blob: 145d3d5253e8cd498b8874017ffc8ef864d94b5c [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,
973 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800975 /*
976 * VM_SOFTDIRTY should not prevent from VMA merging, if we
977 * match the flags but dirty bit -- the caller should mark
978 * merged VMA as dirty. If dirty bit won't be excluded from
979 * comparison, we increase pressue on the memory system forcing
980 * the kernel to generate new VMAs when old one could be
981 * extended instead.
982 */
983 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 return 0;
985 if (vma->vm_file != file)
986 return 0;
987 if (vma->vm_ops && vma->vm_ops->close)
988 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700989 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
990 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 return 1;
992}
993
994static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700995 struct anon_vma *anon_vma2,
996 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
Shaohua Li965f55d2011-05-24 17:11:20 -0700998 /*
999 * The list_is_singular() test is to avoid merging VMA cloned from
1000 * parents. This can improve scalability caused by anon_vma lock.
1001 */
1002 if ((!anon_vma1 || !anon_vma2) && (!vma ||
1003 list_is_singular(&vma->anon_vma_chain)))
1004 return 1;
1005 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
1007
1008/*
1009 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1010 * in front of (at a lower virtual address and file offset than) the vma.
1011 *
1012 * We cannot merge two vmas if they have differently assigned (non-NULL)
1013 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1014 *
1015 * We don't check here for the merged mmap wrapping around the end of pagecache
1016 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1017 * wrap, nor mmaps which cover the final page at index -1UL.
1018 */
1019static int
1020can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001021 struct anon_vma *anon_vma, struct file *file,
1022 pgoff_t vm_pgoff,
1023 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001025 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001026 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 if (vma->vm_pgoff == vm_pgoff)
1028 return 1;
1029 }
1030 return 0;
1031}
1032
1033/*
1034 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1035 * beyond (at a higher virtual address and file offset than) the vma.
1036 *
1037 * We cannot merge two vmas if they have differently assigned (non-NULL)
1038 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1039 */
1040static int
1041can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001042 struct anon_vma *anon_vma, struct file *file,
1043 pgoff_t vm_pgoff,
1044 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001046 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001047 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001049 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1051 return 1;
1052 }
1053 return 0;
1054}
1055
1056/*
1057 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1058 * whether that can be merged with its predecessor or its successor.
1059 * Or both (it neatly fills a hole).
1060 *
1061 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1062 * certain not to be mapped by the time vma_merge is called; but when
1063 * called for mprotect, it is certain to be already mapped (either at
1064 * an offset within prev, or at the start of next), and the flags of
1065 * this area are about to be changed to vm_flags - and the no-change
1066 * case has already been eliminated.
1067 *
1068 * The following mprotect cases have to be considered, where AAAA is
1069 * the area passed down from mprotect_fixup, never extending beyond one
1070 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1071 *
1072 * AAAA AAAA AAAA AAAA
1073 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1074 * cannot merge might become might become might become
1075 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1076 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001077 * mremap move: PPPPXXXXXXXX 8
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 * AAAA
1079 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1080 * might become case 1 below case 2 below case 3 below
1081 *
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001082 * It is important for case 8 that the the vma NNNN overlapping the
1083 * region AAAA is never going to extended over XXXX. Instead XXXX must
1084 * be extended in region AAAA and NNNN must be removed. This way in
1085 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1086 * rmap_locks, the properties of the merged vma will be already
1087 * correct for the whole merged range. Some of those properties like
1088 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1089 * be correct for the whole merged range immediately after the
1090 * rmap_locks are released. Otherwise if XXXX would be removed and
1091 * NNNN would be extended over the XXXX range, remove_migration_ptes
1092 * or other rmap walkers (if working on addresses beyond the "end"
1093 * parameter) may establish ptes with the wrong permissions of NNNN
1094 * instead of the right permissions of XXXX.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 */
1096struct vm_area_struct *vma_merge(struct mm_struct *mm,
1097 struct vm_area_struct *prev, unsigned long addr,
1098 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001099 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001100 pgoff_t pgoff, struct mempolicy *policy,
1101 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102{
1103 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1104 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001105 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
1107 /*
1108 * We later require that vma->vm_flags == vm_flags,
1109 * so this tests vma->vm_flags & VM_SPECIAL, too.
1110 */
1111 if (vm_flags & VM_SPECIAL)
1112 return NULL;
1113
1114 if (prev)
1115 next = prev->vm_next;
1116 else
1117 next = mm->mmap;
1118 area = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001119 if (area && area->vm_end == end) /* cases 6, 7, 8 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 next = next->vm_next;
1121
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001122 /* verify some invariant that must be enforced by the caller */
1123 VM_WARN_ON(prev && addr <= prev->vm_start);
1124 VM_WARN_ON(area && end > area->vm_end);
1125 VM_WARN_ON(addr >= end);
1126
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 /*
1128 * Can it merge with the predecessor?
1129 */
1130 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001131 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001133 anon_vma, file, pgoff,
1134 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 /*
1136 * OK, it can. Can we now merge in the successor as well?
1137 */
1138 if (next && end == next->vm_start &&
1139 mpol_equal(policy, vma_policy(next)) &&
1140 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001141 anon_vma, file,
1142 pgoff+pglen,
1143 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001145 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 /* cases 1, 6 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001147 err = __vma_adjust(prev, prev->vm_start,
1148 next->vm_end, prev->vm_pgoff, NULL,
1149 prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 } else /* cases 2, 5, 7 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001151 err = __vma_adjust(prev, prev->vm_start,
1152 end, prev->vm_pgoff, NULL, prev);
Rik van Riel5beb4932010-03-05 13:42:07 -08001153 if (err)
1154 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001155 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 return prev;
1157 }
1158
1159 /*
1160 * Can this new request be merged in front of next?
1161 */
1162 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001163 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001165 anon_vma, file, pgoff+pglen,
1166 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 if (prev && addr < prev->vm_end) /* case 4 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001168 err = __vma_adjust(prev, prev->vm_start,
1169 addr, prev->vm_pgoff, NULL, next);
1170 else { /* cases 3, 8 */
1171 err = __vma_adjust(area, addr, next->vm_end,
1172 next->vm_pgoff - pglen, NULL, next);
1173 /*
1174 * In case 3 area is already equal to next and
1175 * this is a noop, but in case 8 "area" has
1176 * been removed and next was expanded over it.
1177 */
1178 area = next;
1179 }
Rik van Riel5beb4932010-03-05 13:42:07 -08001180 if (err)
1181 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001182 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 return area;
1184 }
1185
1186 return NULL;
1187}
1188
1189/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001190 * Rough compatbility check to quickly see if it's even worth looking
1191 * at sharing an anon_vma.
1192 *
1193 * They need to have the same vm_file, and the flags can only differ
1194 * in things that mprotect may change.
1195 *
1196 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1197 * we can merge the two vma's. For example, we refuse to merge a vma if
1198 * there is a vm_ops->close() function, because that indicates that the
1199 * driver is doing some kind of reference counting. But that doesn't
1200 * really matter for the anon_vma sharing case.
1201 */
1202static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1203{
1204 return a->vm_end == b->vm_start &&
1205 mpol_equal(vma_policy(a), vma_policy(b)) &&
1206 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001207 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001208 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1209}
1210
1211/*
1212 * Do some basic sanity checking to see if we can re-use the anon_vma
1213 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1214 * the same as 'old', the other will be the new one that is trying
1215 * to share the anon_vma.
1216 *
1217 * NOTE! This runs with mm_sem held for reading, so it is possible that
1218 * the anon_vma of 'old' is concurrently in the process of being set up
1219 * by another page fault trying to merge _that_. But that's ok: if it
1220 * is being set up, that automatically means that it will be a singleton
1221 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001222 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001223 *
1224 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1225 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1226 * is to return an anon_vma that is "complex" due to having gone through
1227 * a fork).
1228 *
1229 * We also make sure that the two vma's are compatible (adjacent,
1230 * and with the same memory policies). That's all stable, even with just
1231 * a read lock on the mm_sem.
1232 */
1233static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1234{
1235 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001236 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001237
1238 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1239 return anon_vma;
1240 }
1241 return NULL;
1242}
1243
1244/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1246 * neighbouring vmas for a suitable anon_vma, before it goes off
1247 * to allocate a new anon_vma. It checks because a repetitive
1248 * sequence of mprotects and faults may otherwise lead to distinct
1249 * anon_vmas being allocated, preventing vma merge in subsequent
1250 * mprotect.
1251 */
1252struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1253{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001254 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
1257 near = vma->vm_next;
1258 if (!near)
1259 goto try_prev;
1260
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001261 anon_vma = reusable_anon_vma(near, vma, near);
1262 if (anon_vma)
1263 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001265 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 if (!near)
1267 goto none;
1268
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001269 anon_vma = reusable_anon_vma(near, near, vma);
1270 if (anon_vma)
1271 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272none:
1273 /*
1274 * There's no absolute need to look only at touching neighbours:
1275 * we could search further afield for "compatible" anon_vmas.
1276 * But it would probably just be a waste of time searching,
1277 * or lead to too many vmas hanging off the same anon_vma.
1278 * We're trying to allow mprotect remerging later on,
1279 * not trying to minimize memory used for anon_vmas.
1280 */
1281 return NULL;
1282}
1283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284/*
Al Viro40401532012-02-13 03:58:52 +00001285 * If a hint addr is less than mmap_min_addr change hint to be as
1286 * low as possible but still greater than mmap_min_addr
1287 */
1288static inline unsigned long round_hint_to_min(unsigned long hint)
1289{
1290 hint &= PAGE_MASK;
1291 if (((void *)hint != NULL) &&
1292 (hint < mmap_min_addr))
1293 return PAGE_ALIGN(mmap_min_addr);
1294 return hint;
1295}
1296
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001297static inline int mlock_future_check(struct mm_struct *mm,
1298 unsigned long flags,
1299 unsigned long len)
1300{
1301 unsigned long locked, lock_limit;
1302
1303 /* mlock MCL_FUTURE? */
1304 if (flags & VM_LOCKED) {
1305 locked = len >> PAGE_SHIFT;
1306 locked += mm->locked_vm;
1307 lock_limit = rlimit(RLIMIT_MEMLOCK);
1308 lock_limit >>= PAGE_SHIFT;
1309 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1310 return -EAGAIN;
1311 }
1312 return 0;
1313}
1314
Al Viro40401532012-02-13 03:58:52 +00001315/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001316 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001318unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001320 unsigned long flags, vm_flags_t vm_flags,
1321 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
vishnu.pscc71aba2014-10-09 15:26:29 -07001323 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001324 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Michel Lespinasse41badc12013-02-22 16:32:47 -08001326 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001327
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001328 if (!len)
1329 return -EINVAL;
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 /*
1332 * Does the application expect PROT_READ to imply PROT_EXEC?
1333 *
1334 * (the exception is when the underlying filesystem is noexec
1335 * mounted, in which case we dont add PROT_EXEC.)
1336 */
1337 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001338 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 prot |= PROT_EXEC;
1340
Eric Paris7cd94142007-11-26 18:47:40 -05001341 if (!(flags & MAP_FIXED))
1342 addr = round_hint_to_min(addr);
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 /* Careful about overflows.. */
1345 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001346 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 return -ENOMEM;
1348
1349 /* offset overflow? */
1350 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001351 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 /* Too many mappings? */
1354 if (mm->map_count > sysctl_max_map_count)
1355 return -ENOMEM;
1356
1357 /* Obtain the address to map to. we verify (or select) it and ensure
1358 * that it represents a valid section of the address space.
1359 */
1360 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001361 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 return addr;
1363
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001364 if (prot == PROT_EXEC) {
1365 pkey = execute_only_pkey(mm);
1366 if (pkey < 0)
1367 pkey = 0;
1368 }
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 /* Do simple checking here so the lower-level routines won't have
1371 * to. we assume access permissions have been handled by the open
1372 * of the memory object, so we don't do any here.
1373 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001374 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1376
Huang Shijiecdf7b342009-09-21 17:03:36 -07001377 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 if (!can_do_mlock())
1379 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001380
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001381 if (mlock_future_check(mm, vm_flags, len))
1382 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001385 struct inode *inode = file_inode(file);
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 switch (flags & MAP_TYPE) {
1388 case MAP_SHARED:
1389 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1390 return -EACCES;
1391
1392 /*
1393 * Make sure we don't allow writing to an append-only
1394 * file..
1395 */
1396 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1397 return -EACCES;
1398
1399 /*
1400 * Make sure there are no mandatory locks on the file.
1401 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001402 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 return -EAGAIN;
1404
1405 vm_flags |= VM_SHARED | VM_MAYSHARE;
1406 if (!(file->f_mode & FMODE_WRITE))
1407 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1408
1409 /* fall through */
1410 case MAP_PRIVATE:
1411 if (!(file->f_mode & FMODE_READ))
1412 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001413 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001414 if (vm_flags & VM_EXEC)
1415 return -EPERM;
1416 vm_flags &= ~VM_MAYEXEC;
1417 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001418
Al Viro72c2d532013-09-22 16:27:52 -04001419 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001420 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001421 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1422 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 break;
1424
1425 default:
1426 return -EINVAL;
1427 }
1428 } else {
1429 switch (flags & MAP_TYPE) {
1430 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001431 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1432 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001433 /*
1434 * Ignore pgoff.
1435 */
1436 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 vm_flags |= VM_SHARED | VM_MAYSHARE;
1438 break;
1439 case MAP_PRIVATE:
1440 /*
1441 * Set pgoff according to addr for anon_vma.
1442 */
1443 pgoff = addr >> PAGE_SHIFT;
1444 break;
1445 default:
1446 return -EINVAL;
1447 }
1448 }
1449
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001450 /*
1451 * Set 'VM_NORESERVE' if we should not account for the
1452 * memory use of this mapping.
1453 */
1454 if (flags & MAP_NORESERVE) {
1455 /* We honor MAP_NORESERVE if allowed to overcommit */
1456 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1457 vm_flags |= VM_NORESERVE;
1458
1459 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1460 if (file && is_file_hugepages(file))
1461 vm_flags |= VM_NORESERVE;
1462 }
1463
1464 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001465 if (!IS_ERR_VALUE(addr) &&
1466 ((vm_flags & VM_LOCKED) ||
1467 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001468 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001469 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001470}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001471
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001472SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1473 unsigned long, prot, unsigned long, flags,
1474 unsigned long, fd, unsigned long, pgoff)
1475{
1476 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001477 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001478
1479 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001480 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001481 file = fget(fd);
1482 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001483 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001484 if (is_file_hugepages(file))
1485 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001486 retval = -EINVAL;
1487 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1488 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001489 } else if (flags & MAP_HUGETLB) {
1490 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001491 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001492
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001493 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001494 if (!hs)
1495 return -EINVAL;
1496
1497 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001498 /*
1499 * VM_NORESERVE is used because the reservations will be
1500 * taken when vm_ops->mmap() is called
1501 * A dummy user value is used because we are not locking
1502 * memory so no accounting is necessary
1503 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001504 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001505 VM_NORESERVE,
1506 &user, HUGETLB_ANONHUGE_INODE,
1507 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001508 if (IS_ERR(file))
1509 return PTR_ERR(file);
1510 }
1511
1512 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1513
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001514 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001515out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001516 if (file)
1517 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001518 return retval;
1519}
1520
Christoph Hellwiga4679372010-03-10 15:21:15 -08001521#ifdef __ARCH_WANT_SYS_OLD_MMAP
1522struct mmap_arg_struct {
1523 unsigned long addr;
1524 unsigned long len;
1525 unsigned long prot;
1526 unsigned long flags;
1527 unsigned long fd;
1528 unsigned long offset;
1529};
1530
1531SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1532{
1533 struct mmap_arg_struct a;
1534
1535 if (copy_from_user(&a, arg, sizeof(a)))
1536 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001537 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001538 return -EINVAL;
1539
1540 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1541 a.offset >> PAGE_SHIFT);
1542}
1543#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1544
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001545/*
1546 * Some shared mappigns will want the pages marked read-only
1547 * to track write events. If so, we'll downgrade vm_page_prot
1548 * to the private version (using protection_map[] without the
1549 * VM_SHARED bit).
1550 */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001551int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001552{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001553 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001554 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001555
1556 /* If it was private or non-writable, the write bit is already clear */
1557 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1558 return 0;
1559
1560 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001561 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001562 return 1;
1563
Peter Feiner64e45502014-10-13 15:55:46 -07001564 /* The open routine did something to the protections that pgprot_modify
1565 * won't preserve? */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001566 if (pgprot_val(vm_page_prot) !=
1567 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001568 return 0;
1569
Peter Feiner64e45502014-10-13 15:55:46 -07001570 /* Do we need to track softdirty? */
1571 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1572 return 1;
1573
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001574 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001575 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001576 return 0;
1577
1578 /* Can the mapping track the dirty pages? */
1579 return vma->vm_file && vma->vm_file->f_mapping &&
1580 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1581}
1582
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001583/*
1584 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001585 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001586 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001587static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001588{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001589 /*
1590 * hugetlb has its own accounting separate from the core VM
1591 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1592 */
1593 if (file && is_file_hugepages(file))
1594 return 0;
1595
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001596 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1597}
1598
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001599unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001600 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001601{
1602 struct mm_struct *mm = current->mm;
1603 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001604 int error;
1605 struct rb_node **rb_link, *rb_parent;
1606 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001607
Cyril Hrubise8420a82013-04-29 15:08:33 -07001608 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001609 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001610 unsigned long nr_pages;
1611
1612 /*
1613 * MAP_FIXED may remove pages of mappings that intersects with
1614 * requested mapping. Account for the pages it would unmap.
1615 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001616 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1617
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001618 if (!may_expand_vm(mm, vm_flags,
1619 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001620 return -ENOMEM;
1621 }
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001624 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1625 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 if (do_munmap(mm, addr, len))
1627 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 }
1629
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001630 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001631 * Private writable mapping: check memory availability
1632 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001633 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001634 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001635 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001636 return -ENOMEM;
1637 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639
1640 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001641 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001643 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1644 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001645 if (vma)
1646 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
1648 /*
1649 * Determine the object being mapped and call the appropriate
1650 * specific mapper. the address has already been validated, but
1651 * not unmapped, but the maps are removed from the list.
1652 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001653 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (!vma) {
1655 error = -ENOMEM;
1656 goto unacct_error;
1657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 vma->vm_mm = mm;
1660 vma->vm_start = addr;
1661 vma->vm_end = addr + len;
1662 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001663 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001665 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 if (vm_flags & VM_DENYWRITE) {
1669 error = deny_write_access(file);
1670 if (error)
1671 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001673 if (vm_flags & VM_SHARED) {
1674 error = mapping_map_writable(file->f_mapping);
1675 if (error)
1676 goto allow_write_and_free_vma;
1677 }
1678
1679 /* ->mmap() can change vma->vm_file, but must guarantee that
1680 * vma_link() below can deny write-access if VM_DENYWRITE is set
1681 * and map writably if VM_SHARED is set. This usually means the
1682 * new file must not have been exposed to user-space, yet.
1683 */
Al Virocb0942b2012-08-27 14:48:26 -04001684 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 error = file->f_op->mmap(file, vma);
1686 if (error)
1687 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001688
1689 /* Can addr have changed??
1690 *
1691 * Answer: Yes, several device drivers can do it in their
1692 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001693 * Bug: If addr is changed, prev, rb_link, rb_parent should
1694 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001695 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001696 WARN_ON_ONCE(addr != vma->vm_start);
1697
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001698 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001699 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 } else if (vm_flags & VM_SHARED) {
1701 error = shmem_zero_setup(vma);
1702 if (error)
1703 goto free_vma;
1704 }
1705
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001706 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001707 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001708 if (file) {
1709 if (vm_flags & VM_SHARED)
1710 mapping_unmap_writable(file->f_mapping);
1711 if (vm_flags & VM_DENYWRITE)
1712 allow_write_access(file);
1713 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001714 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001715out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001716 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001717
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001718 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001720 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1721 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001722 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001723 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001724 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001725 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301726
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001727 if (file)
1728 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301729
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001730 /*
1731 * New (or expanded) vma always get soft dirty status.
1732 * Otherwise user-space soft-dirty page tracker won't
1733 * be able to distinguish situation when vma area unmapped,
1734 * then new mapped in-place (which must be aimed as
1735 * a completely new data area).
1736 */
1737 vma->vm_flags |= VM_SOFTDIRTY;
1738
Peter Feiner64e45502014-10-13 15:55:46 -07001739 vma_set_page_prot(vma);
1740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 return addr;
1742
1743unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 vma->vm_file = NULL;
1745 fput(file);
1746
1747 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001748 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1749 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001750 if (vm_flags & VM_SHARED)
1751 mapping_unmap_writable(file->f_mapping);
1752allow_write_and_free_vma:
1753 if (vm_flags & VM_DENYWRITE)
1754 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755free_vma:
1756 kmem_cache_free(vm_area_cachep, vma);
1757unacct_error:
1758 if (charged)
1759 vm_unacct_memory(charged);
1760 return error;
1761}
1762
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001763unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1764{
1765 /*
1766 * We implement the search by looking for an rbtree node that
1767 * immediately follows a suitable gap. That is,
1768 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1769 * - gap_end = vma->vm_start >= info->low_limit + length;
1770 * - gap_end - gap_start >= length
1771 */
1772
1773 struct mm_struct *mm = current->mm;
1774 struct vm_area_struct *vma;
1775 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1776
1777 /* Adjust search length to account for worst case alignment overhead */
1778 length = info->length + info->align_mask;
1779 if (length < info->length)
1780 return -ENOMEM;
1781
1782 /* Adjust search limits by the desired length */
1783 if (info->high_limit < length)
1784 return -ENOMEM;
1785 high_limit = info->high_limit - length;
1786
1787 if (info->low_limit > high_limit)
1788 return -ENOMEM;
1789 low_limit = info->low_limit + length;
1790
1791 /* Check if rbtree root looks promising */
1792 if (RB_EMPTY_ROOT(&mm->mm_rb))
1793 goto check_highest;
1794 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1795 if (vma->rb_subtree_gap < length)
1796 goto check_highest;
1797
1798 while (true) {
1799 /* Visit left subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001800 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001801 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1802 struct vm_area_struct *left =
1803 rb_entry(vma->vm_rb.rb_left,
1804 struct vm_area_struct, vm_rb);
1805 if (left->rb_subtree_gap >= length) {
1806 vma = left;
1807 continue;
1808 }
1809 }
1810
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001811 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001812check_current:
1813 /* Check if current node has a suitable gap */
1814 if (gap_start > high_limit)
1815 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001816 if (gap_end >= low_limit &&
1817 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001818 goto found;
1819
1820 /* Visit right subtree if it looks promising */
1821 if (vma->vm_rb.rb_right) {
1822 struct vm_area_struct *right =
1823 rb_entry(vma->vm_rb.rb_right,
1824 struct vm_area_struct, vm_rb);
1825 if (right->rb_subtree_gap >= length) {
1826 vma = right;
1827 continue;
1828 }
1829 }
1830
1831 /* Go back up the rbtree to find next candidate node */
1832 while (true) {
1833 struct rb_node *prev = &vma->vm_rb;
1834 if (!rb_parent(prev))
1835 goto check_highest;
1836 vma = rb_entry(rb_parent(prev),
1837 struct vm_area_struct, vm_rb);
1838 if (prev == vma->vm_rb.rb_left) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001839 gap_start = vm_end_gap(vma->vm_prev);
1840 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001841 goto check_current;
1842 }
1843 }
1844 }
1845
1846check_highest:
1847 /* Check highest gap, which does not precede any rbtree node */
1848 gap_start = mm->highest_vm_end;
1849 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1850 if (gap_start > high_limit)
1851 return -ENOMEM;
1852
1853found:
1854 /* We found a suitable gap. Clip it with the original low_limit. */
1855 if (gap_start < info->low_limit)
1856 gap_start = info->low_limit;
1857
1858 /* Adjust gap address to the desired alignment */
1859 gap_start += (info->align_offset - gap_start) & info->align_mask;
1860
1861 VM_BUG_ON(gap_start + info->length > info->high_limit);
1862 VM_BUG_ON(gap_start + info->length > gap_end);
1863 return gap_start;
1864}
1865
1866unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1867{
1868 struct mm_struct *mm = current->mm;
1869 struct vm_area_struct *vma;
1870 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1871
1872 /* Adjust search length to account for worst case alignment overhead */
1873 length = info->length + info->align_mask;
1874 if (length < info->length)
1875 return -ENOMEM;
1876
1877 /*
1878 * Adjust search limits by the desired length.
1879 * See implementation comment at top of unmapped_area().
1880 */
1881 gap_end = info->high_limit;
1882 if (gap_end < length)
1883 return -ENOMEM;
1884 high_limit = gap_end - length;
1885
1886 if (info->low_limit > high_limit)
1887 return -ENOMEM;
1888 low_limit = info->low_limit + length;
1889
1890 /* Check highest gap, which does not precede any rbtree node */
1891 gap_start = mm->highest_vm_end;
1892 if (gap_start <= high_limit)
1893 goto found_highest;
1894
1895 /* Check if rbtree root looks promising */
1896 if (RB_EMPTY_ROOT(&mm->mm_rb))
1897 return -ENOMEM;
1898 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1899 if (vma->rb_subtree_gap < length)
1900 return -ENOMEM;
1901
1902 while (true) {
1903 /* Visit right subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001904 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001905 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1906 struct vm_area_struct *right =
1907 rb_entry(vma->vm_rb.rb_right,
1908 struct vm_area_struct, vm_rb);
1909 if (right->rb_subtree_gap >= length) {
1910 vma = right;
1911 continue;
1912 }
1913 }
1914
1915check_current:
1916 /* Check if current node has a suitable gap */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001917 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001918 if (gap_end < low_limit)
1919 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001920 if (gap_start <= high_limit &&
1921 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001922 goto found;
1923
1924 /* Visit left subtree if it looks promising */
1925 if (vma->vm_rb.rb_left) {
1926 struct vm_area_struct *left =
1927 rb_entry(vma->vm_rb.rb_left,
1928 struct vm_area_struct, vm_rb);
1929 if (left->rb_subtree_gap >= length) {
1930 vma = left;
1931 continue;
1932 }
1933 }
1934
1935 /* Go back up the rbtree to find next candidate node */
1936 while (true) {
1937 struct rb_node *prev = &vma->vm_rb;
1938 if (!rb_parent(prev))
1939 return -ENOMEM;
1940 vma = rb_entry(rb_parent(prev),
1941 struct vm_area_struct, vm_rb);
1942 if (prev == vma->vm_rb.rb_right) {
1943 gap_start = vma->vm_prev ?
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001944 vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001945 goto check_current;
1946 }
1947 }
1948 }
1949
1950found:
1951 /* We found a suitable gap. Clip it with the original high_limit. */
1952 if (gap_end > info->high_limit)
1953 gap_end = info->high_limit;
1954
1955found_highest:
1956 /* Compute highest gap address at the desired alignment */
1957 gap_end -= info->length;
1958 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1959
1960 VM_BUG_ON(gap_end < info->low_limit);
1961 VM_BUG_ON(gap_end < gap_start);
1962 return gap_end;
1963}
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965/* Get an address range which is currently unmapped.
1966 * For shmat() with addr=0.
1967 *
1968 * Ugly calling convention alert:
1969 * Return value with the low bits set means error value,
1970 * ie
1971 * if (ret & ~PAGE_MASK)
1972 * error = ret;
1973 *
1974 * This function "knows" that -ENOMEM has the bits set.
1975 */
1976#ifndef HAVE_ARCH_UNMAPPED_AREA
1977unsigned long
1978arch_get_unmapped_area(struct file *filp, unsigned long addr,
1979 unsigned long len, unsigned long pgoff, unsigned long flags)
1980{
1981 struct mm_struct *mm = current->mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001982 struct vm_area_struct *vma, *prev;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001983 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001985 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 return -ENOMEM;
1987
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001988 if (flags & MAP_FIXED)
1989 return addr;
1990
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 if (addr) {
1992 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001993 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001994 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001995 (!vma || addr + len <= vm_start_gap(vma)) &&
1996 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return addr;
1998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002000 info.flags = 0;
2001 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08002002 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002003 info.high_limit = TASK_SIZE;
2004 info.align_mask = 0;
2005 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006}
vishnu.pscc71aba2014-10-09 15:26:29 -07002007#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009/*
2010 * This mmap-allocator allocates new areas top-down from below the
2011 * stack's low limit (the base):
2012 */
2013#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2014unsigned long
2015arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
2016 const unsigned long len, const unsigned long pgoff,
2017 const unsigned long flags)
2018{
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002019 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002021 unsigned long addr = addr0;
2022 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
2024 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002025 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 return -ENOMEM;
2027
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002028 if (flags & MAP_FIXED)
2029 return addr;
2030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 /* requesting a specific address */
2032 if (addr) {
2033 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002034 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002035 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002036 (!vma || addr + len <= vm_start_gap(vma)) &&
2037 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 return addr;
2039 }
2040
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002041 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2042 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002043 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002044 info.high_limit = mm->mmap_base;
2045 info.align_mask = 0;
2046 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07002047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 /*
2049 * A failed mmap() very likely causes application failure,
2050 * so fall back to the bottom-up function here. This scenario
2051 * can happen with large stack limits and large mmap()
2052 * allocations.
2053 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002054 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002055 VM_BUG_ON(addr != -ENOMEM);
2056 info.flags = 0;
2057 info.low_limit = TASK_UNMAPPED_BASE;
2058 info.high_limit = TASK_SIZE;
2059 addr = vm_unmapped_area(&info);
2060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
2062 return addr;
2063}
2064#endif
2065
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066unsigned long
2067get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2068 unsigned long pgoff, unsigned long flags)
2069{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002070 unsigned long (*get_area)(struct file *, unsigned long,
2071 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
Al Viro9206de92009-12-03 15:23:11 -05002073 unsigned long error = arch_mmap_check(addr, len, flags);
2074 if (error)
2075 return error;
2076
2077 /* Careful about overflows.. */
2078 if (len > TASK_SIZE)
2079 return -ENOMEM;
2080
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002081 get_area = current->mm->get_unmapped_area;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002082 if (file) {
2083 if (file->f_op->get_unmapped_area)
2084 get_area = file->f_op->get_unmapped_area;
2085 } else if (flags & MAP_SHARED) {
2086 /*
2087 * mmap_region() will call shmem_zero_setup() to create a file,
2088 * so use shmem's get_unmapped_area in case it can be huge.
2089 * do_mmap_pgoff() will clear pgoff, so match alignment.
2090 */
2091 pgoff = 0;
2092 get_area = shmem_get_unmapped_area;
2093 }
2094
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002095 addr = get_area(file, addr, len, pgoff, flags);
2096 if (IS_ERR_VALUE(addr))
2097 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002099 if (addr > TASK_SIZE - len)
2100 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002101 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002102 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002103
Al Viro9ac4ed42012-05-30 17:13:15 -04002104 error = security_mmap_addr(addr);
2105 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106}
2107
2108EXPORT_SYMBOL(get_unmapped_area);
2109
2110/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002111struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002113 struct rb_node *rb_node;
2114 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002116 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002117 vma = vmacache_find(mm, addr);
2118 if (likely(vma))
2119 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002121 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002123 while (rb_node) {
2124 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002126 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002127
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002128 if (tmp->vm_end > addr) {
2129 vma = tmp;
2130 if (tmp->vm_start <= addr)
2131 break;
2132 rb_node = rb_node->rb_left;
2133 } else
2134 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002136
2137 if (vma)
2138 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 return vma;
2140}
2141
2142EXPORT_SYMBOL(find_vma);
2143
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002144/*
2145 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002146 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147struct vm_area_struct *
2148find_vma_prev(struct mm_struct *mm, unsigned long addr,
2149 struct vm_area_struct **pprev)
2150{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002151 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002153 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002154 if (vma) {
2155 *pprev = vma->vm_prev;
2156 } else {
2157 struct rb_node *rb_node = mm->mm_rb.rb_node;
2158 *pprev = NULL;
2159 while (rb_node) {
2160 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2161 rb_node = rb_node->rb_right;
2162 }
2163 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002164 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165}
2166
2167/*
2168 * Verify that the stack growth is acceptable and
2169 * update accounting. This is shared with both the
2170 * grow-up and grow-down cases.
2171 */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002172static int acct_stack_growth(struct vm_area_struct *vma,
2173 unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174{
2175 struct mm_struct *mm = vma->vm_mm;
2176 struct rlimit *rlim = current->signal->rlim;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002177 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
2179 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002180 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 return -ENOMEM;
2182
2183 /* Stack limit test */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002184 if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 return -ENOMEM;
2186
2187 /* mlock limit tests */
2188 if (vma->vm_flags & VM_LOCKED) {
2189 unsigned long locked;
2190 unsigned long limit;
2191 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002192 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002193 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 if (locked > limit && !capable(CAP_IPC_LOCK))
2195 return -ENOMEM;
2196 }
2197
Adam Litke0d59a012007-01-30 14:35:39 -08002198 /* Check to ensure the stack will not grow into a hugetlb-only region */
2199 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2200 vma->vm_end - size;
2201 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2202 return -EFAULT;
2203
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 /*
2205 * Overcommit.. This must be the final test, as it will
2206 * update security statistics.
2207 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002208 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 return -ENOMEM;
2210
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 return 0;
2212}
2213
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002214#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002216 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2217 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002219int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220{
Oleg Nesterov09357812015-11-05 18:48:17 -08002221 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002222 struct vm_area_struct *next;
2223 unsigned long gap_addr;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002224 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
2226 if (!(vma->vm_flags & VM_GROWSUP))
2227 return -EFAULT;
2228
Helge Deller5d10ad622017-06-19 17:34:05 +02002229 /* Guard against exceeding limits of the address space. */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002230 address &= PAGE_MASK;
Helge Deller5d10ad622017-06-19 17:34:05 +02002231 if (address >= TASK_SIZE)
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002232 return -ENOMEM;
Helge Deller5d10ad622017-06-19 17:34:05 +02002233 address += PAGE_SIZE;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002234
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002235 /* Enforce stack_guard_gap */
2236 gap_addr = address + stack_guard_gap;
Helge Deller5d10ad622017-06-19 17:34:05 +02002237
2238 /* Guard against overflow */
2239 if (gap_addr < address || gap_addr > TASK_SIZE)
2240 gap_addr = TASK_SIZE;
2241
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002242 next = vma->vm_next;
2243 if (next && next->vm_start < gap_addr) {
2244 if (!(next->vm_flags & VM_GROWSUP))
2245 return -ENOMEM;
2246 /* Check that both stack segments have the same anon_vma? */
2247 }
2248
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002249 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 if (unlikely(anon_vma_prepare(vma)))
2251 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253 /*
2254 * vma->vm_start/vm_end cannot change under us because the caller
2255 * is required to hold the mmap_sem in read mode. We need the
2256 * anon_vma lock to serialize against concurrent expand_stacks.
2257 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002258 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259
2260 /* Somebody else might have raced and expanded it already */
2261 if (address > vma->vm_end) {
2262 unsigned long size, grow;
2263
2264 size = address - vma->vm_start;
2265 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2266
Hugh Dickins42c36f62011-05-09 17:44:42 -07002267 error = -ENOMEM;
2268 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2269 error = acct_stack_growth(vma, size, grow);
2270 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002271 /*
2272 * vma_gap_update() doesn't support concurrent
2273 * updates, but we only hold a shared mmap_sem
2274 * lock here, so we need to protect against
2275 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002276 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002277 * we don't guarantee that all growable vmas
2278 * in a mm share the same root anon vma.
2279 * So, we reuse mm->page_table_lock to guard
2280 * against concurrent vma expansions.
2281 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002282 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002283 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002284 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002285 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002286 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002287 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002288 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002289 if (vma->vm_next)
2290 vma_gap_update(vma->vm_next);
2291 else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002292 mm->highest_vm_end = vm_end_gap(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002293 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002294
Hugh Dickins42c36f62011-05-09 17:44:42 -07002295 perf_event_mmap(vma);
2296 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002299 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002300 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002301 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 return error;
2303}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002304#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2305
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306/*
2307 * vma is the first one with address < vma->vm_start. Have to extend vma.
2308 */
Michal Hockod05f3162011-05-24 17:11:44 -07002309int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002310 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311{
Oleg Nesterov09357812015-11-05 18:48:17 -08002312 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002313 struct vm_area_struct *prev;
2314 unsigned long gap_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 int error;
2316
Eric Paris88694772007-11-26 18:47:26 -05002317 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002318 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002319 if (error)
2320 return error;
2321
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002322 /* Enforce stack_guard_gap */
2323 gap_addr = address - stack_guard_gap;
2324 if (gap_addr > address)
2325 return -ENOMEM;
2326 prev = vma->vm_prev;
2327 if (prev && prev->vm_end > gap_addr) {
2328 if (!(prev->vm_flags & VM_GROWSDOWN))
2329 return -ENOMEM;
2330 /* Check that both stack segments have the same anon_vma? */
2331 }
2332
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002333 /* We must make sure the anon_vma is allocated. */
2334 if (unlikely(anon_vma_prepare(vma)))
2335 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
2337 /*
2338 * vma->vm_start/vm_end cannot change under us because the caller
2339 * is required to hold the mmap_sem in read mode. We need the
2340 * anon_vma lock to serialize against concurrent expand_stacks.
2341 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002342 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
2344 /* Somebody else might have raced and expanded it already */
2345 if (address < vma->vm_start) {
2346 unsigned long size, grow;
2347
2348 size = vma->vm_end - address;
2349 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2350
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002351 error = -ENOMEM;
2352 if (grow <= vma->vm_pgoff) {
2353 error = acct_stack_growth(vma, size, grow);
2354 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002355 /*
2356 * vma_gap_update() doesn't support concurrent
2357 * updates, but we only hold a shared mmap_sem
2358 * lock here, so we need to protect against
2359 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002360 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002361 * we don't guarantee that all growable vmas
2362 * in a mm share the same root anon vma.
2363 * So, we reuse mm->page_table_lock to guard
2364 * against concurrent vma expansions.
2365 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002366 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002367 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002368 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002369 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002370 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002371 vma->vm_start = address;
2372 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002373 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002374 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002375 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002376
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002377 perf_event_mmap(vma);
2378 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 }
2380 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002381 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002382 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002383 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 return error;
2385}
2386
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002387/* enforced gap between the expanding stack and other mappings. */
2388unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2389
2390static int __init cmdline_parse_stack_guard_gap(char *p)
2391{
2392 unsigned long val;
2393 char *endptr;
2394
2395 val = simple_strtoul(p, &endptr, 10);
2396 if (!*endptr)
2397 stack_guard_gap = val << PAGE_SHIFT;
2398
2399 return 0;
2400}
2401__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2402
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002403#ifdef CONFIG_STACK_GROWSUP
2404int expand_stack(struct vm_area_struct *vma, unsigned long address)
2405{
2406 return expand_upwards(vma, address);
2407}
2408
2409struct vm_area_struct *
2410find_extend_vma(struct mm_struct *mm, unsigned long addr)
2411{
2412 struct vm_area_struct *vma, *prev;
2413
2414 addr &= PAGE_MASK;
2415 vma = find_vma_prev(mm, addr, &prev);
2416 if (vma && (vma->vm_start <= addr))
2417 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002418 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002419 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002420 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002421 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002422 return prev;
2423}
2424#else
2425int expand_stack(struct vm_area_struct *vma, unsigned long address)
2426{
2427 return expand_downwards(vma, address);
2428}
2429
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002431find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432{
vishnu.pscc71aba2014-10-09 15:26:29 -07002433 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 unsigned long start;
2435
2436 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002437 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (!vma)
2439 return NULL;
2440 if (vma->vm_start <= addr)
2441 return vma;
2442 if (!(vma->vm_flags & VM_GROWSDOWN))
2443 return NULL;
2444 start = vma->vm_start;
2445 if (expand_stack(vma, addr))
2446 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002447 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002448 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 return vma;
2450}
2451#endif
2452
Jesse Barnese1d6d012014-12-12 16:55:27 -08002453EXPORT_SYMBOL_GPL(find_extend_vma);
2454
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002456 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002458 *
2459 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002461static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002463 unsigned long nr_accounted = 0;
2464
Hugh Dickins365e9c872005-10-29 18:16:18 -07002465 /* Update high watermark before we lower total_vm */
2466 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002468 long nrpages = vma_pages(vma);
2469
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002470 if (vma->vm_flags & VM_ACCOUNT)
2471 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002472 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002473 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002474 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002475 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 validate_mm(mm);
2477}
2478
2479/*
2480 * Get rid of page table information in the indicated region.
2481 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002482 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 */
2484static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002485 struct vm_area_struct *vma, struct vm_area_struct *prev,
2486 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487{
vishnu.pscc71aba2014-10-09 15:26:29 -07002488 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002489 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490
2491 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002492 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002493 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002494 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002495 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002496 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002497 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498}
2499
2500/*
2501 * Create a list of vma's touched by the unmap, removing them from the mm's
2502 * vma list as we go..
2503 */
2504static void
2505detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2506 struct vm_area_struct *prev, unsigned long end)
2507{
2508 struct vm_area_struct **insertion_point;
2509 struct vm_area_struct *tail_vma = NULL;
2510
2511 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002512 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002514 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 mm->map_count--;
2516 tail_vma = vma;
2517 vma = vma->vm_next;
2518 } while (vma && vma->vm_start < end);
2519 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002520 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002521 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002522 vma_gap_update(vma);
2523 } else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002524 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002526
2527 /* Kill the cache */
2528 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529}
2530
2531/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002532 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2533 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002535static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 unsigned long addr, int new_below)
2537{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002539 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
Andi Kleena5516432008-07-23 21:27:41 -07002541 if (is_vm_hugetlb_page(vma) && (addr &
2542 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 return -EINVAL;
2544
Christoph Lametere94b1762006-12-06 20:33:17 -08002545 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002547 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
2549 /* most fields are the same, copy all, and then fixup */
2550 *new = *vma;
2551
Rik van Riel5beb4932010-03-05 13:42:07 -08002552 INIT_LIST_HEAD(&new->anon_vma_chain);
2553
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 if (new_below)
2555 new->vm_end = addr;
2556 else {
2557 new->vm_start = addr;
2558 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2559 }
2560
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002561 err = vma_dup_policy(vma, new);
2562 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002563 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002565 err = anon_vma_clone(new, vma);
2566 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002567 goto out_free_mpol;
2568
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002569 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 get_file(new->vm_file);
2571
2572 if (new->vm_ops && new->vm_ops->open)
2573 new->vm_ops->open(new);
2574
2575 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002576 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2578 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002579 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
Rik van Riel5beb4932010-03-05 13:42:07 -08002581 /* Success. */
2582 if (!err)
2583 return 0;
2584
2585 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002586 if (new->vm_ops && new->vm_ops->close)
2587 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002588 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002589 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002590 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002591 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002592 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002593 out_free_vma:
2594 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002595 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596}
2597
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002598/*
2599 * Split a vma into two pieces at address 'addr', a new vma is allocated
2600 * either for the first part or the tail.
2601 */
2602int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2603 unsigned long addr, int new_below)
2604{
2605 if (mm->map_count >= sysctl_max_map_count)
2606 return -ENOMEM;
2607
2608 return __split_vma(mm, vma, addr, new_below);
2609}
2610
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611/* Munmap is split into 2 main parts -- this part which finds
2612 * what needs doing, and the areas themselves, which do the
2613 * work. This now handles partial unmappings.
2614 * Jeremy Fitzhardinge <jeremy@goop.org>
2615 */
2616int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2617{
2618 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002619 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002621 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 return -EINVAL;
2623
vishnu.pscc71aba2014-10-09 15:26:29 -07002624 len = PAGE_ALIGN(len);
2625 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 return -EINVAL;
2627
2628 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002629 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002630 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002632 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002633 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
2635 /* if it doesn't overlap, we have nothing.. */
2636 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002637 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 return 0;
2639
2640 /*
2641 * If we need to split any vma, do it now to save pain later.
2642 *
2643 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2644 * unmapped vm_area_struct will remain in use: so lower split_vma
2645 * places tmp vma above, and higher split_vma places tmp vma below.
2646 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002647 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002648 int error;
2649
2650 /*
2651 * Make sure that map_count on return from munmap() will
2652 * not exceed its limit; but let map_count go just above
2653 * its limit temporarily, to help free resources as expected.
2654 */
2655 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2656 return -ENOMEM;
2657
2658 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 if (error)
2660 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002661 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 }
2663
2664 /* Does it split the last one? */
2665 last = find_vma(mm, end);
2666 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002667 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 if (error)
2669 return error;
2670 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002671 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
2673 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002674 * unlock any mlock()ed ranges before detaching vmas
2675 */
2676 if (mm->locked_vm) {
2677 struct vm_area_struct *tmp = vma;
2678 while (tmp && tmp->vm_start < end) {
2679 if (tmp->vm_flags & VM_LOCKED) {
2680 mm->locked_vm -= vma_pages(tmp);
2681 munlock_vma_pages_all(tmp);
2682 }
2683 tmp = tmp->vm_next;
2684 }
2685 }
2686
2687 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 * Remove the vma's, and unmap the actual pages
2689 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002690 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2691 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Dave Hansen1de4fa12014-11-14 07:18:31 -08002693 arch_unmap(mm, vma, start, end);
2694
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002696 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697
2698 return 0;
2699}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700
Al Virobfce2812012-04-20 21:57:04 -04002701int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002702{
2703 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002704 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002705
Michal Hockoae798782016-05-23 16:25:33 -07002706 if (down_write_killable(&mm->mmap_sem))
2707 return -EINTR;
2708
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002709 ret = do_munmap(mm, start, len);
2710 up_write(&mm->mmap_sem);
2711 return ret;
2712}
2713EXPORT_SYMBOL(vm_munmap);
2714
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002715SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716{
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002717 int ret;
2718 struct mm_struct *mm = current->mm;
2719
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 profile_munmap(addr);
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002721 if (down_write_killable(&mm->mmap_sem))
2722 return -EINTR;
2723 ret = do_munmap(mm, addr, len);
2724 up_write(&mm->mmap_sem);
2725 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726}
2727
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002728
2729/*
2730 * Emulation of deprecated remap_file_pages() syscall.
2731 */
2732SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2733 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2734{
2735
2736 struct mm_struct *mm = current->mm;
2737 struct vm_area_struct *vma;
2738 unsigned long populate = 0;
2739 unsigned long ret = -EINVAL;
2740 struct file *file;
2741
Joe Perches756a025f02016-03-17 14:19:47 -07002742 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2743 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002744
2745 if (prot)
2746 return ret;
2747 start = start & PAGE_MASK;
2748 size = size & PAGE_MASK;
2749
2750 if (start + size <= start)
2751 return ret;
2752
2753 /* Does pgoff wrap? */
2754 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2755 return ret;
2756
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002757 if (down_write_killable(&mm->mmap_sem))
2758 return -EINTR;
2759
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002760 vma = find_vma(mm, start);
2761
2762 if (!vma || !(vma->vm_flags & VM_SHARED))
2763 goto out;
2764
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002765 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002766 goto out;
2767
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002768 if (start + size > vma->vm_end) {
2769 struct vm_area_struct *next;
2770
2771 for (next = vma->vm_next; next; next = next->vm_next) {
2772 /* hole between vmas ? */
2773 if (next->vm_start != next->vm_prev->vm_end)
2774 goto out;
2775
2776 if (next->vm_file != vma->vm_file)
2777 goto out;
2778
2779 if (next->vm_flags != vma->vm_flags)
2780 goto out;
2781
2782 if (start + size <= next->vm_end)
2783 break;
2784 }
2785
2786 if (!next)
2787 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002788 }
2789
2790 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2791 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2792 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2793
2794 flags &= MAP_NONBLOCK;
2795 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2796 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002797 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002798 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002799
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002800 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002801 for (tmp = vma; tmp->vm_start >= start + size;
2802 tmp = tmp->vm_next) {
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07002803 /*
2804 * Split pmd and munlock page on the border
2805 * of the range.
2806 */
2807 vma_adjust_trans_huge(tmp, start, start + size, 0);
2808
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002809 munlock_vma_pages_range(tmp,
2810 max(tmp->vm_start, start),
2811 min(tmp->vm_end, start + size));
2812 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002813 }
2814
2815 file = get_file(vma->vm_file);
2816 ret = do_mmap_pgoff(vma->vm_file, start, size,
2817 prot, flags, pgoff, &populate);
2818 fput(file);
2819out:
2820 up_write(&mm->mmap_sem);
2821 if (populate)
2822 mm_populate(ret, populate);
2823 if (!IS_ERR_VALUE(ret))
2824 ret = 0;
2825 return ret;
2826}
2827
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828static inline void verify_mm_writelocked(struct mm_struct *mm)
2829{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002830#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2832 WARN_ON(1);
2833 up_read(&mm->mmap_sem);
2834 }
2835#endif
2836}
2837
2838/*
2839 * this is really a simplified "do_mmap". it only handles
2840 * anonymous maps. eventually we may be able to do some
2841 * brk-specific accounting here.
2842 */
Kees Cookba093a62016-08-02 14:04:54 -07002843static int do_brk(unsigned long addr, unsigned long request)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844{
vishnu.pscc71aba2014-10-09 15:26:29 -07002845 struct mm_struct *mm = current->mm;
2846 struct vm_area_struct *vma, *prev;
Kees Cookba093a62016-08-02 14:04:54 -07002847 unsigned long flags, len;
vishnu.pscc71aba2014-10-09 15:26:29 -07002848 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002850 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
Kees Cookba093a62016-08-02 14:04:54 -07002852 len = PAGE_ALIGN(request);
2853 if (len < request)
2854 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002856 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857
Kirill Korotaev3a459752006-09-07 14:17:04 +04002858 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2859
Al Viro2c6a1012009-12-03 19:40:46 -05002860 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002861 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002862 return error;
2863
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002864 error = mlock_future_check(mm, mm->def_flags, len);
2865 if (error)
2866 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
2868 /*
2869 * mm->mmap_sem is required to protect against another thread
2870 * changing the mappings in case we sleep.
2871 */
2872 verify_mm_writelocked(mm);
2873
2874 /*
2875 * Clear old maps. this also does some error checking for us
2876 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002877 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2878 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 if (do_munmap(mm, addr, len))
2880 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 }
2882
2883 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002884 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 return -ENOMEM;
2886
2887 if (mm->map_count > sysctl_max_map_count)
2888 return -ENOMEM;
2889
Al Viro191c5422012-02-13 03:58:52 +00002890 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 return -ENOMEM;
2892
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002894 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002895 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002896 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 goto out;
2898
2899 /*
2900 * create a vma struct for an anonymous mapping
2901 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002902 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 if (!vma) {
2904 vm_unacct_memory(len >> PAGE_SHIFT);
2905 return -ENOMEM;
2906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907
Rik van Riel5beb4932010-03-05 13:42:07 -08002908 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 vma->vm_mm = mm;
2910 vma->vm_start = addr;
2911 vma->vm_end = addr + len;
2912 vma->vm_pgoff = pgoff;
2913 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002914 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 vma_link(mm, vma, prev, rb_link, rb_parent);
2916out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002917 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002919 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002920 if (flags & VM_LOCKED)
2921 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002922 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002923 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924}
2925
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002926int vm_brk(unsigned long addr, unsigned long len)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002927{
2928 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002929 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002930 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002931
Michal Hocko2d6c9282016-05-23 16:25:42 -07002932 if (down_write_killable(&mm->mmap_sem))
2933 return -EINTR;
2934
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002935 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002936 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002937 up_write(&mm->mmap_sem);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002938 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002939 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002940 return ret;
2941}
2942EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943
2944/* Release all mmaps. */
2945void exit_mmap(struct mm_struct *mm)
2946{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002947 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002948 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 unsigned long nr_accounted = 0;
2950
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002951 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002952 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002953
Rik van Rielba470de2008-10-18 20:26:50 -07002954 if (mm->locked_vm) {
2955 vma = mm->mmap;
2956 while (vma) {
2957 if (vma->vm_flags & VM_LOCKED)
2958 munlock_vma_pages_all(vma);
2959 vma = vma->vm_next;
2960 }
2961 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002962
2963 arch_exit_mmap(mm);
2964
Rik van Rielba470de2008-10-18 20:26:50 -07002965 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002966 if (!vma) /* Can happen if dup_mmap() received an OOM */
2967 return;
2968
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002971 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002972 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002973 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002974 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002975
Hugh Dickins6ee86302013-04-29 15:07:44 -07002976 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002977 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002980 * Walk the list again, actually closing and freeing it,
2981 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002983 while (vma) {
2984 if (vma->vm_flags & VM_ACCOUNT)
2985 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002986 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002987 }
2988 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989}
2990
2991/* Insert vm structure into process list sorted by address
2992 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002993 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002995int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996{
Hugh Dickins6597d782012-10-08 16:29:07 -07002997 struct vm_area_struct *prev;
2998 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999
Chen Gangc9d13f52015-09-08 15:04:08 -07003000 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3001 &prev, &rb_link, &rb_parent))
3002 return -ENOMEM;
3003 if ((vma->vm_flags & VM_ACCOUNT) &&
3004 security_vm_enough_memory_mm(mm, vma_pages(vma)))
3005 return -ENOMEM;
3006
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 /*
3008 * The vm_pgoff of a purely anonymous vma should be irrelevant
3009 * until its first write fault, when page's anon_vma and index
3010 * are set. But now set the vm_pgoff it will almost certainly
3011 * end up with (unless mremap moves it elsewhere before that
3012 * first wfault), so /proc/pid/maps tells a consistent story.
3013 *
3014 * By setting it to reflect the virtual start address of the
3015 * vma, merges and splits can happen in a seamless way, just
3016 * using the existing file pgoff checks and manipulations.
3017 * Similarly in do_mmap_pgoff and in do_brk.
3018 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003019 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 BUG_ON(vma->anon_vma);
3021 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3022 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05303023
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 vma_link(mm, vma, prev, rb_link, rb_parent);
3025 return 0;
3026}
3027
3028/*
3029 * Copy the vma structure to a new location in the same mm,
3030 * prior to moving page table entries, to effect an mremap move.
3031 */
3032struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07003033 unsigned long addr, unsigned long len, pgoff_t pgoff,
3034 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035{
3036 struct vm_area_struct *vma = *vmap;
3037 unsigned long vma_start = vma->vm_start;
3038 struct mm_struct *mm = vma->vm_mm;
3039 struct vm_area_struct *new_vma, *prev;
3040 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003041 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042
3043 /*
3044 * If anonymous vma has not yet been faulted, update new pgoff
3045 * to match new location, to increase its chance of merging.
3046 */
Oleg Nesterovce757992015-09-08 14:58:34 -07003047 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003049 faulted_in_anon_vma = false;
3050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051
Hugh Dickins6597d782012-10-08 16:29:07 -07003052 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3053 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07003055 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3056 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 if (new_vma) {
3058 /*
3059 * Source vma may have been merged into new_vma
3060 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003061 if (unlikely(vma_start >= new_vma->vm_start &&
3062 vma_start < new_vma->vm_end)) {
3063 /*
3064 * The only way we can get a vma_merge with
3065 * self during an mremap is if the vma hasn't
3066 * been faulted in yet and we were allowed to
3067 * reset the dst vma->vm_pgoff to the
3068 * destination address of the mremap to allow
3069 * the merge to happen. mremap must change the
3070 * vm_pgoff linearity between src and dst vmas
3071 * (in turn preventing a vma_merge) to be
3072 * safe. It is only safe to keep the vm_pgoff
3073 * linear if there are no pages mapped yet.
3074 */
Sasha Levin81d1b092014-10-09 15:28:10 -07003075 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07003076 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07003077 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07003078 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08003080 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07003081 if (!new_vma)
3082 goto out;
3083 *new_vma = *vma;
3084 new_vma->vm_start = addr;
3085 new_vma->vm_end = addr + len;
3086 new_vma->vm_pgoff = pgoff;
3087 if (vma_dup_policy(vma, new_vma))
3088 goto out_free_vma;
3089 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3090 if (anon_vma_clone(new_vma, vma))
3091 goto out_free_mempol;
3092 if (new_vma->vm_file)
3093 get_file(new_vma->vm_file);
3094 if (new_vma->vm_ops && new_vma->vm_ops->open)
3095 new_vma->vm_ops->open(new_vma);
3096 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3097 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 }
3099 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003100
Chen Gange3975892015-09-08 15:03:38 -07003101out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07003102 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07003103out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08003104 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07003105out:
Rik van Riel5beb4932010-03-05 13:42:07 -08003106 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003108
3109/*
3110 * Return true if the calling process may expand its vm space by the passed
3111 * number of pages
3112 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003113bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003114{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003115 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3116 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003117
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003118 if (is_data_mapping(flags) &&
3119 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003120 /* Workaround for Valgrind */
3121 if (rlimit(RLIMIT_DATA) == 0 &&
3122 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3123 return true;
3124 if (!ignore_rlimit_data) {
3125 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 -08003126 current->comm, current->pid,
3127 (mm->data_vm + npages) << PAGE_SHIFT,
3128 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003129 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003130 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003131 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07003132
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003133 return true;
3134}
3135
3136void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3137{
3138 mm->total_vm += npages;
3139
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003140 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003141 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003142 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003143 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003144 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003145 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003146}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003147
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003148static int special_mapping_fault(struct vm_area_struct *vma,
3149 struct vm_fault *vmf);
3150
3151/*
3152 * Having a close hook prevents vma merging regardless of flags.
3153 */
3154static void special_mapping_close(struct vm_area_struct *vma)
3155{
3156}
3157
3158static const char *special_mapping_name(struct vm_area_struct *vma)
3159{
3160 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3161}
3162
Dmitry Safonovb059a452016-06-28 14:35:38 +03003163static int special_mapping_mremap(struct vm_area_struct *new_vma)
3164{
3165 struct vm_special_mapping *sm = new_vma->vm_private_data;
3166
3167 if (sm->mremap)
3168 return sm->mremap(sm, new_vma);
3169 return 0;
3170}
3171
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003172static const struct vm_operations_struct special_mapping_vmops = {
3173 .close = special_mapping_close,
3174 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03003175 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003176 .name = special_mapping_name,
3177};
3178
3179static const struct vm_operations_struct legacy_special_mapping_vmops = {
3180 .close = special_mapping_close,
3181 .fault = special_mapping_fault,
3182};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003183
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003184static int special_mapping_fault(struct vm_area_struct *vma,
3185 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003186{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003187 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003188 struct page **pages;
3189
Andy Lutomirskif872f542015-12-29 20:12:19 -08003190 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003191 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08003192 } else {
3193 struct vm_special_mapping *sm = vma->vm_private_data;
3194
3195 if (sm->fault)
3196 return sm->fault(sm, vma, vmf);
3197
3198 pages = sm->pages;
3199 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003200
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003201 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003202 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003203
3204 if (*pages) {
3205 struct page *page = *pages;
3206 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003207 vmf->page = page;
3208 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003209 }
3210
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003211 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003212}
3213
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003214static struct vm_area_struct *__install_special_mapping(
3215 struct mm_struct *mm,
3216 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003217 unsigned long vm_flags, void *priv,
3218 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003219{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003220 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003221 struct vm_area_struct *vma;
3222
3223 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3224 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003225 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003226
Rik van Riel5beb4932010-03-05 13:42:07 -08003227 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003228 vma->vm_mm = mm;
3229 vma->vm_start = addr;
3230 vma->vm_end = addr + len;
3231
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003232 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003233 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003234
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003235 vma->vm_ops = ops;
3236 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003237
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003238 ret = insert_vm_struct(mm, vma);
3239 if (ret)
3240 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003241
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003242 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003243
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003244 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003245
Stefani Seibold3935ed62014-03-17 23:22:02 +01003246 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003247
3248out:
3249 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003250 return ERR_PTR(ret);
3251}
3252
Dmitry Safonov2eefd872016-09-05 16:33:05 +03003253bool vma_is_special_mapping(const struct vm_area_struct *vma,
3254 const struct vm_special_mapping *sm)
3255{
3256 return vma->vm_private_data == sm &&
3257 (vma->vm_ops == &special_mapping_vmops ||
3258 vma->vm_ops == &legacy_special_mapping_vmops);
3259}
3260
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003261/*
3262 * Called with mm->mmap_sem held for writing.
3263 * Insert a new vma covering the given region, with the given flags.
3264 * Its pages are supplied by the given array of struct page *.
3265 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3266 * The region past the last page supplied will always produce SIGBUS.
3267 * The array pointer and the pages it points to are assumed to stay alive
3268 * for as long as this mapping might exist.
3269 */
3270struct vm_area_struct *_install_special_mapping(
3271 struct mm_struct *mm,
3272 unsigned long addr, unsigned long len,
3273 unsigned long vm_flags, const struct vm_special_mapping *spec)
3274{
Chen Gang27f28b92015-11-05 18:48:41 -08003275 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3276 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003277}
3278
Stefani Seibold3935ed62014-03-17 23:22:02 +01003279int install_special_mapping(struct mm_struct *mm,
3280 unsigned long addr, unsigned long len,
3281 unsigned long vm_flags, struct page **pages)
3282{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003283 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003284 mm, addr, len, vm_flags, (void *)pages,
3285 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003286
Duan Jiong14bd5b42014-06-04 16:07:05 -07003287 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003288}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003289
3290static DEFINE_MUTEX(mm_all_locks_mutex);
3291
Peter Zijlstra454ed842008-08-11 09:30:25 +02003292static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003293{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003294 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003295 /*
3296 * The LSB of head.next can't change from under us
3297 * because we hold the mm_all_locks_mutex.
3298 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003299 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003300 /*
3301 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003302 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003303 * the same anon_vma we won't take it again.
3304 *
3305 * No need of atomic instructions here, head.next
3306 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003307 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003308 */
3309 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003310 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003311 BUG();
3312 }
3313}
3314
Peter Zijlstra454ed842008-08-11 09:30:25 +02003315static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003316{
3317 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3318 /*
3319 * AS_MM_ALL_LOCKS can't change from under us because
3320 * we hold the mm_all_locks_mutex.
3321 *
3322 * Operations on ->flags have to be atomic because
3323 * even if AS_MM_ALL_LOCKS is stable thanks to the
3324 * mm_all_locks_mutex, there may be other cpus
3325 * changing other bitflags in parallel to us.
3326 */
3327 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3328 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003329 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003330 }
3331}
3332
3333/*
3334 * This operation locks against the VM for all pte/vma/mm related
3335 * operations that could ever happen on a certain mm. This includes
3336 * vmtruncate, try_to_unmap, and all page faults.
3337 *
3338 * The caller must take the mmap_sem in write mode before calling
3339 * mm_take_all_locks(). The caller isn't allowed to release the
3340 * mmap_sem until mm_drop_all_locks() returns.
3341 *
3342 * mmap_sem in write mode is required in order to block all operations
3343 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003344 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003345 * anon_vmas to be associated with existing vmas.
3346 *
3347 * A single task can't take more than one mm_take_all_locks() in a row
3348 * or it would deadlock.
3349 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003350 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003351 * mapping->flags avoid to take the same lock twice, if more than one
3352 * vma in this mm is backed by the same anon_vma or address_space.
3353 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003354 * We take locks in following order, accordingly to comment at beginning
3355 * of mm/rmap.c:
3356 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3357 * hugetlb mapping);
3358 * - all i_mmap_rwsem locks;
3359 * - all anon_vma->rwseml
3360 *
3361 * We can take all locks within these types randomly because the VM code
3362 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3363 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003364 *
3365 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3366 * that may have to take thousand of locks.
3367 *
3368 * mm_take_all_locks() can fail if it's interrupted by signals.
3369 */
3370int mm_take_all_locks(struct mm_struct *mm)
3371{
3372 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003373 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003374
3375 BUG_ON(down_read_trylock(&mm->mmap_sem));
3376
3377 mutex_lock(&mm_all_locks_mutex);
3378
3379 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3380 if (signal_pending(current))
3381 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003382 if (vma->vm_file && vma->vm_file->f_mapping &&
3383 is_vm_hugetlb_page(vma))
3384 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3385 }
3386
3387 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3388 if (signal_pending(current))
3389 goto out_unlock;
3390 if (vma->vm_file && vma->vm_file->f_mapping &&
3391 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003392 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003393 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003394
3395 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3396 if (signal_pending(current))
3397 goto out_unlock;
3398 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003399 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3400 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003401 }
3402
Kautuk Consul584cff52011-10-31 17:08:59 -07003403 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003404
3405out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003406 mm_drop_all_locks(mm);
3407 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003408}
3409
3410static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3411{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003412 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003413 /*
3414 * The LSB of head.next can't change to 0 from under
3415 * us because we hold the mm_all_locks_mutex.
3416 *
3417 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003418 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003419 * never see our bitflag.
3420 *
3421 * No need of atomic instructions here, head.next
3422 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003423 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003424 */
3425 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003426 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003427 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003428 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003429 }
3430}
3431
3432static void vm_unlock_mapping(struct address_space *mapping)
3433{
3434 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3435 /*
3436 * AS_MM_ALL_LOCKS can't change to 0 from under us
3437 * because we hold the mm_all_locks_mutex.
3438 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003439 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003440 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3441 &mapping->flags))
3442 BUG();
3443 }
3444}
3445
3446/*
3447 * The mmap_sem cannot be released by the caller until
3448 * mm_drop_all_locks() returns.
3449 */
3450void mm_drop_all_locks(struct mm_struct *mm)
3451{
3452 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003453 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003454
3455 BUG_ON(down_read_trylock(&mm->mmap_sem));
3456 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3457
3458 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3459 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003460 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3461 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003462 if (vma->vm_file && vma->vm_file->f_mapping)
3463 vm_unlock_mapping(vma->vm_file->f_mapping);
3464 }
3465
3466 mutex_unlock(&mm_all_locks_mutex);
3467}
David Howells8feae132009-01-08 12:04:47 +00003468
3469/*
3470 * initialise the VMA slab
3471 */
3472void __init mmap_init(void)
3473{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003474 int ret;
3475
Tejun Heo908c7f12014-09-08 09:51:29 +09003476 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003477 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003478}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003479
3480/*
3481 * Initialise sysctl_user_reserve_kbytes.
3482 *
3483 * This is intended to prevent a user from starting a single memory hogging
3484 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3485 * mode.
3486 *
3487 * The default value is min(3% of free memory, 128MB)
3488 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3489 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003490static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003491{
3492 unsigned long free_kbytes;
3493
3494 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3495
3496 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3497 return 0;
3498}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003499subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003500
3501/*
3502 * Initialise sysctl_admin_reserve_kbytes.
3503 *
3504 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3505 * to log in and kill a memory hogging process.
3506 *
3507 * Systems with more than 256MB will reserve 8MB, enough to recover
3508 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3509 * only reserve 3% of free pages by default.
3510 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003511static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003512{
3513 unsigned long free_kbytes;
3514
3515 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3516
3517 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3518 return 0;
3519}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003520subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003521
3522/*
3523 * Reinititalise user and admin reserves if memory is added or removed.
3524 *
3525 * The default user reserve max is 128MB, and the default max for the
3526 * admin reserve is 8MB. These are usually, but not always, enough to
3527 * enable recovery from a memory hogging process using login/sshd, a shell,
3528 * and tools like top. It may make sense to increase or even disable the
3529 * reserve depending on the existence of swap or variations in the recovery
3530 * tools. So, the admin may have changed them.
3531 *
3532 * If memory is added and the reserves have been eliminated or increased above
3533 * the default max, then we'll trust the admin.
3534 *
3535 * If memory is removed and there isn't enough free memory, then we
3536 * need to reset the reserves.
3537 *
3538 * Otherwise keep the reserve set by the admin.
3539 */
3540static int reserve_mem_notifier(struct notifier_block *nb,
3541 unsigned long action, void *data)
3542{
3543 unsigned long tmp, free_kbytes;
3544
3545 switch (action) {
3546 case MEM_ONLINE:
3547 /* Default max is 128MB. Leave alone if modified by operator. */
3548 tmp = sysctl_user_reserve_kbytes;
3549 if (0 < tmp && tmp < (1UL << 17))
3550 init_user_reserve();
3551
3552 /* Default max is 8MB. Leave alone if modified by operator. */
3553 tmp = sysctl_admin_reserve_kbytes;
3554 if (0 < tmp && tmp < (1UL << 13))
3555 init_admin_reserve();
3556
3557 break;
3558 case MEM_OFFLINE:
3559 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3560
3561 if (sysctl_user_reserve_kbytes > free_kbytes) {
3562 init_user_reserve();
3563 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3564 sysctl_user_reserve_kbytes);
3565 }
3566
3567 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3568 init_admin_reserve();
3569 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3570 sysctl_admin_reserve_kbytes);
3571 }
3572 break;
3573 default:
3574 break;
3575 }
3576 return NOTIFY_OK;
3577}
3578
3579static struct notifier_block reserve_mem_nb = {
3580 .notifier_call = reserve_mem_notifier,
3581};
3582
3583static int __meminit init_reserve_notifier(void)
3584{
3585 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003586 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003587
3588 return 0;
3589}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003590subsys_initcall(init_reserve_notifier);