:
249 retval = mm->brk;
250 up_write(&mm->mmap_sem);
251 return retval;
252}
253
Michel Lespinassed3737182012-12-11 16:01:38 -0800254static long vma_compute_subtree_gap(struct vm_area_struct *vma)
255{
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700256 unsigned long max, prev_end, subtree_gap;
257
258 /*
259 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
260 * allow two stack_guard_gaps between them here, and when choosing
261 * an unmapped area; whereas when expanding we only require one.
262 * That's a little inconsistent, but keeps the code here simpler.
263 */
264 max = vm_start_gap(vma);
265 if (vma->vm_prev) {
266 prev_end = vm_end_gap(vma->vm_prev);
267 if (max > prev_end)
268 max -= prev_end;
269 else
270 max = 0;
271 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800272 if (vma->vm_rb.rb_left) {
273 subtree_gap = rb_entry(vma->vm_rb.rb_left,
274 struct vm_area_struct, vm_rb)->rb_subtree_gap;
275 if (subtree_gap > max)
276 max = subtree_gap;
277 }
278 if (vma->vm_rb.rb_right) {
279 subtree_gap = rb_entry(vma->vm_rb.rb_right,
280 struct vm_area_struct, vm_rb)->rb_subtree_gap;
281 if (subtree_gap > max)
282 max = subtree_gap;
283 }
284 return max;
285}
286
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700287#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800288static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800290 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800291 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 struct rb_node *nd, *pn = NULL;
293 unsigned long prev = 0, pend = 0;
294
295 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
296 struct vm_area_struct *vma;
297 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800298 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700299 pr_emerg("vm_start %lx < prev %lx\n",
300 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800301 bug = 1;
302 }
303 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700304 pr_emerg("vm_start %lx < pend %lx\n",
305 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800306 bug = 1;
307 }
308 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700309 pr_emerg("vm_start %lx > vm_end %lx\n",
310 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800311 bug = 1;
312 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800313 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800314 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700315 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800316 vma->rb_subtree_gap,
317 vma_compute_subtree_gap(vma));
318 bug = 1;
319 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800320 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 i++;
322 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800323 prev = vma->vm_start;
324 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 }
326 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800327 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800329 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700330 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800331 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800333 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334}
335
Michel Lespinassed3737182012-12-11 16:01:38 -0800336static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
337{
338 struct rb_node *nd;
339
340 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
341 struct vm_area_struct *vma;
342 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700343 VM_BUG_ON_VMA(vma != ignore &&
344 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
345 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347}
348
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700349static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
351 int bug = 0;
352 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800353 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700354 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700355
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700356 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800357 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700358 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700359
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800360 if (anon_vma) {
361 anon_vma_lock_read(anon_vma);
362 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
363 anon_vma_interval_tree_verify(avc);
364 anon_vma_unlock_read(anon_vma);
365 }
366
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700367 highest_address = vm_end_gap(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700368 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 i++;
370 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800371 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700372 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800373 bug = 1;
374 }
375 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700376 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700377 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800378 bug = 1;
379 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800380 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800381 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700382 if (i != -1)
383 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800384 bug = 1;
385 }
Sasha Levin96dad672014-10-09 15:28:39 -0700386 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800389#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390#define validate_mm(mm) do { } while (0)
391#endif
392
Michel Lespinassed3737182012-12-11 16:01:38 -0800393RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
394 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
395
396/*
397 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
398 * vma->vm_prev->vm_end values changed, without modifying the vma's position
399 * in the rbtree.
400 */
401static void vma_gap_update(struct vm_area_struct *vma)
402{
403 /*
404 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
405 * function that does exacltly what we want.
406 */
407 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
408}
409
410static inline void vma_rb_insert(struct vm_area_struct *vma,
411 struct rb_root *root)
412{
413 /* All rb_subtree_gap values must be consistent prior to insertion */
414 validate_mm_rb(root, NULL);
415
416 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
417}
418
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700419static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
420{
421 /*
422 * Note rb_erase_augmented is a fairly large inline function,
423 * so make sure we instantiate it only once with our desired
424 * augmented rbtree callbacks.
425 */
426 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
427}
428
429static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
430 struct rb_root *root,
431 struct vm_area_struct *ignore)
432{
433 /*
434 * All rb_subtree_gap values must be consistent prior to erase,
435 * with the possible exception of the "next" vma being erased if
436 * next->vm_start was reduced.
437 */
438 validate_mm_rb(root, ignore);
439
440 __vma_rb_erase(vma, root);
441}
442
443static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
444 struct rb_root *root)
Michel Lespinassed3737182012-12-11 16:01:38 -0800445{
446 /*
447 * All rb_subtree_gap values must be consistent prior to erase,
448 * with the possible exception of the vma being erased.
449 */
450 validate_mm_rb(root, vma);
451
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700452 __vma_rb_erase(vma, root);
Michel Lespinassed3737182012-12-11 16:01:38 -0800453}
454
Michel Lespinassebf181b92012-10-08 16:31:39 -0700455/*
456 * vma has some anon_vma assigned, and is already inserted on that
457 * anon_vma's interval trees.
458 *
459 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
460 * vma must be removed from the anon_vma's interval trees using
461 * anon_vma_interval_tree_pre_update_vma().
462 *
463 * After the update, the vma will be reinserted using
464 * anon_vma_interval_tree_post_update_vma().
465 *
466 * The entire update must be protected by exclusive mmap_sem and by
467 * the root anon_vma's mutex.
468 */
469static inline void
470anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
471{
472 struct anon_vma_chain *avc;
473
474 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
475 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
476}
477
478static inline void
479anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
480{
481 struct anon_vma_chain *avc;
482
483 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
484 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
485}
486
Hugh Dickins6597d782012-10-08 16:29:07 -0700487static int find_vma_links(struct mm_struct *mm, unsigned long addr,
488 unsigned long end, struct vm_area_struct **pprev,
489 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
Hugh Dickins6597d782012-10-08 16:29:07 -0700491 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493 __rb_link = &mm->mm_rb.rb_node;
494 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 while (*__rb_link) {
497 struct vm_area_struct *vma_tmp;
498
499 __rb_parent = *__rb_link;
500 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
501
502 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700503 /* Fail if an existing vma overlaps the area */
504 if (vma_tmp->vm_start < end)
505 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 __rb_link = &__rb_parent->rb_left;
507 } else {
508 rb_prev = __rb_parent;
509 __rb_link = &__rb_parent->rb_right;
510 }
511 }
512
513 *pprev = NULL;
514 if (rb_prev)
515 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
516 *rb_link = __rb_link;
517 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700518 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
520
Cyril Hrubise8420a82013-04-29 15:08:33 -0700521static unsigned long count_vma_pages_range(struct mm_struct *mm,
522 unsigned long addr, unsigned long end)
523{
524 unsigned long nr_pages = 0;
525 struct vm_area_struct *vma;
526
527 /* Find first overlaping mapping */
528 vma = find_vma_intersection(mm, addr, end);
529 if (!vma)
530 return 0;
531
532 nr_pages = (min(end, vma->vm_end) -
533 max(addr, vma->vm_start)) >> PAGE_SHIFT;
534
535 /* Iterate over the rest of the overlaps */
536 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
537 unsigned long overlap_len;
538
539 if (vma->vm_start > end)
540 break;
541
542 overlap_len = min(end, vma->vm_end) - vma->vm_start;
543 nr_pages += overlap_len >> PAGE_SHIFT;
544 }
545
546 return nr_pages;
547}
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
550 struct rb_node **rb_link, struct rb_node *rb_parent)
551{
Michel Lespinassed3737182012-12-11 16:01:38 -0800552 /* Update tracking information for the gap following the new vma. */
553 if (vma->vm_next)
554 vma_gap_update(vma->vm_next);
555 else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700556 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800557
558 /*
559 * vma->vm_prev wasn't known when we followed the rbtree to find the
560 * correct insertion point for that vma. As a result, we could not
561 * update the vma vm_rb parents rb_subtree_gap values on the way down.
562 * So, we first insert the vma with a zero rb_subtree_gap value
563 * (to be consistent with what we did on the way down), and then
564 * immediately update the gap to the correct value. Finally we
565 * rebalance the rbtree after all augmented values have been set.
566 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800568 vma->rb_subtree_gap = 0;
569 vma_gap_update(vma);
570 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571}
572
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700573static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
ZhenwenXu48aae422009-01-06 14:40:21 -0800575 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 file = vma->vm_file;
578 if (file) {
579 struct address_space *mapping = file->f_mapping;
580
581 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500582 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700584 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800587 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 flush_dcache_mmap_unlock(mapping);
589 }
590}
591
592static void
593__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
594 struct vm_area_struct *prev, struct rb_node **rb_link,
595 struct rb_node *rb_parent)
596{
597 __vma_link_list(mm, vma, prev, rb_parent);
598 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
601static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
602 struct vm_area_struct *prev, struct rb_node **rb_link,
603 struct rb_node *rb_parent)
604{
605 struct address_space *mapping = NULL;
606
Huang Shijie64ac4942014-06-04 16:07:33 -0700607 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800609 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700610 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 __vma_link(mm, vma, prev, rb_link, rb_parent);
613 __vma_link_file(vma);
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800616 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
618 mm->map_count++;
619 validate_mm(mm);
620}
621
622/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700623 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700624 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800626static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
Hugh Dickins6597d782012-10-08 16:29:07 -0700628 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800629 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Hugh Dickins6597d782012-10-08 16:29:07 -0700631 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
632 &prev, &rb_link, &rb_parent))
633 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 __vma_link(mm, vma, prev, rb_link, rb_parent);
635 mm->map_count++;
636}
637
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700638static __always_inline void __vma_unlink_common(struct mm_struct *mm,
639 struct vm_area_struct *vma,
640 struct vm_area_struct *prev,
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700641 bool has_prev,
642 struct vm_area_struct *ignore)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
Michel Lespinassed3737182012-12-11 16:01:38 -0800644 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700645
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700646 vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700647 next = vma->vm_next;
648 if (has_prev)
649 prev->vm_next = next;
650 else {
651 prev = vma->vm_prev;
652 if (prev)
653 prev->vm_next = next;
654 else
655 mm->mmap = next;
656 }
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700657 if (next)
658 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700659
660 /* Kill the cache */
661 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662}
663
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700664static inline void __vma_unlink_prev(struct mm_struct *mm,
665 struct vm_area_struct *vma,
666 struct vm_area_struct *prev)
667{
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700668 __vma_unlink_common(mm, vma, prev, true, vma);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700669}
670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671/*
672 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
673 * is already present in an i_mmap tree without adjusting the tree.
674 * The following helper function should be used when such adjustments
675 * are necessary. The "insert" vma (if any) is to be inserted
676 * before we drop the necessary locks.
677 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700678int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
679 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
680 struct vm_area_struct *expand)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct mm_struct *mm = vma->vm_mm;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700683 struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700685 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700686 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800688 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 long adjust_next = 0;
690 int remove_next = 0;
691
692 if (next && !insert) {
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700693 struct vm_area_struct *exporter = NULL, *importer = NULL;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700694
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 if (end >= next->vm_end) {
696 /*
697 * vma expands, overlapping all the next, and
698 * perhaps the one after too (mprotect case 6).
Andrea Arcangeli86d12e42016-10-07 17:01:34 -0700699 * The only other cases that gets here are
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700700 * case 1, case 7 and case 8.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700702 if (next == expand) {
703 /*
704 * The only case where we don't expand "vma"
705 * and we expand "next" instead is case 8.
706 */
707 VM_WARN_ON(end != next->vm_end);
708 /*
709 * remove_next == 3 means we're
710 * removing "vma" and that to do so we
711 * swapped "vma" and "next".
712 */
713 remove_next = 3;
714 VM_WARN_ON(file != next->vm_file);
715 swap(vma, next);
716 } else {
717 VM_WARN_ON(expand != vma);
718 /*
719 * case 1, 6, 7, remove_next == 2 is case 6,
720 * remove_next == 1 is case 1 or 7.
721 */
722 remove_next = 1 + (end > next->vm_end);
723 VM_WARN_ON(remove_next == 2 &&
724 end != next->vm_next->vm_end);
725 VM_WARN_ON(remove_next == 1 &&
726 end != next->vm_end);
727 /* trim end to next, for case 6 first pass */
728 end = next->vm_end;
729 }
730
Linus Torvalds287d97a2010-04-10 15:22:30 -0700731 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 importer = vma;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700733
734 /*
735 * If next doesn't have anon_vma, import from vma after
736 * next, if the vma overlaps with it.
737 */
Andrea Arcangeli97a42cd2016-10-07 17:01:31 -0700738 if (remove_next == 2 && !next->anon_vma)
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700739 exporter = next->vm_next;
740
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 } else if (end > next->vm_start) {
742 /*
743 * vma expands, overlapping part of the next:
744 * mprotect case 5 shifting the boundary up.
745 */
746 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700747 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 importer = vma;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700749 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 } else if (end < vma->vm_end) {
751 /*
752 * vma shrinks, and !insert tells it's not
753 * split_vma inserting another: so it must be
754 * mprotect case 4 shifting the boundary down.
755 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700756 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700757 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 importer = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700759 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Rik van Riel5beb4932010-03-05 13:42:07 -0800762 /*
763 * Easily overlooked: when mprotect shifts the boundary,
764 * make sure the expanding vma has anon_vma set if the
765 * shrinking vma had, to cover any anon pages imported.
766 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700767 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800768 int error;
769
Linus Torvalds287d97a2010-04-10 15:22:30 -0700770 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300771 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700772 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300773 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800774 }
775 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700776again:
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700777 vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
Kirill A. Shutemov37f9f552016-07-26 15:25:48 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (file) {
780 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800781 root = &mapping->i_mmap;
782 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530783
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800784 if (adjust_next)
785 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530786
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800787 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700790 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 * are visible to arm/parisc __flush_dcache_page
792 * throughout; but we cannot insert into address
793 * space until vma start or end is updated.
794 */
795 __vma_link_file(insert);
796 }
797 }
798
Michel Lespinassebf181b92012-10-08 16:31:39 -0700799 anon_vma = vma->anon_vma;
800 if (!anon_vma && adjust_next)
801 anon_vma = next->anon_vma;
802 if (anon_vma) {
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700803 VM_WARN_ON(adjust_next && next->anon_vma &&
804 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000805 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700806 anon_vma_interval_tree_pre_update_vma(vma);
807 if (adjust_next)
808 anon_vma_interval_tree_pre_update_vma(next);
809 }
Rik van Riel012f18002010-08-09 17:18:40 -0700810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 if (root) {
812 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700813 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700815 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 }
817
Michel Lespinassed3737182012-12-11 16:01:38 -0800818 if (start != vma->vm_start) {
819 vma->vm_start = start;
820 start_changed = true;
821 }
822 if (end != vma->vm_end) {
823 vma->vm_end = end;
824 end_changed = true;
825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 vma->vm_pgoff = pgoff;
827 if (adjust_next) {
828 next->vm_start += adjust_next << PAGE_SHIFT;
829 next->vm_pgoff += adjust_next;
830 }
831
832 if (root) {
833 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700834 vma_interval_tree_insert(next, root);
835 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 flush_dcache_mmap_unlock(mapping);
837 }
838
839 if (remove_next) {
840 /*
841 * vma_merge has merged next into vma, and needs
842 * us to remove next before dropping the locks.
843 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700844 if (remove_next != 3)
845 __vma_unlink_prev(mm, next, vma);
846 else
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700847 /*
848 * vma is not before next if they've been
849 * swapped.
850 *
851 * pre-swap() next->vm_start was reduced so
852 * tell validate_mm_rb to ignore pre-swap()
853 * "next" (which is stored in post-swap()
854 * "vma").
855 */
856 __vma_unlink_common(mm, next, NULL, false, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (file)
858 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 } else if (insert) {
860 /*
861 * split_vma has split insert from vma, and needs
862 * us to insert it before dropping the locks
863 * (it may either follow vma or precede it).
864 */
865 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800866 } else {
867 if (start_changed)
868 vma_gap_update(vma);
869 if (end_changed) {
870 if (!next)
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700871 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800872 else if (!adjust_next)
873 vma_gap_update(next);
874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 }
876
Michel Lespinassebf181b92012-10-08 16:31:39 -0700877 if (anon_vma) {
878 anon_vma_interval_tree_post_update_vma(vma);
879 if (adjust_next)
880 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800881 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800884 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530886 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100887 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530888
889 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100890 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530891 }
892
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700894 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530895 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700897 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800898 if (next->anon_vma)
899 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700901 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 kmem_cache_free(vm_area_cachep, next);
903 /*
904 * In mprotect's case 6 (see comments on vma_merge),
905 * we must remove another next too. It would clutter
906 * up the code too much to do both in one go.
907 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700908 if (remove_next != 3) {
909 /*
910 * If "next" was removed and vma->vm_end was
911 * expanded (up) over it, in turn
912 * "next->vm_prev->vm_end" changed and the
913 * "vma->vm_next" gap must be updated.
914 */
915 next = vma->vm_next;
916 } else {
917 /*
918 * For the scope of the comment "next" and
919 * "vma" considered pre-swap(): if "vma" was
920 * removed, next->vm_start was expanded (down)
921 * over it and the "next" gap must be updated.
922 * Because of the swap() the post-swap() "vma"
923 * actually points to pre-swap() "next"
924 * (post-swap() "next" as opposed is now a
925 * dangling pointer).
926 */
927 next = vma;
928 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700929 if (remove_next == 2) {
930 remove_next = 1;
931 end = next->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 goto again;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700933 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800934 else if (next)
935 vma_gap_update(next);
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700936 else {
937 /*
938 * If remove_next == 2 we obviously can't
939 * reach this path.
940 *
941 * If remove_next == 3 we can't reach this
942 * path because pre-swap() next is always not
943 * NULL. pre-swap() "next" is not being
944 * removed and its next->vm_end is not altered
945 * (and furthermore "end" already matches
946 * next->vm_end in remove_next == 3).
947 *
948 * We reach this only in the remove_next == 1
949 * case if the "next" vma that was removed was
950 * the highest vma of the mm. However in such
951 * case next->vm_end == "end" and the extended
952 * "vma" has vma->vm_end == next->vm_end so
953 * mm->highest_vm_end doesn't need any update
954 * in remove_next == 1 case.
955 */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -0700956 VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530959 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100960 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
962 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800963
964 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
967/*
968 * If the vma has a ->close operation then the driver probably needs to release
969 * per-vma resources, so we don't attempt to merge those.
970 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700972 struct file *file, unsigned long vm_flags,
Colin Cross3e4578f2015-10-27 16:42:08 -0700973 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
974 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800976 /*
977 * VM_SOFTDIRTY should not prevent from VMA merging, if we
978 * match the flags but dirty bit -- the caller should mark
979 * merged VMA as dirty. If dirty bit won't be excluded from
980 * comparison, we increase pressue on the memory system forcing
981 * the kernel to generate new VMAs when old one could be
982 * extended instead.
983 */
984 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 return 0;
986 if (vma->vm_file != file)
987 return 0;
988 if (vma->vm_ops && vma->vm_ops->close)
989 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700990 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
991 return 0;
Colin Cross3e4578f2015-10-27 16:42:08 -0700992 if (vma_get_anon_name(vma) != anon_name)
993 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 return 1;
995}
996
997static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700998 struct anon_vma *anon_vma2,
999 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000{
Shaohua Li965f55d2011-05-24 17:11:20 -07001001 /*
1002 * The list_is_singular() test is to avoid merging VMA cloned from
1003 * parents. This can improve scalability caused by anon_vma lock.
1004 */
1005 if ((!anon_vma1 || !anon_vma2) && (!vma ||
1006 list_is_singular(&vma->anon_vma_chain)))
1007 return 1;
1008 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009}
1010
1011/*
1012 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1013 * in front of (at a lower virtual address and file offset than) the vma.
1014 *
1015 * We cannot merge two vmas if they have differently assigned (non-NULL)
1016 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1017 *
1018 * We don't check here for the merged mmap wrapping around the end of pagecache
1019 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1020 * wrap, nor mmaps which cover the final page at index -1UL.
1021 */
1022static int
1023can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001024 struct anon_vma *anon_vma, struct file *file,
1025 pgoff_t vm_pgoff,
Colin Cross3e4578f2015-10-27 16:42:08 -07001026 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1027 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
Colin Cross3e4578f2015-10-27 16:42:08 -07001029 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001030 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 if (vma->vm_pgoff == vm_pgoff)
1032 return 1;
1033 }
1034 return 0;
1035}
1036
1037/*
1038 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1039 * beyond (at a higher virtual address and file offset than) the vma.
1040 *
1041 * We cannot merge two vmas if they have differently assigned (non-NULL)
1042 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1043 */
1044static int
1045can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001046 struct anon_vma *anon_vma, struct file *file,
1047 pgoff_t vm_pgoff,
Colin Cross3e4578f2015-10-27 16:42:08 -07001048 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1049 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050{
Colin Cross3e4578f2015-10-27 16:42:08 -07001051 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001052 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001054 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1056 return 1;
1057 }
1058 return 0;
1059}
1060
1061/*
Colin Cross3e4578f2015-10-27 16:42:08 -07001062 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
1063 * figure out whether that can be merged with its predecessor or its
1064 * successor. Or both (it neatly fills a hole).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 *
1066 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1067 * certain not to be mapped by the time vma_merge is called; but when
1068 * called for mprotect, it is certain to be already mapped (either at
1069 * an offset within prev, or at the start of next), and the flags of
1070 * this area are about to be changed to vm_flags - and the no-change
1071 * case has already been eliminated.
1072 *
1073 * The following mprotect cases have to be considered, where AAAA is
1074 * the area passed down from mprotect_fixup, never extending beyond one
1075 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1076 *
1077 * AAAA AAAA AAAA AAAA
1078 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1079 * cannot merge might become might become might become
1080 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1081 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001082 * mremap move: PPPPXXXXXXXX 8
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 * AAAA
1084 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1085 * might become case 1 below case 2 below case 3 below
1086 *
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001087 * It is important for case 8 that the the vma NNNN overlapping the
1088 * region AAAA is never going to extended over XXXX. Instead XXXX must
1089 * be extended in region AAAA and NNNN must be removed. This way in
1090 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1091 * rmap_locks, the properties of the merged vma will be already
1092 * correct for the whole merged range. Some of those properties like
1093 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1094 * be correct for the whole merged range immediately after the
1095 * rmap_locks are released. Otherwise if XXXX would be removed and
1096 * NNNN would be extended over the XXXX range, remove_migration_ptes
1097 * or other rmap walkers (if working on addresses beyond the "end"
1098 * parameter) may establish ptes with the wrong permissions of NNNN
1099 * instead of the right permissions of XXXX.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 */
1101struct vm_area_struct *vma_merge(struct mm_struct *mm,
1102 struct vm_area_struct *prev, unsigned long addr,
1103 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001104 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001105 pgoff_t pgoff, struct mempolicy *policy,
Colin Cross3e4578f2015-10-27 16:42:08 -07001106 struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1107 const char __user *anon_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108{
1109 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1110 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001111 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 /*
1114 * We later require that vma->vm_flags == vm_flags,
1115 * so this tests vma->vm_flags & VM_SPECIAL, too.
1116 */
1117 if (vm_flags & VM_SPECIAL)
1118 return NULL;
1119
1120 if (prev)
1121 next = prev->vm_next;
1122 else
1123 next = mm->mmap;
1124 area = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001125 if (area && area->vm_end == end) /* cases 6, 7, 8 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 next = next->vm_next;
1127
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001128 /* verify some invariant that must be enforced by the caller */
1129 VM_WARN_ON(prev && addr <= prev->vm_start);
1130 VM_WARN_ON(area && end > area->vm_end);
1131 VM_WARN_ON(addr >= end);
1132
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 /*
1134 * Can it merge with the predecessor?
1135 */
1136 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001137 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001139 anon_vma, file, pgoff,
Colin Cross3e4578f2015-10-27 16:42:08 -07001140 vm_userfaultfd_ctx,
1141 anon_name)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 /*
1143 * OK, it can. Can we now merge in the successor as well?
1144 */
1145 if (next && end == next->vm_start &&
1146 mpol_equal(policy, vma_policy(next)) &&
1147 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001148 anon_vma, file,
1149 pgoff+pglen,
Colin Cross3e4578f2015-10-27 16:42:08 -07001150 vm_userfaultfd_ctx,
1151 anon_name) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001153 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 /* cases 1, 6 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001155 err = __vma_adjust(prev, prev->vm_start,
1156 next->vm_end, prev->vm_pgoff, NULL,
1157 prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 } else /* cases 2, 5, 7 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001159 err = __vma_adjust(prev, prev->vm_start,
1160 end, prev->vm_pgoff, NULL, prev);
Rik van Riel5beb4932010-03-05 13:42:07 -08001161 if (err)
1162 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001163 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 return prev;
1165 }
1166
1167 /*
1168 * Can this new request be merged in front of next?
1169 */
1170 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001171 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001173 anon_vma, file, pgoff+pglen,
Colin Cross3e4578f2015-10-27 16:42:08 -07001174 vm_userfaultfd_ctx,
1175 anon_name)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 if (prev && addr < prev->vm_end) /* case 4 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001177 err = __vma_adjust(prev, prev->vm_start,
1178 addr, prev->vm_pgoff, NULL, next);
1179 else { /* cases 3, 8 */
1180 err = __vma_adjust(area, addr, next->vm_end,
1181 next->vm_pgoff - pglen, NULL, next);
1182 /*
1183 * In case 3 area is already equal to next and
1184 * this is a noop, but in case 8 "area" has
1185 * been removed and next was expanded over it.
1186 */
1187 area = next;
1188 }
Rik van Riel5beb4932010-03-05 13:42:07 -08001189 if (err)
1190 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001191 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 return area;
1193 }
1194
1195 return NULL;
1196}
1197
1198/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001199 * Rough compatbility check to quickly see if it's even worth looking
1200 * at sharing an anon_vma.
1201 *
1202 * They need to have the same vm_file, and the flags can only differ
1203 * in things that mprotect may change.
1204 *
1205 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1206 * we can merge the two vma's. For example, we refuse to merge a vma if
1207 * there is a vm_ops->close() function, because that indicates that the
1208 * driver is doing some kind of reference counting. But that doesn't
1209 * really matter for the anon_vma sharing case.
1210 */
1211static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1212{
1213 return a->vm_end == b->vm_start &&
1214 mpol_equal(vma_policy(a), vma_policy(b)) &&
1215 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001216 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001217 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1218}
1219
1220/*
1221 * Do some basic sanity checking to see if we can re-use the anon_vma
1222 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1223 * the same as 'old', the other will be the new one that is trying
1224 * to share the anon_vma.
1225 *
1226 * NOTE! This runs with mm_sem held for reading, so it is possible that
1227 * the anon_vma of 'old' is concurrently in the process of being set up
1228 * by another page fault trying to merge _that_. But that's ok: if it
1229 * is being set up, that automatically means that it will be a singleton
1230 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001231 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001232 *
1233 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1234 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1235 * is to return an anon_vma that is "complex" due to having gone through
1236 * a fork).
1237 *
1238 * We also make sure that the two vma's are compatible (adjacent,
1239 * and with the same memory policies). That's all stable, even with just
1240 * a read lock on the mm_sem.
1241 */
1242static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1243{
1244 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001245 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001246
1247 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1248 return anon_vma;
1249 }
1250 return NULL;
1251}
1252
1253/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1255 * neighbouring vmas for a suitable anon_vma, before it goes off
1256 * to allocate a new anon_vma. It checks because a repetitive
1257 * sequence of mprotects and faults may otherwise lead to distinct
1258 * anon_vmas being allocated, preventing vma merge in subsequent
1259 * mprotect.
1260 */
1261struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1262{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001263 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266 near = vma->vm_next;
1267 if (!near)
1268 goto try_prev;
1269
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001270 anon_vma = reusable_anon_vma(near, vma, near);
1271 if (anon_vma)
1272 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001274 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 if (!near)
1276 goto none;
1277
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001278 anon_vma = reusable_anon_vma(near, near, vma);
1279 if (anon_vma)
1280 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281none:
1282 /*
1283 * There's no absolute need to look only at touching neighbours:
1284 * we could search further afield for "compatible" anon_vmas.
1285 * But it would probably just be a waste of time searching,
1286 * or lead to too many vmas hanging off the same anon_vma.
1287 * We're trying to allow mprotect remerging later on,
1288 * not trying to minimize memory used for anon_vmas.
1289 */
1290 return NULL;
1291}
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293/*
Al Viro40401532012-02-13 03:58:52 +00001294 * If a hint addr is less than mmap_min_addr change hint to be as
1295 * low as possible but still greater than mmap_min_addr
1296 */
1297static inline unsigned long round_hint_to_min(unsigned long hint)
1298{
1299 hint &= PAGE_MASK;
1300 if (((void *)hint != NULL) &&
1301 (hint < mmap_min_addr))
1302 return PAGE_ALIGN(mmap_min_addr);
1303 return hint;
1304}
1305
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001306static inline int mlock_future_check(struct mm_struct *mm,
1307 unsigned long flags,
1308 unsigned long len)
1309{
1310 unsigned long locked, lock_limit;
1311
1312 /* mlock MCL_FUTURE? */
1313 if (flags & VM_LOCKED) {
1314 locked = len >> PAGE_SHIFT;
1315 locked += mm->locked_vm;
1316 lock_limit = rlimit(RLIMIT_MEMLOCK);
1317 lock_limit >>= PAGE_SHIFT;
1318 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1319 return -EAGAIN;
1320 }
1321 return 0;
1322}
1323
Al Viro40401532012-02-13 03:58:52 +00001324/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001325 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001327unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001329 unsigned long flags, vm_flags_t vm_flags,
1330 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
vishnu.pscc71aba2014-10-09 15:26:29 -07001332 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001333 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Michel Lespinasse41badc12013-02-22 16:32:47 -08001335 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001336
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001337 if (!len)
1338 return -EINVAL;
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 /*
1341 * Does the application expect PROT_READ to imply PROT_EXEC?
1342 *
1343 * (the exception is when the underlying filesystem is noexec
1344 * mounted, in which case we dont add PROT_EXEC.)
1345 */
1346 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001347 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 prot |= PROT_EXEC;
1349
Eric Paris7cd94142007-11-26 18:47:40 -05001350 if (!(flags & MAP_FIXED))
1351 addr = round_hint_to_min(addr);
1352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 /* Careful about overflows.. */
1354 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001355 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 return -ENOMEM;
1357
1358 /* offset overflow? */
1359 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001360 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362 /* Too many mappings? */
1363 if (mm->map_count > sysctl_max_map_count)
1364 return -ENOMEM;
1365
1366 /* Obtain the address to map to. we verify (or select) it and ensure
1367 * that it represents a valid section of the address space.
1368 */
1369 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001370 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 return addr;
1372
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001373 if (prot == PROT_EXEC) {
1374 pkey = execute_only_pkey(mm);
1375 if (pkey < 0)
1376 pkey = 0;
1377 }
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 /* Do simple checking here so the lower-level routines won't have
1380 * to. we assume access permissions have been handled by the open
1381 * of the memory object, so we don't do any here.
1382 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001383 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1385
Huang Shijiecdf7b342009-09-21 17:03:36 -07001386 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 if (!can_do_mlock())
1388 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001389
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001390 if (mlock_future_check(mm, vm_flags, len))
1391 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001394 struct inode *inode = file_inode(file);
1395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 switch (flags & MAP_TYPE) {
1397 case MAP_SHARED:
1398 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1399 return -EACCES;
1400
1401 /*
1402 * Make sure we don't allow writing to an append-only
1403 * file..
1404 */
1405 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1406 return -EACCES;
1407
1408 /*
1409 * Make sure there are no mandatory locks on the file.
1410 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001411 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 return -EAGAIN;
1413
1414 vm_flags |= VM_SHARED | VM_MAYSHARE;
1415 if (!(file->f_mode & FMODE_WRITE))
1416 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1417
1418 /* fall through */
1419 case MAP_PRIVATE:
1420 if (!(file->f_mode & FMODE_READ))
1421 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001422 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001423 if (vm_flags & VM_EXEC)
1424 return -EPERM;
1425 vm_flags &= ~VM_MAYEXEC;
1426 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001427
Al Viro72c2d532013-09-22 16:27:52 -04001428 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001429 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001430 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1431 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 break;
1433
1434 default:
1435 return -EINVAL;
1436 }
1437 } else {
1438 switch (flags & MAP_TYPE) {
1439 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001440 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1441 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001442 /*
1443 * Ignore pgoff.
1444 */
1445 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 vm_flags |= VM_SHARED | VM_MAYSHARE;
1447 break;
1448 case MAP_PRIVATE:
1449 /*
1450 * Set pgoff according to addr for anon_vma.
1451 */
1452 pgoff = addr >> PAGE_SHIFT;
1453 break;
1454 default:
1455 return -EINVAL;
1456 }
1457 }
1458
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001459 /*
1460 * Set 'VM_NORESERVE' if we should not account for the
1461 * memory use of this mapping.
1462 */
1463 if (flags & MAP_NORESERVE) {
1464 /* We honor MAP_NORESERVE if allowed to overcommit */
1465 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1466 vm_flags |= VM_NORESERVE;
1467
1468 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1469 if (file && is_file_hugepages(file))
1470 vm_flags |= VM_NORESERVE;
1471 }
1472
1473 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001474 if (!IS_ERR_VALUE(addr) &&
1475 ((vm_flags & VM_LOCKED) ||
1476 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001477 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001478 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001479}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001480
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001481SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1482 unsigned long, prot, unsigned long, flags,
1483 unsigned long, fd, unsigned long, pgoff)
1484{
1485 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001486 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001487
1488 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001489 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001490 file = fget(fd);
1491 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001492 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001493 if (is_file_hugepages(file))
1494 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001495 retval = -EINVAL;
1496 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1497 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001498 } else if (flags & MAP_HUGETLB) {
1499 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001500 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001501
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001502 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001503 if (!hs)
1504 return -EINVAL;
1505
1506 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001507 /*
1508 * VM_NORESERVE is used because the reservations will be
1509 * taken when vm_ops->mmap() is called
1510 * A dummy user value is used because we are not locking
1511 * memory so no accounting is necessary
1512 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001513 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001514 VM_NORESERVE,
1515 &user, HUGETLB_ANONHUGE_INODE,
1516 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001517 if (IS_ERR(file))
1518 return PTR_ERR(file);
1519 }
1520
1521 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1522
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001523 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001524out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001525 if (file)
1526 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001527 return retval;
1528}
1529
Christoph Hellwiga4679372010-03-10 15:21:15 -08001530#ifdef __ARCH_WANT_SYS_OLD_MMAP
1531struct mmap_arg_struct {
1532 unsigned long addr;
1533 unsigned long len;
1534 unsigned long prot;
1535 unsigned long flags;
1536 unsigned long fd;
1537 unsigned long offset;
1538};
1539
1540SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1541{
1542 struct mmap_arg_struct a;
1543
1544 if (copy_from_user(&a, arg, sizeof(a)))
1545 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001546 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001547 return -EINVAL;
1548
1549 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1550 a.offset >> PAGE_SHIFT);
1551}
1552#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1553
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001554/*
1555 * Some shared mappigns will want the pages marked read-only
1556 * to track write events. If so, we'll downgrade vm_page_prot
1557 * to the private version (using protection_map[] without the
1558 * VM_SHARED bit).
1559 */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001560int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001561{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001562 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001563 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001564
1565 /* If it was private or non-writable, the write bit is already clear */
1566 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1567 return 0;
1568
1569 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001570 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001571 return 1;
1572
Peter Feiner64e45502014-10-13 15:55:46 -07001573 /* The open routine did something to the protections that pgprot_modify
1574 * won't preserve? */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001575 if (pgprot_val(vm_page_prot) !=
1576 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001577 return 0;
1578
Peter Feiner64e45502014-10-13 15:55:46 -07001579 /* Do we need to track softdirty? */
1580 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1581 return 1;
1582
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001583 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001584 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001585 return 0;
1586
1587 /* Can the mapping track the dirty pages? */
1588 return vma->vm_file && vma->vm_file->f_mapping &&
1589 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1590}
1591
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001592/*
1593 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001594 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001595 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001596static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001597{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001598 /*
1599 * hugetlb has its own accounting separate from the core VM
1600 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1601 */
1602 if (file && is_file_hugepages(file))
1603 return 0;
1604
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001605 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1606}
1607
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001608unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001609 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001610{
1611 struct mm_struct *mm = current->mm;
1612 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001613 int error;
1614 struct rb_node **rb_link, *rb_parent;
1615 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001616
Cyril Hrubise8420a82013-04-29 15:08:33 -07001617 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001618 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001619 unsigned long nr_pages;
1620
1621 /*
1622 * MAP_FIXED may remove pages of mappings that intersects with
1623 * requested mapping. Account for the pages it would unmap.
1624 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001625 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1626
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001627 if (!may_expand_vm(mm, vm_flags,
1628 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001629 return -ENOMEM;
1630 }
1631
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001633 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1634 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 if (do_munmap(mm, addr, len))
1636 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 }
1638
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001639 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001640 * Private writable mapping: check memory availability
1641 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001642 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001643 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001644 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001645 return -ENOMEM;
1646 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 }
1648
1649 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001650 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001652 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
Colin Cross3e4578f2015-10-27 16:42:08 -07001653 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001654 if (vma)
1655 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657 /*
1658 * Determine the object being mapped and call the appropriate
1659 * specific mapper. the address has already been validated, but
1660 * not unmapped, but the maps are removed from the list.
1661 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001662 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (!vma) {
1664 error = -ENOMEM;
1665 goto unacct_error;
1666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
1668 vma->vm_mm = mm;
1669 vma->vm_start = addr;
1670 vma->vm_end = addr + len;
1671 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001672 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001674 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (vm_flags & VM_DENYWRITE) {
1678 error = deny_write_access(file);
1679 if (error)
1680 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001682 if (vm_flags & VM_SHARED) {
1683 error = mapping_map_writable(file->f_mapping);
1684 if (error)
1685 goto allow_write_and_free_vma;
1686 }
1687
1688 /* ->mmap() can change vma->vm_file, but must guarantee that
1689 * vma_link() below can deny write-access if VM_DENYWRITE is set
1690 * and map writably if VM_SHARED is set. This usually means the
1691 * new file must not have been exposed to user-space, yet.
1692 */
Al Virocb0942b2012-08-27 14:48:26 -04001693 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 error = file->f_op->mmap(file, vma);
1695 if (error)
1696 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001697
1698 /* Can addr have changed??
1699 *
1700 * Answer: Yes, several device drivers can do it in their
1701 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001702 * Bug: If addr is changed, prev, rb_link, rb_parent should
1703 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001704 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001705 WARN_ON_ONCE(addr != vma->vm_start);
1706
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001707 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001708 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 } else if (vm_flags & VM_SHARED) {
1710 error = shmem_zero_setup(vma);
1711 if (error)
1712 goto free_vma;
1713 }
1714
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001715 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001716 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001717 if (file) {
1718 if (vm_flags & VM_SHARED)
1719 mapping_unmap_writable(file->f_mapping);
1720 if (vm_flags & VM_DENYWRITE)
1721 allow_write_access(file);
1722 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001723 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001724out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001725 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001726
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001727 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001729 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1730 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001731 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001732 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001733 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001734 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301735
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001736 if (file)
1737 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301738
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001739 /*
1740 * New (or expanded) vma always get soft dirty status.
1741 * Otherwise user-space soft-dirty page tracker won't
1742 * be able to distinguish situation when vma area unmapped,
1743 * then new mapped in-place (which must be aimed as
1744 * a completely new data area).
1745 */
1746 vma->vm_flags |= VM_SOFTDIRTY;
1747
Peter Feiner64e45502014-10-13 15:55:46 -07001748 vma_set_page_prot(vma);
1749
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 return addr;
1751
1752unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 vma->vm_file = NULL;
1754 fput(file);
1755
1756 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001757 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1758 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001759 if (vm_flags & VM_SHARED)
1760 mapping_unmap_writable(file->f_mapping);
1761allow_write_and_free_vma:
1762 if (vm_flags & VM_DENYWRITE)
1763 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764free_vma:
1765 kmem_cache_free(vm_area_cachep, vma);
1766unacct_error:
1767 if (charged)
1768 vm_unacct_memory(charged);
1769 return error;
1770}
1771
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001772unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1773{
1774 /*
1775 * We implement the search by looking for an rbtree node that
1776 * immediately follows a suitable gap. That is,
1777 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1778 * - gap_end = vma->vm_start >= info->low_limit + length;
1779 * - gap_end - gap_start >= length
1780 */
1781
1782 struct mm_struct *mm = current->mm;
1783 struct vm_area_struct *vma;
1784 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1785
1786 /* Adjust search length to account for worst case alignment overhead */
1787 length = info->length + info->align_mask;
1788 if (length < info->length)
1789 return -ENOMEM;
1790
1791 /* Adjust search limits by the desired length */
1792 if (info->high_limit < length)
1793 return -ENOMEM;
1794 high_limit = info->high_limit - length;
1795
1796 if (info->low_limit > high_limit)
1797 return -ENOMEM;
1798 low_limit = info->low_limit + length;
1799
1800 /* Check if rbtree root looks promising */
1801 if (RB_EMPTY_ROOT(&mm->mm_rb))
1802 goto check_highest;
1803 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1804 if (vma->rb_subtree_gap < length)
1805 goto check_highest;
1806
1807 while (true) {
1808 /* Visit left subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001809 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001810 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1811 struct vm_area_struct *left =
1812 rb_entry(vma->vm_rb.rb_left,
1813 struct vm_area_struct, vm_rb);
1814 if (left->rb_subtree_gap >= length) {
1815 vma = left;
1816 continue;
1817 }
1818 }
1819
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001820 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001821check_current:
1822 /* Check if current node has a suitable gap */
1823 if (gap_start > high_limit)
1824 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001825 if (gap_end >= low_limit &&
1826 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001827 goto found;
1828
1829 /* Visit right subtree if it looks promising */
1830 if (vma->vm_rb.rb_right) {
1831 struct vm_area_struct *right =
1832 rb_entry(vma->vm_rb.rb_right,
1833 struct vm_area_struct, vm_rb);
1834 if (right->rb_subtree_gap >= length) {
1835 vma = right;
1836 continue;
1837 }
1838 }
1839
1840 /* Go back up the rbtree to find next candidate node */
1841 while (true) {
1842 struct rb_node *prev = &vma->vm_rb;
1843 if (!rb_parent(prev))
1844 goto check_highest;
1845 vma = rb_entry(rb_parent(prev),
1846 struct vm_area_struct, vm_rb);
1847 if (prev == vma->vm_rb.rb_left) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001848 gap_start = vm_end_gap(vma->vm_prev);
1849 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001850 goto check_current;
1851 }
1852 }
1853 }
1854
1855check_highest:
1856 /* Check highest gap, which does not precede any rbtree node */
1857 gap_start = mm->highest_vm_end;
1858 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1859 if (gap_start > high_limit)
1860 return -ENOMEM;
1861
1862found:
1863 /* We found a suitable gap. Clip it with the original low_limit. */
1864 if (gap_start < info->low_limit)
1865 gap_start = info->low_limit;
1866
1867 /* Adjust gap address to the desired alignment */
1868 gap_start += (info->align_offset - gap_start) & info->align_mask;
1869
1870 VM_BUG_ON(gap_start + info->length > info->high_limit);
1871 VM_BUG_ON(gap_start + info->length > gap_end);
1872 return gap_start;
1873}
1874
1875unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1876{
1877 struct mm_struct *mm = current->mm;
1878 struct vm_area_struct *vma;
1879 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1880
1881 /* Adjust search length to account for worst case alignment overhead */
1882 length = info->length + info->align_mask;
1883 if (length < info->length)
1884 return -ENOMEM;
1885
1886 /*
1887 * Adjust search limits by the desired length.
1888 * See implementation comment at top of unmapped_area().
1889 */
1890 gap_end = info->high_limit;
1891 if (gap_end < length)
1892 return -ENOMEM;
1893 high_limit = gap_end - length;
1894
1895 if (info->low_limit > high_limit)
1896 return -ENOMEM;
1897 low_limit = info->low_limit + length;
1898
1899 /* Check highest gap, which does not precede any rbtree node */
1900 gap_start = mm->highest_vm_end;
1901 if (gap_start <= high_limit)
1902 goto found_highest;
1903
1904 /* Check if rbtree root looks promising */
1905 if (RB_EMPTY_ROOT(&mm->mm_rb))
1906 return -ENOMEM;
1907 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1908 if (vma->rb_subtree_gap < length)
1909 return -ENOMEM;
1910
1911 while (true) {
1912 /* Visit right subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001913 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001914 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1915 struct vm_area_struct *right =
1916 rb_entry(vma->vm_rb.rb_right,
1917 struct vm_area_struct, vm_rb);
1918 if (right->rb_subtree_gap >= length) {
1919 vma = right;
1920 continue;
1921 }
1922 }
1923
1924check_current:
1925 /* Check if current node has a suitable gap */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001926 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001927 if (gap_end < low_limit)
1928 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001929 if (gap_start <= high_limit &&
1930 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001931 goto found;
1932
1933 /* Visit left subtree if it looks promising */
1934 if (vma->vm_rb.rb_left) {
1935 struct vm_area_struct *left =
1936 rb_entry(vma->vm_rb.rb_left,
1937 struct vm_area_struct, vm_rb);
1938 if (left->rb_subtree_gap >= length) {
1939 vma = left;
1940 continue;
1941 }
1942 }
1943
1944 /* Go back up the rbtree to find next candidate node */
1945 while (true) {
1946 struct rb_node *prev = &vma->vm_rb;
1947 if (!rb_parent(prev))
1948 return -ENOMEM;
1949 vma = rb_entry(rb_parent(prev),
1950 struct vm_area_struct, vm_rb);
1951 if (prev == vma->vm_rb.rb_right) {
1952 gap_start = vma->vm_prev ?
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001953 vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001954 goto check_current;
1955 }
1956 }
1957 }
1958
1959found:
1960 /* We found a suitable gap. Clip it with the original high_limit. */
1961 if (gap_end > info->high_limit)
1962 gap_end = info->high_limit;
1963
1964found_highest:
1965 /* Compute highest gap address at the desired alignment */
1966 gap_end -= info->length;
1967 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1968
1969 VM_BUG_ON(gap_end < info->low_limit);
1970 VM_BUG_ON(gap_end < gap_start);
1971 return gap_end;
1972}
1973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974/* Get an address range which is currently unmapped.
1975 * For shmat() with addr=0.
1976 *
1977 * Ugly calling convention alert:
1978 * Return value with the low bits set means error value,
1979 * ie
1980 * if (ret & ~PAGE_MASK)
1981 * error = ret;
1982 *
1983 * This function "knows" that -ENOMEM has the bits set.
1984 */
1985#ifndef HAVE_ARCH_UNMAPPED_AREA
1986unsigned long
1987arch_get_unmapped_area(struct file *filp, unsigned long addr,
1988 unsigned long len, unsigned long pgoff, unsigned long flags)
1989{
1990 struct mm_struct *mm = current->mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001991 struct vm_area_struct *vma, *prev;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001992 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001994 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 return -ENOMEM;
1996
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001997 if (flags & MAP_FIXED)
1998 return addr;
1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 if (addr) {
2001 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002002 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002003 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002004 (!vma || addr + len <= vm_start_gap(vma)) &&
2005 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 return addr;
2007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002009 info.flags = 0;
2010 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08002011 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002012 info.high_limit = TASK_SIZE;
2013 info.align_mask = 0;
2014 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015}
vishnu.pscc71aba2014-10-09 15:26:29 -07002016#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018/*
2019 * This mmap-allocator allocates new areas top-down from below the
2020 * stack's low limit (the base):
2021 */
2022#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2023unsigned long
2024arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
2025 const unsigned long len, const unsigned long pgoff,
2026 const unsigned long flags)
2027{
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002028 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002030 unsigned long addr = addr0;
2031 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
2033 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002034 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 return -ENOMEM;
2036
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002037 if (flags & MAP_FIXED)
2038 return addr;
2039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 /* requesting a specific address */
2041 if (addr) {
2042 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002043 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002044 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002045 (!vma || addr + len <= vm_start_gap(vma)) &&
2046 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return addr;
2048 }
2049
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002050 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2051 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002052 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002053 info.high_limit = mm->mmap_base;
2054 info.align_mask = 0;
2055 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07002056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 /*
2058 * A failed mmap() very likely causes application failure,
2059 * so fall back to the bottom-up function here. This scenario
2060 * can happen with large stack limits and large mmap()
2061 * allocations.
2062 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002063 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002064 VM_BUG_ON(addr != -ENOMEM);
2065 info.flags = 0;
2066 info.low_limit = TASK_UNMAPPED_BASE;
2067 info.high_limit = TASK_SIZE;
2068 addr = vm_unmapped_area(&info);
2069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071 return addr;
2072}
2073#endif
2074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075unsigned long
2076get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2077 unsigned long pgoff, unsigned long flags)
2078{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002079 unsigned long (*get_area)(struct file *, unsigned long,
2080 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Al Viro9206de92009-12-03 15:23:11 -05002082 unsigned long error = arch_mmap_check(addr, len, flags);
2083 if (error)
2084 return error;
2085
2086 /* Careful about overflows.. */
2087 if (len > TASK_SIZE)
2088 return -ENOMEM;
2089
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002090 get_area = current->mm->get_unmapped_area;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002091 if (file) {
2092 if (file->f_op->get_unmapped_area)
2093 get_area = file->f_op->get_unmapped_area;
2094 } else if (flags & MAP_SHARED) {
2095 /*
2096 * mmap_region() will call shmem_zero_setup() to create a file,
2097 * so use shmem's get_unmapped_area in case it can be huge.
2098 * do_mmap_pgoff() will clear pgoff, so match alignment.
2099 */
2100 pgoff = 0;
2101 get_area = shmem_get_unmapped_area;
2102 }
2103
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002104 addr = get_area(file, addr, len, pgoff, flags);
2105 if (IS_ERR_VALUE(addr))
2106 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002108 if (addr > TASK_SIZE - len)
2109 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002110 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002111 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002112
Al Viro9ac4ed42012-05-30 17:13:15 -04002113 error = security_mmap_addr(addr);
2114 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115}
2116
2117EXPORT_SYMBOL(get_unmapped_area);
2118
2119/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002120struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002122 struct rb_node *rb_node;
2123 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002125 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002126 vma = vmacache_find(mm, addr);
2127 if (likely(vma))
2128 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002130 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002132 while (rb_node) {
2133 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002135 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002136
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002137 if (tmp->vm_end > addr) {
2138 vma = tmp;
2139 if (tmp->vm_start <= addr)
2140 break;
2141 rb_node = rb_node->rb_left;
2142 } else
2143 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002145
2146 if (vma)
2147 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 return vma;
2149}
2150
2151EXPORT_SYMBOL(find_vma);
2152
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002153/*
2154 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002155 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156struct vm_area_struct *
2157find_vma_prev(struct mm_struct *mm, unsigned long addr,
2158 struct vm_area_struct **pprev)
2159{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002160 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002162 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002163 if (vma) {
2164 *pprev = vma->vm_prev;
2165 } else {
2166 struct rb_node *rb_node = mm->mm_rb.rb_node;
2167 *pprev = NULL;
2168 while (rb_node) {
2169 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2170 rb_node = rb_node->rb_right;
2171 }
2172 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002173 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174}
2175
2176/*
2177 * Verify that the stack growth is acceptable and
2178 * update accounting. This is shared with both the
2179 * grow-up and grow-down cases.
2180 */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002181static int acct_stack_growth(struct vm_area_struct *vma,
2182 unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183{
2184 struct mm_struct *mm = vma->vm_mm;
2185 struct rlimit *rlim = current->signal->rlim;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002186 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
2188 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002189 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 return -ENOMEM;
2191
2192 /* Stack limit test */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002193 if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 return -ENOMEM;
2195
2196 /* mlock limit tests */
2197 if (vma->vm_flags & VM_LOCKED) {
2198 unsigned long locked;
2199 unsigned long limit;
2200 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002201 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002202 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 if (locked > limit && !capable(CAP_IPC_LOCK))
2204 return -ENOMEM;
2205 }
2206
Adam Litke0d59a012007-01-30 14:35:39 -08002207 /* Check to ensure the stack will not grow into a hugetlb-only region */
2208 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2209 vma->vm_end - size;
2210 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2211 return -EFAULT;
2212
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 /*
2214 * Overcommit.. This must be the final test, as it will
2215 * update security statistics.
2216 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002217 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 return -ENOMEM;
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 return 0;
2221}
2222
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002223#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002225 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2226 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002228int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229{
Oleg Nesterov09357812015-11-05 18:48:17 -08002230 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002231 struct vm_area_struct *next;
2232 unsigned long gap_addr;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002233 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234
2235 if (!(vma->vm_flags & VM_GROWSUP))
2236 return -EFAULT;
2237
Helge Deller5d10ad622017-06-19 17:34:05 +02002238 /* Guard against exceeding limits of the address space. */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002239 address &= PAGE_MASK;
Helge Deller38dfd2e2017-07-14 14:49:38 -07002240 if (address >= (TASK_SIZE & PAGE_MASK))
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002241 return -ENOMEM;
Helge Deller5d10ad622017-06-19 17:34:05 +02002242 address += PAGE_SIZE;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002243
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002244 /* Enforce stack_guard_gap */
2245 gap_addr = address + stack_guard_gap;
Helge Deller5d10ad622017-06-19 17:34:05 +02002246
2247 /* Guard against overflow */
2248 if (gap_addr < address || gap_addr > TASK_SIZE)
2249 gap_addr = TASK_SIZE;
2250
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002251 next = vma->vm_next;
2252 if (next && next->vm_start < gap_addr) {
2253 if (!(next->vm_flags & VM_GROWSUP))
2254 return -ENOMEM;
2255 /* Check that both stack segments have the same anon_vma? */
2256 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002257
2258 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 if (unlikely(anon_vma_prepare(vma)))
2260 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261
2262 /*
2263 * vma->vm_start/vm_end cannot change under us because the caller
2264 * is required to hold the mmap_sem in read mode. We need the
2265 * anon_vma lock to serialize against concurrent expand_stacks.
2266 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002267 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
2269 /* Somebody else might have raced and expanded it already */
2270 if (address > vma->vm_end) {
2271 unsigned long size, grow;
2272
2273 size = address - vma->vm_start;
2274 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2275
Hugh Dickins42c36f62011-05-09 17:44:42 -07002276 error = -ENOMEM;
2277 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2278 error = acct_stack_growth(vma, size, grow);
2279 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002280 /*
2281 * vma_gap_update() doesn't support concurrent
2282 * updates, but we only hold a shared mmap_sem
2283 * lock here, so we need to protect against
2284 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002285 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002286 * we don't guarantee that all growable vmas
2287 * in a mm share the same root anon vma.
2288 * So, we reuse mm->page_table_lock to guard
2289 * against concurrent vma expansions.
2290 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002291 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002292 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002293 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002294 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002295 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002296 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002297 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002298 if (vma->vm_next)
2299 vma_gap_update(vma->vm_next);
2300 else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002301 mm->highest_vm_end = vm_end_gap(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002302 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002303
Hugh Dickins42c36f62011-05-09 17:44:42 -07002304 perf_event_mmap(vma);
2305 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002308 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002309 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002310 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 return error;
2312}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002313#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315/*
2316 * vma is the first one with address < vma->vm_start. Have to extend vma.
2317 */
Michal Hockod05f3162011-05-24 17:11:44 -07002318int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002319 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320{
Oleg Nesterov09357812015-11-05 18:48:17 -08002321 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002322 struct vm_area_struct *prev;
2323 unsigned long gap_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 int error;
2325
Eric Paris88694772007-11-26 18:47:26 -05002326 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002327 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002328 if (error)
2329 return error;
2330
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002331 /* Enforce stack_guard_gap */
2332 gap_addr = address - stack_guard_gap;
2333 if (gap_addr > address)
2334 return -ENOMEM;
2335 prev = vma->vm_prev;
2336 if (prev && prev->vm_end > gap_addr) {
2337 if (!(prev->vm_flags & VM_GROWSDOWN))
2338 return -ENOMEM;
2339 /* Check that both stack segments have the same anon_vma? */
2340 }
2341
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002342 /* We must make sure the anon_vma is allocated. */
2343 if (unlikely(anon_vma_prepare(vma)))
2344 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346 /*
2347 * vma->vm_start/vm_end cannot change under us because the caller
2348 * is required to hold the mmap_sem in read mode. We need the
2349 * anon_vma lock to serialize against concurrent expand_stacks.
2350 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002351 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
2353 /* Somebody else might have raced and expanded it already */
2354 if (address < vma->vm_start) {
2355 unsigned long size, grow;
2356
2357 size = vma->vm_end - address;
2358 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2359
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002360 error = -ENOMEM;
2361 if (grow <= vma->vm_pgoff) {
2362 error = acct_stack_growth(vma, size, grow);
2363 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002364 /*
2365 * vma_gap_update() doesn't support concurrent
2366 * updates, but we only hold a shared mmap_sem
2367 * lock here, so we need to protect against
2368 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002369 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002370 * we don't guarantee that all growable vmas
2371 * in a mm share the same root anon vma.
2372 * So, we reuse mm->page_table_lock to guard
2373 * against concurrent vma expansions.
2374 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002375 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002376 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002377 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002378 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002379 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002380 vma->vm_start = address;
2381 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002382 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002383 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002384 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002385
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002386 perf_event_mmap(vma);
2387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 }
2389 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002390 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002391 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002392 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 return error;
2394}
2395
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002396/* enforced gap between the expanding stack and other mappings. */
2397unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2398
2399static int __init cmdline_parse_stack_guard_gap(char *p)
2400{
2401 unsigned long val;
2402 char *endptr;
2403
2404 val = simple_strtoul(p, &endptr, 10);
2405 if (!*endptr)
2406 stack_guard_gap = val << PAGE_SHIFT;
2407
2408 return 0;
2409}
2410__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2411
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002412#ifdef CONFIG_STACK_GROWSUP
2413int expand_stack(struct vm_area_struct *vma, unsigned long address)
2414{
2415 return expand_upwards(vma, address);
2416}
2417
2418struct vm_area_struct *
2419find_extend_vma(struct mm_struct *mm, unsigned long addr)
2420{
2421 struct vm_area_struct *vma, *prev;
2422
2423 addr &= PAGE_MASK;
2424 vma = find_vma_prev(mm, addr, &prev);
2425 if (vma && (vma->vm_start <= addr))
2426 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002427 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002428 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002429 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002430 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002431 return prev;
2432}
2433#else
2434int expand_stack(struct vm_area_struct *vma, unsigned long address)
2435{
2436 return expand_downwards(vma, address);
2437}
2438
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002440find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441{
vishnu.pscc71aba2014-10-09 15:26:29 -07002442 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 unsigned long start;
2444
2445 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002446 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 if (!vma)
2448 return NULL;
2449 if (vma->vm_start <= addr)
2450 return vma;
2451 if (!(vma->vm_flags & VM_GROWSDOWN))
2452 return NULL;
2453 start = vma->vm_start;
2454 if (expand_stack(vma, addr))
2455 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002456 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002457 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 return vma;
2459}
2460#endif
2461
Jesse Barnese1d6d012014-12-12 16:55:27 -08002462EXPORT_SYMBOL_GPL(find_extend_vma);
2463
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002465 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002467 *
2468 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002470static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002472 unsigned long nr_accounted = 0;
2473
Hugh Dickins365e9c872005-10-29 18:16:18 -07002474 /* Update high watermark before we lower total_vm */
2475 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002477 long nrpages = vma_pages(vma);
2478
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002479 if (vma->vm_flags & VM_ACCOUNT)
2480 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002481 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002482 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002483 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002484 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 validate_mm(mm);
2486}
2487
2488/*
2489 * Get rid of page table information in the indicated region.
2490 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002491 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 */
2493static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002494 struct vm_area_struct *vma, struct vm_area_struct *prev,
2495 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496{
vishnu.pscc71aba2014-10-09 15:26:29 -07002497 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002498 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499
2500 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002501 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002502 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002503 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002504 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002505 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002506 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507}
2508
2509/*
2510 * Create a list of vma's touched by the unmap, removing them from the mm's
2511 * vma list as we go..
2512 */
2513static void
2514detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2515 struct vm_area_struct *prev, unsigned long end)
2516{
2517 struct vm_area_struct **insertion_point;
2518 struct vm_area_struct *tail_vma = NULL;
2519
2520 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002521 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002523 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 mm->map_count--;
2525 tail_vma = vma;
2526 vma = vma->vm_next;
2527 } while (vma && vma->vm_start < end);
2528 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002529 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002530 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002531 vma_gap_update(vma);
2532 } else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002533 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002535
2536 /* Kill the cache */
2537 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538}
2539
2540/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002541 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2542 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002544static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 unsigned long addr, int new_below)
2546{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002548 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Andi Kleena5516432008-07-23 21:27:41 -07002550 if (is_vm_hugetlb_page(vma) && (addr &
2551 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 return -EINVAL;
2553
Christoph Lametere94b1762006-12-06 20:33:17 -08002554 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002556 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
2558 /* most fields are the same, copy all, and then fixup */
2559 *new = *vma;
2560
Rik van Riel5beb4932010-03-05 13:42:07 -08002561 INIT_LIST_HEAD(&new->anon_vma_chain);
2562
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 if (new_below)
2564 new->vm_end = addr;
2565 else {
2566 new->vm_start = addr;
2567 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2568 }
2569
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002570 err = vma_dup_policy(vma, new);
2571 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002572 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002574 err = anon_vma_clone(new, vma);
2575 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002576 goto out_free_mpol;
2577
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002578 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 get_file(new->vm_file);
2580
2581 if (new->vm_ops && new->vm_ops->open)
2582 new->vm_ops->open(new);
2583
2584 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002585 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2587 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002588 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
Rik van Riel5beb4932010-03-05 13:42:07 -08002590 /* Success. */
2591 if (!err)
2592 return 0;
2593
2594 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002595 if (new->vm_ops && new->vm_ops->close)
2596 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002597 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002598 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002599 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002600 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002601 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002602 out_free_vma:
2603 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002604 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605}
2606
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002607/*
2608 * Split a vma into two pieces at address 'addr', a new vma is allocated
2609 * either for the first part or the tail.
2610 */
2611int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2612 unsigned long addr, int new_below)
2613{
2614 if (mm->map_count >= sysctl_max_map_count)
2615 return -ENOMEM;
2616
2617 return __split_vma(mm, vma, addr, new_below);
2618}
2619
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620/* Munmap is split into 2 main parts -- this part which finds
2621 * what needs doing, and the areas themselves, which do the
2622 * work. This now handles partial unmappings.
2623 * Jeremy Fitzhardinge <jeremy@goop.org>
2624 */
2625int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2626{
2627 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002628 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002630 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 return -EINVAL;
2632
vishnu.pscc71aba2014-10-09 15:26:29 -07002633 len = PAGE_ALIGN(len);
2634 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 return -EINVAL;
2636
2637 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002638 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002639 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002641 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002642 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
2644 /* if it doesn't overlap, we have nothing.. */
2645 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002646 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 return 0;
2648
2649 /*
2650 * If we need to split any vma, do it now to save pain later.
2651 *
2652 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2653 * unmapped vm_area_struct will remain in use: so lower split_vma
2654 * places tmp vma above, and higher split_vma places tmp vma below.
2655 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002656 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002657 int error;
2658
2659 /*
2660 * Make sure that map_count on return from munmap() will
2661 * not exceed its limit; but let map_count go just above
2662 * its limit temporarily, to help free resources as expected.
2663 */
2664 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2665 return -ENOMEM;
2666
2667 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 if (error)
2669 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002670 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 }
2672
2673 /* Does it split the last one? */
2674 last = find_vma(mm, end);
2675 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002676 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 if (error)
2678 return error;
2679 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002680 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
2682 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002683 * unlock any mlock()ed ranges before detaching vmas
2684 */
2685 if (mm->locked_vm) {
2686 struct vm_area_struct *tmp = vma;
2687 while (tmp && tmp->vm_start < end) {
2688 if (tmp->vm_flags & VM_LOCKED) {
2689 mm->locked_vm -= vma_pages(tmp);
2690 munlock_vma_pages_all(tmp);
2691 }
2692 tmp = tmp->vm_next;
2693 }
2694 }
2695
2696 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 * Remove the vma's, and unmap the actual pages
2698 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002699 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2700 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
Dave Hansen1de4fa12014-11-14 07:18:31 -08002702 arch_unmap(mm, vma, start, end);
2703
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002705 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
2707 return 0;
2708}
Guenter Roeckabfb4862016-03-24 10:39:14 -07002709EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
Al Virobfce2812012-04-20 21:57:04 -04002711int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002712{
2713 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002714 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002715
Michal Hockoae798782016-05-23 16:25:33 -07002716 if (down_write_killable(&mm->mmap_sem))
2717 return -EINTR;
2718
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002719 ret = do_munmap(mm, start, len);
2720 up_write(&mm->mmap_sem);
2721 return ret;
2722}
2723EXPORT_SYMBOL(vm_munmap);
2724
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002725SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726{
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002727 int ret;
2728 struct mm_struct *mm = current->mm;
2729
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 profile_munmap(addr);
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002731 if (down_write_killable(&mm->mmap_sem))
2732 return -EINTR;
2733 ret = do_munmap(mm, addr, len);
2734 up_write(&mm->mmap_sem);
2735 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736}
2737
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002738
2739/*
2740 * Emulation of deprecated remap_file_pages() syscall.
2741 */
2742SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2743 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2744{
2745
2746 struct mm_struct *mm = current->mm;
2747 struct vm_area_struct *vma;
2748 unsigned long populate = 0;
2749 unsigned long ret = -EINVAL;
2750 struct file *file;
2751
Joe Perches756a025f02016-03-17 14:19:47 -07002752 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2753 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002754
2755 if (prot)
2756 return ret;
2757 start = start & PAGE_MASK;
2758 size = size & PAGE_MASK;
2759
2760 if (start + size <= start)
2761 return ret;
2762
2763 /* Does pgoff wrap? */
2764 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2765 return ret;
2766
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002767 if (down_write_killable(&mm->mmap_sem))
2768 return -EINTR;
2769
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002770 vma = find_vma(mm, start);
2771
2772 if (!vma || !(vma->vm_flags & VM_SHARED))
2773 goto out;
2774
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002775 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002776 goto out;
2777
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002778 if (start + size > vma->vm_end) {
2779 struct vm_area_struct *next;
2780
2781 for (next = vma->vm_next; next; next = next->vm_next) {
2782 /* hole between vmas ? */
2783 if (next->vm_start != next->vm_prev->vm_end)
2784 goto out;
2785
2786 if (next->vm_file != vma->vm_file)
2787 goto out;
2788
2789 if (next->vm_flags != vma->vm_flags)
2790 goto out;
2791
2792 if (start + size <= next->vm_end)
2793 break;
2794 }
2795
2796 if (!next)
2797 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002798 }
2799
2800 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2801 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2802 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2803
2804 flags &= MAP_NONBLOCK;
2805 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2806 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002807 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002808 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002809
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002810 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002811 for (tmp = vma; tmp->vm_start >= start + size;
2812 tmp = tmp->vm_next) {
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07002813 /*
2814 * Split pmd and munlock page on the border
2815 * of the range.
2816 */
2817 vma_adjust_trans_huge(tmp, start, start + size, 0);
2818
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002819 munlock_vma_pages_range(tmp,
2820 max(tmp->vm_start, start),
2821 min(tmp->vm_end, start + size));
2822 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002823 }
2824
2825 file = get_file(vma->vm_file);
2826 ret = do_mmap_pgoff(vma->vm_file, start, size,
2827 prot, flags, pgoff, &populate);
2828 fput(file);
2829out:
2830 up_write(&mm->mmap_sem);
2831 if (populate)
2832 mm_populate(ret, populate);
2833 if (!IS_ERR_VALUE(ret))
2834 ret = 0;
2835 return ret;
2836}
2837
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838static inline void verify_mm_writelocked(struct mm_struct *mm)
2839{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002840#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2842 WARN_ON(1);
2843 up_read(&mm->mmap_sem);
2844 }
2845#endif
2846}
2847
2848/*
2849 * this is really a simplified "do_mmap". it only handles
2850 * anonymous maps. eventually we may be able to do some
2851 * brk-specific accounting here.
2852 */
Kees Cookba093a62016-08-02 14:04:54 -07002853static int do_brk(unsigned long addr, unsigned long request)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854{
vishnu.pscc71aba2014-10-09 15:26:29 -07002855 struct mm_struct *mm = current->mm;
2856 struct vm_area_struct *vma, *prev;
Kees Cookba093a62016-08-02 14:04:54 -07002857 unsigned long flags, len;
vishnu.pscc71aba2014-10-09 15:26:29 -07002858 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002860 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
Kees Cookba093a62016-08-02 14:04:54 -07002862 len = PAGE_ALIGN(request);
2863 if (len < request)
2864 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002866 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
Kirill Korotaev3a459752006-09-07 14:17:04 +04002868 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2869
Al Viro2c6a1012009-12-03 19:40:46 -05002870 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002871 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002872 return error;
2873
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002874 error = mlock_future_check(mm, mm->def_flags, len);
2875 if (error)
2876 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877
2878 /*
2879 * mm->mmap_sem is required to protect against another thread
2880 * changing the mappings in case we sleep.
2881 */
2882 verify_mm_writelocked(mm);
2883
2884 /*
2885 * Clear old maps. this also does some error checking for us
2886 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002887 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2888 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 if (do_munmap(mm, addr, len))
2890 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 }
2892
2893 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002894 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 return -ENOMEM;
2896
2897 if (mm->map_count > sysctl_max_map_count)
2898 return -ENOMEM;
2899
Al Viro191c5422012-02-13 03:58:52 +00002900 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 return -ENOMEM;
2902
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002904 vma = vma_merge(mm, prev, addr, addr + len, flags,
Colin Cross3e4578f2015-10-27 16:42:08 -07002905 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
Rik van Rielba470de2008-10-18 20:26:50 -07002906 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 goto out;
2908
2909 /*
2910 * create a vma struct for an anonymous mapping
2911 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002912 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 if (!vma) {
2914 vm_unacct_memory(len >> PAGE_SHIFT);
2915 return -ENOMEM;
2916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917
Rik van Riel5beb4932010-03-05 13:42:07 -08002918 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 vma->vm_mm = mm;
2920 vma->vm_start = addr;
2921 vma->vm_end = addr + len;
2922 vma->vm_pgoff = pgoff;
2923 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002924 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 vma_link(mm, vma, prev, rb_link, rb_parent);
2926out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002927 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002929 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002930 if (flags & VM_LOCKED)
2931 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002932 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934}
2935
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002936int vm_brk(unsigned long addr, unsigned long len)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002937{
2938 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002939 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002940 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002941
Michal Hocko2d6c9282016-05-23 16:25:42 -07002942 if (down_write_killable(&mm->mmap_sem))
2943 return -EINTR;
2944
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002945 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002946 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002947 up_write(&mm->mmap_sem);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002948 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002949 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002950 return ret;
2951}
2952EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
2954/* Release all mmaps. */
2955void exit_mmap(struct mm_struct *mm)
2956{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002957 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002958 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 unsigned long nr_accounted = 0;
2960
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002961 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002962 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002963
Rik van Rielba470de2008-10-18 20:26:50 -07002964 if (mm->locked_vm) {
2965 vma = mm->mmap;
2966 while (vma) {
2967 if (vma->vm_flags & VM_LOCKED)
2968 munlock_vma_pages_all(vma);
2969 vma = vma->vm_next;
2970 }
2971 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002972
2973 arch_exit_mmap(mm);
2974
Rik van Rielba470de2008-10-18 20:26:50 -07002975 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002976 if (!vma) /* Can happen if dup_mmap() received an OOM */
2977 return;
2978
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002981 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002982 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002983 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002984 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002985
Hugh Dickins6ee86302013-04-29 15:07:44 -07002986 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002987 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002990 * Walk the list again, actually closing and freeing it,
2991 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002993 while (vma) {
2994 if (vma->vm_flags & VM_ACCOUNT)
2995 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002996 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002997 }
2998 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999}
3000
3001/* Insert vm structure into process list sorted by address
3002 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003003 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 */
Hugh Dickins6597d782012-10-08 16:29:07 -07003005int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006{
Hugh Dickins6597d782012-10-08 16:29:07 -07003007 struct vm_area_struct *prev;
3008 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009
Chen Gangc9d13f52015-09-08 15:04:08 -07003010 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3011 &prev, &rb_link, &rb_parent))
3012 return -ENOMEM;
3013 if ((vma->vm_flags & VM_ACCOUNT) &&
3014 security_vm_enough_memory_mm(mm, vma_pages(vma)))
3015 return -ENOMEM;
3016
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 /*
3018 * The vm_pgoff of a purely anonymous vma should be irrelevant
3019 * until its first write fault, when page's anon_vma and index
3020 * are set. But now set the vm_pgoff it will almost certainly
3021 * end up with (unless mremap moves it elsewhere before that
3022 * first wfault), so /proc/pid/maps tells a consistent story.
3023 *
3024 * By setting it to reflect the virtual start address of the
3025 * vma, merges and splits can happen in a seamless way, just
3026 * using the existing file pgoff checks and manipulations.
3027 * Similarly in do_mmap_pgoff and in do_brk.
3028 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003029 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 BUG_ON(vma->anon_vma);
3031 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3032 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05303033
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 vma_link(mm, vma, prev, rb_link, rb_parent);
3035 return 0;
3036}
3037
3038/*
3039 * Copy the vma structure to a new location in the same mm,
3040 * prior to moving page table entries, to effect an mremap move.
3041 */
3042struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07003043 unsigned long addr, unsigned long len, pgoff_t pgoff,
3044 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045{
3046 struct vm_area_struct *vma = *vmap;
3047 unsigned long vma_start = vma->vm_start;
3048 struct mm_struct *mm = vma->vm_mm;
3049 struct vm_area_struct *new_vma, *prev;
3050 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003051 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052
3053 /*
3054 * If anonymous vma has not yet been faulted, update new pgoff
3055 * to match new location, to increase its chance of merging.
3056 */
Oleg Nesterovce757992015-09-08 14:58:34 -07003057 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003059 faulted_in_anon_vma = false;
3060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061
Hugh Dickins6597d782012-10-08 16:29:07 -07003062 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3063 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07003065 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
Colin Cross3e4578f2015-10-27 16:42:08 -07003066 vma->vm_userfaultfd_ctx, vma_get_anon_name(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 if (new_vma) {
3068 /*
3069 * Source vma may have been merged into new_vma
3070 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003071 if (unlikely(vma_start >= new_vma->vm_start &&
3072 vma_start < new_vma->vm_end)) {
3073 /*
3074 * The only way we can get a vma_merge with
3075 * self during an mremap is if the vma hasn't
3076 * been faulted in yet and we were allowed to
3077 * reset the dst vma->vm_pgoff to the
3078 * destination address of the mremap to allow
3079 * the merge to happen. mremap must change the
3080 * vm_pgoff linearity between src and dst vmas
3081 * (in turn preventing a vma_merge) to be
3082 * safe. It is only safe to keep the vm_pgoff
3083 * linear if there are no pages mapped yet.
3084 */
Sasha Levin81d1b092014-10-09 15:28:10 -07003085 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07003086 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07003087 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07003088 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08003090 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07003091 if (!new_vma)
3092 goto out;
3093 *new_vma = *vma;
3094 new_vma->vm_start = addr;
3095 new_vma->vm_end = addr + len;
3096 new_vma->vm_pgoff = pgoff;
3097 if (vma_dup_policy(vma, new_vma))
3098 goto out_free_vma;
3099 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3100 if (anon_vma_clone(new_vma, vma))
3101 goto out_free_mempol;
3102 if (new_vma->vm_file)
3103 get_file(new_vma->vm_file);
3104 if (new_vma->vm_ops && new_vma->vm_ops->open)
3105 new_vma->vm_ops->open(new_vma);
3106 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3107 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 }
3109 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003110
Chen Gange3975892015-09-08 15:03:38 -07003111out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07003112 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07003113out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08003114 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07003115out:
Rik van Riel5beb4932010-03-05 13:42:07 -08003116 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003118
3119/*
3120 * Return true if the calling process may expand its vm space by the passed
3121 * number of pages
3122 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003123bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003124{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003125 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3126 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003127
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003128 if (is_data_mapping(flags) &&
3129 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003130 /* Workaround for Valgrind */
3131 if (rlimit(RLIMIT_DATA) == 0 &&
3132 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3133 return true;
3134 if (!ignore_rlimit_data) {
3135 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 -08003136 current->comm, current->pid,
3137 (mm->data_vm + npages) << PAGE_SHIFT,
3138 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003139 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003140 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003141 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07003142
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003143 return true;
3144}
3145
3146void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3147{
3148 mm->total_vm += npages;
3149
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003150 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003151 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003152 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003153 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003154 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003155 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003156}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003157
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003158static int special_mapping_fault(struct vm_area_struct *vma,
3159 struct vm_fault *vmf);
3160
3161/*
3162 * Having a close hook prevents vma merging regardless of flags.
3163 */
3164static void special_mapping_close(struct vm_area_struct *vma)
3165{
3166}
3167
3168static const char *special_mapping_name(struct vm_area_struct *vma)
3169{
3170 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3171}
3172
Dmitry Safonovb059a452016-06-28 14:35:38 +03003173static int special_mapping_mremap(struct vm_area_struct *new_vma)
3174{
3175 struct vm_special_mapping *sm = new_vma->vm_private_data;
3176
3177 if (sm->mremap)
3178 return sm->mremap(sm, new_vma);
3179 return 0;
3180}
3181
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003182static const struct vm_operations_struct special_mapping_vmops = {
3183 .close = special_mapping_close,
3184 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03003185 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003186 .name = special_mapping_name,
3187};
3188
3189static const struct vm_operations_struct legacy_special_mapping_vmops = {
3190 .close = special_mapping_close,
3191 .fault = special_mapping_fault,
3192};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003193
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003194static int special_mapping_fault(struct vm_area_struct *vma,
3195 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003196{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003197 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003198 struct page **pages;
3199
Andy Lutomirskif872f542015-12-29 20:12:19 -08003200 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003201 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08003202 } else {
3203 struct vm_special_mapping *sm = vma->vm_private_data;
3204
3205 if (sm->fault)
3206 return sm->fault(sm, vma, vmf);
3207
3208 pages = sm->pages;
3209 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003210
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003211 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003212 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003213
3214 if (*pages) {
3215 struct page *page = *pages;
3216 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003217 vmf->page = page;
3218 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003219 }
3220
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003221 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003222}
3223
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003224static struct vm_area_struct *__install_special_mapping(
3225 struct mm_struct *mm,
3226 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003227 unsigned long vm_flags, void *priv,
3228 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003229{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003230 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003231 struct vm_area_struct *vma;
3232
3233 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3234 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003235 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003236
Rik van Riel5beb4932010-03-05 13:42:07 -08003237 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003238 vma->vm_mm = mm;
3239 vma->vm_start = addr;
3240 vma->vm_end = addr + len;
3241
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003242 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003243 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003244
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003245 vma->vm_ops = ops;
3246 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003247
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003248 ret = insert_vm_struct(mm, vma);
3249 if (ret)
3250 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003251
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003252 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003253
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003254 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003255
Stefani Seibold3935ed62014-03-17 23:22:02 +01003256 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003257
3258out:
3259 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003260 return ERR_PTR(ret);
3261}
3262
Dmitry Safonov2eefd872016-09-05 16:33:05 +03003263bool vma_is_special_mapping(const struct vm_area_struct *vma,
3264 const struct vm_special_mapping *sm)
3265{
3266 return vma->vm_private_data == sm &&
3267 (vma->vm_ops == &special_mapping_vmops ||
3268 vma->vm_ops == &legacy_special_mapping_vmops);
3269}
3270
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003271/*
3272 * Called with mm->mmap_sem held for writing.
3273 * Insert a new vma covering the given region, with the given flags.
3274 * Its pages are supplied by the given array of struct page *.
3275 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3276 * The region past the last page supplied will always produce SIGBUS.
3277 * The array pointer and the pages it points to are assumed to stay alive
3278 * for as long as this mapping might exist.
3279 */
3280struct vm_area_struct *_install_special_mapping(
3281 struct mm_struct *mm,
3282 unsigned long addr, unsigned long len,
3283 unsigned long vm_flags, const struct vm_special_mapping *spec)
3284{
Chen Gang27f28b92015-11-05 18:48:41 -08003285 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3286 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003287}
3288
Stefani Seibold3935ed62014-03-17 23:22:02 +01003289int install_special_mapping(struct mm_struct *mm,
3290 unsigned long addr, unsigned long len,
3291 unsigned long vm_flags, struct page **pages)
3292{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003293 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003294 mm, addr, len, vm_flags, (void *)pages,
3295 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003296
Duan Jiong14bd5b42014-06-04 16:07:05 -07003297 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003298}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003299
3300static DEFINE_MUTEX(mm_all_locks_mutex);
3301
Peter Zijlstra454ed842008-08-11 09:30:25 +02003302static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003303{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003304 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003305 /*
3306 * The LSB of head.next can't change from under us
3307 * because we hold the mm_all_locks_mutex.
3308 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003309 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003310 /*
3311 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003312 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003313 * the same anon_vma we won't take it again.
3314 *
3315 * No need of atomic instructions here, head.next
3316 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003317 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003318 */
3319 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003320 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003321 BUG();
3322 }
3323}
3324
Peter Zijlstra454ed842008-08-11 09:30:25 +02003325static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003326{
3327 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3328 /*
3329 * AS_MM_ALL_LOCKS can't change from under us because
3330 * we hold the mm_all_locks_mutex.
3331 *
3332 * Operations on ->flags have to be atomic because
3333 * even if AS_MM_ALL_LOCKS is stable thanks to the
3334 * mm_all_locks_mutex, there may be other cpus
3335 * changing other bitflags in parallel to us.
3336 */
3337 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3338 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003339 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003340 }
3341}
3342
3343/*
3344 * This operation locks against the VM for all pte/vma/mm related
3345 * operations that could ever happen on a certain mm. This includes
3346 * vmtruncate, try_to_unmap, and all page faults.
3347 *
3348 * The caller must take the mmap_sem in write mode before calling
3349 * mm_take_all_locks(). The caller isn't allowed to release the
3350 * mmap_sem until mm_drop_all_locks() returns.
3351 *
3352 * mmap_sem in write mode is required in order to block all operations
3353 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003354 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003355 * anon_vmas to be associated with existing vmas.
3356 *
3357 * A single task can't take more than one mm_take_all_locks() in a row
3358 * or it would deadlock.
3359 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003360 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003361 * mapping->flags avoid to take the same lock twice, if more than one
3362 * vma in this mm is backed by the same anon_vma or address_space.
3363 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003364 * We take locks in following order, accordingly to comment at beginning
3365 * of mm/rmap.c:
3366 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3367 * hugetlb mapping);
3368 * - all i_mmap_rwsem locks;
3369 * - all anon_vma->rwseml
3370 *
3371 * We can take all locks within these types randomly because the VM code
3372 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3373 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003374 *
3375 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3376 * that may have to take thousand of locks.
3377 *
3378 * mm_take_all_locks() can fail if it's interrupted by signals.
3379 */
3380int mm_take_all_locks(struct mm_struct *mm)
3381{
3382 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003383 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003384
3385 BUG_ON(down_read_trylock(&mm->mmap_sem));
3386
3387 mutex_lock(&mm_all_locks_mutex);
3388
3389 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3390 if (signal_pending(current))
3391 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003392 if (vma->vm_file && vma->vm_file->f_mapping &&
3393 is_vm_hugetlb_page(vma))
3394 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3395 }
3396
3397 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3398 if (signal_pending(current))
3399 goto out_unlock;
3400 if (vma->vm_file && vma->vm_file->f_mapping &&
3401 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003402 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003403 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003404
3405 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3406 if (signal_pending(current))
3407 goto out_unlock;
3408 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003409 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3410 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003411 }
3412
Kautuk Consul584cff52011-10-31 17:08:59 -07003413 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003414
3415out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003416 mm_drop_all_locks(mm);
3417 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003418}
3419
3420static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3421{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003422 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003423 /*
3424 * The LSB of head.next can't change to 0 from under
3425 * us because we hold the mm_all_locks_mutex.
3426 *
3427 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003428 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003429 * never see our bitflag.
3430 *
3431 * No need of atomic instructions here, head.next
3432 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003433 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003434 */
3435 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003436 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003437 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003438 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003439 }
3440}
3441
3442static void vm_unlock_mapping(struct address_space *mapping)
3443{
3444 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3445 /*
3446 * AS_MM_ALL_LOCKS can't change to 0 from under us
3447 * because we hold the mm_all_locks_mutex.
3448 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003449 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003450 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3451 &mapping->flags))
3452 BUG();
3453 }
3454}
3455
3456/*
3457 * The mmap_sem cannot be released by the caller until
3458 * mm_drop_all_locks() returns.
3459 */
3460void mm_drop_all_locks(struct mm_struct *mm)
3461{
3462 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003463 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003464
3465 BUG_ON(down_read_trylock(&mm->mmap_sem));
3466 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3467
3468 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3469 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003470 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3471 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003472 if (vma->vm_file && vma->vm_file->f_mapping)
3473 vm_unlock_mapping(vma->vm_file->f_mapping);
3474 }
3475
3476 mutex_unlock(&mm_all_locks_mutex);
3477}
David Howells8feae132009-01-08 12:04:47 +00003478
3479/*
3480 * initialise the VMA slab
3481 */
3482void __init mmap_init(void)
3483{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003484 int ret;
3485
Tejun Heo908c7f12014-09-08 09:51:29 +09003486 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003487 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003488}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003489
3490/*
3491 * Initialise sysctl_user_reserve_kbytes.
3492 *
3493 * This is intended to prevent a user from starting a single memory hogging
3494 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3495 * mode.
3496 *
3497 * The default value is min(3% of free memory, 128MB)
3498 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3499 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003500static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003501{
3502 unsigned long free_kbytes;
3503
3504 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3505
3506 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3507 return 0;
3508}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003509subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003510
3511/*
3512 * Initialise sysctl_admin_reserve_kbytes.
3513 *
3514 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3515 * to log in and kill a memory hogging process.
3516 *
3517 * Systems with more than 256MB will reserve 8MB, enough to recover
3518 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3519 * only reserve 3% of free pages by default.
3520 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003521static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003522{
3523 unsigned long free_kbytes;
3524
3525 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3526
3527 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3528 return 0;
3529}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003530subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003531
3532/*
3533 * Reinititalise user and admin reserves if memory is added or removed.
3534 *
3535 * The default user reserve max is 128MB, and the default max for the
3536 * admin reserve is 8MB. These are usually, but not always, enough to
3537 * enable recovery from a memory hogging process using login/sshd, a shell,
3538 * and tools like top. It may make sense to increase or even disable the
3539 * reserve depending on the existence of swap or variations in the recovery
3540 * tools. So, the admin may have changed them.
3541 *
3542 * If memory is added and the reserves have been eliminated or increased above
3543 * the default max, then we'll trust the admin.
3544 *
3545 * If memory is removed and there isn't enough free memory, then we
3546 * need to reset the reserves.
3547 *
3548 * Otherwise keep the reserve set by the admin.
3549 */
3550static int reserve_mem_notifier(struct notifier_block *nb,
3551 unsigned long action, void *data)
3552{
3553 unsigned long tmp, free_kbytes;
3554
3555 switch (action) {
3556 case MEM_ONLINE:
3557 /* Default max is 128MB. Leave alone if modified by operator. */
3558 tmp = sysctl_user_reserve_kbytes;
3559 if (0 < tmp && tmp < (1UL << 17))
3560 init_user_reserve();
3561
3562 /* Default max is 8MB. Leave alone if modified by operator. */
3563 tmp = sysctl_admin_reserve_kbytes;
3564 if (0 < tmp && tmp < (1UL << 13))
3565 init_admin_reserve();
3566
3567 break;
3568 case MEM_OFFLINE:
3569 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3570
3571 if (sysctl_user_reserve_kbytes > free_kbytes) {
3572 init_user_reserve();
3573 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3574 sysctl_user_reserve_kbytes);
3575 }
3576
3577 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3578 init_admin_reserve();
3579 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3580 sysctl_admin_reserve_kbytes);
3581 }
3582 break;
3583 default:
3584 break;
3585 }
3586 return NOTIFY_OK;
3587}
3588
3589static struct notifier_block reserve_mem_nb = {
3590 .notifier_call = reserve_mem_notifier,
3591};
3592
3593static int __meminit init_reserve_notifier(void)
3594{
3595 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003596 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003597
3598 return 0;
3599}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003600subsys_initcall(init_reserve_notifier);