blob: aa97074a4a9904655d5e5634c6923d12e5d39e31 [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
Linus Torvalds7a403742018-05-11 09:52:01 -07001315static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1316{
1317 if (S_ISREG(inode->i_mode))
Linus Torvalds4be65292018-05-19 09:29:11 -07001318 return MAX_LFS_FILESIZE;
Linus Torvalds7a403742018-05-11 09:52:01 -07001319
1320 if (S_ISBLK(inode->i_mode))
1321 return MAX_LFS_FILESIZE;
1322
1323 /* Special "we do even unsigned file positions" case */
1324 if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1325 return 0;
1326
1327 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1328 return ULONG_MAX;
1329}
1330
1331static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1332 unsigned long pgoff, unsigned long len)
1333{
1334 u64 maxsize = file_mmap_size_max(file, inode);
1335
1336 if (maxsize && len > maxsize)
1337 return false;
1338 maxsize -= len;
1339 if (pgoff > maxsize >> PAGE_SHIFT)
1340 return false;
1341 return true;
1342}
1343
Al Viro40401532012-02-13 03:58:52 +00001344/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001345 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001347unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001349 unsigned long flags, vm_flags_t vm_flags,
1350 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
vishnu.pscc71aba2014-10-09 15:26:29 -07001352 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001353 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Michel Lespinasse41badc12013-02-22 16:32:47 -08001355 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001356
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001357 if (!len)
1358 return -EINVAL;
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 /*
1361 * Does the application expect PROT_READ to imply PROT_EXEC?
1362 *
1363 * (the exception is when the underlying filesystem is noexec
1364 * mounted, in which case we dont add PROT_EXEC.)
1365 */
1366 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001367 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 prot |= PROT_EXEC;
1369
Eric Paris7cd94142007-11-26 18:47:40 -05001370 if (!(flags & MAP_FIXED))
1371 addr = round_hint_to_min(addr);
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 /* Careful about overflows.. */
1374 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001375 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 return -ENOMEM;
1377
1378 /* offset overflow? */
1379 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001380 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
1382 /* Too many mappings? */
1383 if (mm->map_count > sysctl_max_map_count)
1384 return -ENOMEM;
1385
1386 /* Obtain the address to map to. we verify (or select) it and ensure
1387 * that it represents a valid section of the address space.
1388 */
1389 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001390 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 return addr;
1392
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001393 if (prot == PROT_EXEC) {
1394 pkey = execute_only_pkey(mm);
1395 if (pkey < 0)
1396 pkey = 0;
1397 }
1398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 /* Do simple checking here so the lower-level routines won't have
1400 * to. we assume access permissions have been handled by the open
1401 * of the memory object, so we don't do any here.
1402 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001403 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1405
Huang Shijiecdf7b342009-09-21 17:03:36 -07001406 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (!can_do_mlock())
1408 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001409
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001410 if (mlock_future_check(mm, vm_flags, len))
1411 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001414 struct inode *inode = file_inode(file);
1415
Linus Torvalds7a403742018-05-11 09:52:01 -07001416 if (!file_mmap_ok(file, inode, pgoff, len))
1417 return -EOVERFLOW;
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 switch (flags & MAP_TYPE) {
1420 case MAP_SHARED:
1421 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1422 return -EACCES;
1423
1424 /*
1425 * Make sure we don't allow writing to an append-only
1426 * file..
1427 */
1428 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1429 return -EACCES;
1430
1431 /*
1432 * Make sure there are no mandatory locks on the file.
1433 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001434 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return -EAGAIN;
1436
1437 vm_flags |= VM_SHARED | VM_MAYSHARE;
1438 if (!(file->f_mode & FMODE_WRITE))
1439 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1440
1441 /* fall through */
1442 case MAP_PRIVATE:
1443 if (!(file->f_mode & FMODE_READ))
1444 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001445 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001446 if (vm_flags & VM_EXEC)
1447 return -EPERM;
1448 vm_flags &= ~VM_MAYEXEC;
1449 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001450
Al Viro72c2d532013-09-22 16:27:52 -04001451 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001452 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001453 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1454 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 break;
1456
1457 default:
1458 return -EINVAL;
1459 }
1460 } else {
1461 switch (flags & MAP_TYPE) {
1462 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001463 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1464 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001465 /*
1466 * Ignore pgoff.
1467 */
1468 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 vm_flags |= VM_SHARED | VM_MAYSHARE;
1470 break;
1471 case MAP_PRIVATE:
1472 /*
1473 * Set pgoff according to addr for anon_vma.
1474 */
1475 pgoff = addr >> PAGE_SHIFT;
1476 break;
1477 default:
1478 return -EINVAL;
1479 }
1480 }
1481
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001482 /*
1483 * Set 'VM_NORESERVE' if we should not account for the
1484 * memory use of this mapping.
1485 */
1486 if (flags & MAP_NORESERVE) {
1487 /* We honor MAP_NORESERVE if allowed to overcommit */
1488 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1489 vm_flags |= VM_NORESERVE;
1490
1491 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1492 if (file && is_file_hugepages(file))
1493 vm_flags |= VM_NORESERVE;
1494 }
1495
1496 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001497 if (!IS_ERR_VALUE(addr) &&
1498 ((vm_flags & VM_LOCKED) ||
1499 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001500 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001501 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001502}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001503
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001504SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1505 unsigned long, prot, unsigned long, flags,
1506 unsigned long, fd, unsigned long, pgoff)
1507{
1508 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001509 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001510
1511 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001512 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001513 file = fget(fd);
1514 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001515 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001516 if (is_file_hugepages(file))
1517 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001518 retval = -EINVAL;
1519 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1520 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001521 } else if (flags & MAP_HUGETLB) {
1522 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001523 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001524
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001525 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001526 if (!hs)
1527 return -EINVAL;
1528
1529 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001530 /*
1531 * VM_NORESERVE is used because the reservations will be
1532 * taken when vm_ops->mmap() is called
1533 * A dummy user value is used because we are not locking
1534 * memory so no accounting is necessary
1535 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001536 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001537 VM_NORESERVE,
1538 &user, HUGETLB_ANONHUGE_INODE,
1539 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001540 if (IS_ERR(file))
1541 return PTR_ERR(file);
1542 }
1543
1544 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1545
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001546 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001547out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001548 if (file)
1549 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001550 return retval;
1551}
1552
Christoph Hellwiga4679372010-03-10 15:21:15 -08001553#ifdef __ARCH_WANT_SYS_OLD_MMAP
1554struct mmap_arg_struct {
1555 unsigned long addr;
1556 unsigned long len;
1557 unsigned long prot;
1558 unsigned long flags;
1559 unsigned long fd;
1560 unsigned long offset;
1561};
1562
1563SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1564{
1565 struct mmap_arg_struct a;
1566
1567 if (copy_from_user(&a, arg, sizeof(a)))
1568 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001569 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001570 return -EINVAL;
1571
1572 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1573 a.offset >> PAGE_SHIFT);
1574}
1575#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1576
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001577/*
1578 * Some shared mappigns will want the pages marked read-only
1579 * to track write events. If so, we'll downgrade vm_page_prot
1580 * to the private version (using protection_map[] without the
1581 * VM_SHARED bit).
1582 */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001583int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001584{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001585 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001586 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001587
1588 /* If it was private or non-writable, the write bit is already clear */
1589 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1590 return 0;
1591
1592 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001593 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001594 return 1;
1595
Peter Feiner64e45502014-10-13 15:55:46 -07001596 /* The open routine did something to the protections that pgprot_modify
1597 * won't preserve? */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001598 if (pgprot_val(vm_page_prot) !=
1599 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001600 return 0;
1601
Peter Feiner64e45502014-10-13 15:55:46 -07001602 /* Do we need to track softdirty? */
1603 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1604 return 1;
1605
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001606 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001607 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001608 return 0;
1609
1610 /* Can the mapping track the dirty pages? */
1611 return vma->vm_file && vma->vm_file->f_mapping &&
1612 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1613}
1614
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001615/*
1616 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001617 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001618 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001619static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001620{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001621 /*
1622 * hugetlb has its own accounting separate from the core VM
1623 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1624 */
1625 if (file && is_file_hugepages(file))
1626 return 0;
1627
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001628 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1629}
1630
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001631unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001632 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001633{
1634 struct mm_struct *mm = current->mm;
1635 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001636 int error;
1637 struct rb_node **rb_link, *rb_parent;
1638 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001639
Cyril Hrubise8420a82013-04-29 15:08:33 -07001640 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001641 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001642 unsigned long nr_pages;
1643
1644 /*
1645 * MAP_FIXED may remove pages of mappings that intersects with
1646 * requested mapping. Account for the pages it would unmap.
1647 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001648 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1649
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001650 if (!may_expand_vm(mm, vm_flags,
1651 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001652 return -ENOMEM;
1653 }
1654
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001656 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1657 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 if (do_munmap(mm, addr, len))
1659 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 }
1661
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001662 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001663 * Private writable mapping: check memory availability
1664 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001665 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001666 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001667 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001668 return -ENOMEM;
1669 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
1671
1672 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001673 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001675 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1676 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001677 if (vma)
1678 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
1680 /*
1681 * Determine the object being mapped and call the appropriate
1682 * specific mapper. the address has already been validated, but
1683 * not unmapped, but the maps are removed from the list.
1684 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001685 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (!vma) {
1687 error = -ENOMEM;
1688 goto unacct_error;
1689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691 vma->vm_mm = mm;
1692 vma->vm_start = addr;
1693 vma->vm_end = addr + len;
1694 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001695 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001697 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
1699 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 if (vm_flags & VM_DENYWRITE) {
1701 error = deny_write_access(file);
1702 if (error)
1703 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001705 if (vm_flags & VM_SHARED) {
1706 error = mapping_map_writable(file->f_mapping);
1707 if (error)
1708 goto allow_write_and_free_vma;
1709 }
1710
1711 /* ->mmap() can change vma->vm_file, but must guarantee that
1712 * vma_link() below can deny write-access if VM_DENYWRITE is set
1713 * and map writably if VM_SHARED is set. This usually means the
1714 * new file must not have been exposed to user-space, yet.
1715 */
Al Virocb0942b2012-08-27 14:48:26 -04001716 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 error = file->f_op->mmap(file, vma);
1718 if (error)
1719 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001720
1721 /* Can addr have changed??
1722 *
1723 * Answer: Yes, several device drivers can do it in their
1724 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001725 * Bug: If addr is changed, prev, rb_link, rb_parent should
1726 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001727 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001728 WARN_ON_ONCE(addr != vma->vm_start);
1729
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001730 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001731 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 } else if (vm_flags & VM_SHARED) {
1733 error = shmem_zero_setup(vma);
1734 if (error)
1735 goto free_vma;
1736 }
1737
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001738 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001739 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001740 if (file) {
1741 if (vm_flags & VM_SHARED)
1742 mapping_unmap_writable(file->f_mapping);
1743 if (vm_flags & VM_DENYWRITE)
1744 allow_write_access(file);
1745 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001746 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001747out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001748 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001749
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001750 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001752 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1753 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001754 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001755 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001756 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001757 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301758
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001759 if (file)
1760 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301761
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001762 /*
1763 * New (or expanded) vma always get soft dirty status.
1764 * Otherwise user-space soft-dirty page tracker won't
1765 * be able to distinguish situation when vma area unmapped,
1766 * then new mapped in-place (which must be aimed as
1767 * a completely new data area).
1768 */
1769 vma->vm_flags |= VM_SOFTDIRTY;
1770
Peter Feiner64e45502014-10-13 15:55:46 -07001771 vma_set_page_prot(vma);
1772
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 return addr;
1774
1775unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 vma->vm_file = NULL;
1777 fput(file);
1778
1779 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001780 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1781 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001782 if (vm_flags & VM_SHARED)
1783 mapping_unmap_writable(file->f_mapping);
1784allow_write_and_free_vma:
1785 if (vm_flags & VM_DENYWRITE)
1786 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787free_vma:
1788 kmem_cache_free(vm_area_cachep, vma);
1789unacct_error:
1790 if (charged)
1791 vm_unacct_memory(charged);
1792 return error;
1793}
1794
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001795unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1796{
1797 /*
1798 * We implement the search by looking for an rbtree node that
1799 * immediately follows a suitable gap. That is,
1800 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1801 * - gap_end = vma->vm_start >= info->low_limit + length;
1802 * - gap_end - gap_start >= length
1803 */
1804
1805 struct mm_struct *mm = current->mm;
1806 struct vm_area_struct *vma;
1807 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1808
1809 /* Adjust search length to account for worst case alignment overhead */
1810 length = info->length + info->align_mask;
1811 if (length < info->length)
1812 return -ENOMEM;
1813
1814 /* Adjust search limits by the desired length */
1815 if (info->high_limit < length)
1816 return -ENOMEM;
1817 high_limit = info->high_limit - length;
1818
1819 if (info->low_limit > high_limit)
1820 return -ENOMEM;
1821 low_limit = info->low_limit + length;
1822
1823 /* Check if rbtree root looks promising */
1824 if (RB_EMPTY_ROOT(&mm->mm_rb))
1825 goto check_highest;
1826 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1827 if (vma->rb_subtree_gap < length)
1828 goto check_highest;
1829
1830 while (true) {
1831 /* Visit left subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001832 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001833 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1834 struct vm_area_struct *left =
1835 rb_entry(vma->vm_rb.rb_left,
1836 struct vm_area_struct, vm_rb);
1837 if (left->rb_subtree_gap >= length) {
1838 vma = left;
1839 continue;
1840 }
1841 }
1842
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001843 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001844check_current:
1845 /* Check if current node has a suitable gap */
1846 if (gap_start > high_limit)
1847 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001848 if (gap_end >= low_limit &&
1849 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001850 goto found;
1851
1852 /* Visit right subtree if it looks promising */
1853 if (vma->vm_rb.rb_right) {
1854 struct vm_area_struct *right =
1855 rb_entry(vma->vm_rb.rb_right,
1856 struct vm_area_struct, vm_rb);
1857 if (right->rb_subtree_gap >= length) {
1858 vma = right;
1859 continue;
1860 }
1861 }
1862
1863 /* Go back up the rbtree to find next candidate node */
1864 while (true) {
1865 struct rb_node *prev = &vma->vm_rb;
1866 if (!rb_parent(prev))
1867 goto check_highest;
1868 vma = rb_entry(rb_parent(prev),
1869 struct vm_area_struct, vm_rb);
1870 if (prev == vma->vm_rb.rb_left) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001871 gap_start = vm_end_gap(vma->vm_prev);
1872 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001873 goto check_current;
1874 }
1875 }
1876 }
1877
1878check_highest:
1879 /* Check highest gap, which does not precede any rbtree node */
1880 gap_start = mm->highest_vm_end;
1881 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1882 if (gap_start > high_limit)
1883 return -ENOMEM;
1884
1885found:
1886 /* We found a suitable gap. Clip it with the original low_limit. */
1887 if (gap_start < info->low_limit)
1888 gap_start = info->low_limit;
1889
1890 /* Adjust gap address to the desired alignment */
1891 gap_start += (info->align_offset - gap_start) & info->align_mask;
1892
1893 VM_BUG_ON(gap_start + info->length > info->high_limit);
1894 VM_BUG_ON(gap_start + info->length > gap_end);
1895 return gap_start;
1896}
1897
1898unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1899{
1900 struct mm_struct *mm = current->mm;
1901 struct vm_area_struct *vma;
1902 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1903
1904 /* Adjust search length to account for worst case alignment overhead */
1905 length = info->length + info->align_mask;
1906 if (length < info->length)
1907 return -ENOMEM;
1908
1909 /*
1910 * Adjust search limits by the desired length.
1911 * See implementation comment at top of unmapped_area().
1912 */
1913 gap_end = info->high_limit;
1914 if (gap_end < length)
1915 return -ENOMEM;
1916 high_limit = gap_end - length;
1917
1918 if (info->low_limit > high_limit)
1919 return -ENOMEM;
1920 low_limit = info->low_limit + length;
1921
1922 /* Check highest gap, which does not precede any rbtree node */
1923 gap_start = mm->highest_vm_end;
1924 if (gap_start <= high_limit)
1925 goto found_highest;
1926
1927 /* Check if rbtree root looks promising */
1928 if (RB_EMPTY_ROOT(&mm->mm_rb))
1929 return -ENOMEM;
1930 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1931 if (vma->rb_subtree_gap < length)
1932 return -ENOMEM;
1933
1934 while (true) {
1935 /* Visit right subtree if it looks promising */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001936 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001937 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1938 struct vm_area_struct *right =
1939 rb_entry(vma->vm_rb.rb_right,
1940 struct vm_area_struct, vm_rb);
1941 if (right->rb_subtree_gap >= length) {
1942 vma = right;
1943 continue;
1944 }
1945 }
1946
1947check_current:
1948 /* Check if current node has a suitable gap */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001949 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001950 if (gap_end < low_limit)
1951 return -ENOMEM;
Hugh Dickinsce7fe852017-06-20 02:10:44 -07001952 if (gap_start <= high_limit &&
1953 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001954 goto found;
1955
1956 /* Visit left subtree if it looks promising */
1957 if (vma->vm_rb.rb_left) {
1958 struct vm_area_struct *left =
1959 rb_entry(vma->vm_rb.rb_left,
1960 struct vm_area_struct, vm_rb);
1961 if (left->rb_subtree_gap >= length) {
1962 vma = left;
1963 continue;
1964 }
1965 }
1966
1967 /* Go back up the rbtree to find next candidate node */
1968 while (true) {
1969 struct rb_node *prev = &vma->vm_rb;
1970 if (!rb_parent(prev))
1971 return -ENOMEM;
1972 vma = rb_entry(rb_parent(prev),
1973 struct vm_area_struct, vm_rb);
1974 if (prev == vma->vm_rb.rb_right) {
1975 gap_start = vma->vm_prev ?
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07001976 vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001977 goto check_current;
1978 }
1979 }
1980 }
1981
1982found:
1983 /* We found a suitable gap. Clip it with the original high_limit. */
1984 if (gap_end > info->high_limit)
1985 gap_end = info->high_limit;
1986
1987found_highest:
1988 /* Compute highest gap address at the desired alignment */
1989 gap_end -= info->length;
1990 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1991
1992 VM_BUG_ON(gap_end < info->low_limit);
1993 VM_BUG_ON(gap_end < gap_start);
1994 return gap_end;
1995}
1996
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997/* Get an address range which is currently unmapped.
1998 * For shmat() with addr=0.
1999 *
2000 * Ugly calling convention alert:
2001 * Return value with the low bits set means error value,
2002 * ie
2003 * if (ret & ~PAGE_MASK)
2004 * error = ret;
2005 *
2006 * This function "knows" that -ENOMEM has the bits set.
2007 */
2008#ifndef HAVE_ARCH_UNMAPPED_AREA
2009unsigned long
2010arch_get_unmapped_area(struct file *filp, unsigned long addr,
2011 unsigned long len, unsigned long pgoff, unsigned long flags)
2012{
2013 struct mm_struct *mm = current->mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002014 struct vm_area_struct *vma, *prev;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002015 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002017 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 return -ENOMEM;
2019
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002020 if (flags & MAP_FIXED)
2021 return addr;
2022
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 if (addr) {
2024 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002025 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002026 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002027 (!vma || addr + len <= vm_start_gap(vma)) &&
2028 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 return addr;
2030 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002032 info.flags = 0;
2033 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08002034 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002035 info.high_limit = TASK_SIZE;
2036 info.align_mask = 0;
2037 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038}
vishnu.pscc71aba2014-10-09 15:26:29 -07002039#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041/*
2042 * This mmap-allocator allocates new areas top-down from below the
2043 * stack's low limit (the base):
2044 */
2045#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2046unsigned long
2047arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
2048 const unsigned long len, const unsigned long pgoff,
2049 const unsigned long flags)
2050{
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002051 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002053 unsigned long addr = addr0;
2054 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
2056 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002057 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 return -ENOMEM;
2059
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002060 if (flags & MAP_FIXED)
2061 return addr;
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 /* requesting a specific address */
2064 if (addr) {
2065 addr = PAGE_ALIGN(addr);
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002066 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002067 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002068 (!vma || addr + len <= vm_start_gap(vma)) &&
2069 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 return addr;
2071 }
2072
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002073 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2074 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002075 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002076 info.high_limit = mm->mmap_base;
2077 info.align_mask = 0;
2078 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07002079
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 /*
2081 * A failed mmap() very likely causes application failure,
2082 * so fall back to the bottom-up function here. This scenario
2083 * can happen with large stack limits and large mmap()
2084 * allocations.
2085 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002086 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002087 VM_BUG_ON(addr != -ENOMEM);
2088 info.flags = 0;
2089 info.low_limit = TASK_UNMAPPED_BASE;
2090 info.high_limit = TASK_SIZE;
2091 addr = vm_unmapped_area(&info);
2092 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
2094 return addr;
2095}
2096#endif
2097
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098unsigned long
2099get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2100 unsigned long pgoff, unsigned long flags)
2101{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002102 unsigned long (*get_area)(struct file *, unsigned long,
2103 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Al Viro9206de92009-12-03 15:23:11 -05002105 unsigned long error = arch_mmap_check(addr, len, flags);
2106 if (error)
2107 return error;
2108
2109 /* Careful about overflows.. */
2110 if (len > TASK_SIZE)
2111 return -ENOMEM;
2112
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002113 get_area = current->mm->get_unmapped_area;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002114 if (file) {
2115 if (file->f_op->get_unmapped_area)
2116 get_area = file->f_op->get_unmapped_area;
2117 } else if (flags & MAP_SHARED) {
2118 /*
2119 * mmap_region() will call shmem_zero_setup() to create a file,
2120 * so use shmem's get_unmapped_area in case it can be huge.
2121 * do_mmap_pgoff() will clear pgoff, so match alignment.
2122 */
2123 pgoff = 0;
2124 get_area = shmem_get_unmapped_area;
2125 }
2126
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002127 addr = get_area(file, addr, len, pgoff, flags);
2128 if (IS_ERR_VALUE(addr))
2129 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002131 if (addr > TASK_SIZE - len)
2132 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002133 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002134 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002135
Al Viro9ac4ed42012-05-30 17:13:15 -04002136 error = security_mmap_addr(addr);
2137 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138}
2139
2140EXPORT_SYMBOL(get_unmapped_area);
2141
2142/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002143struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002145 struct rb_node *rb_node;
2146 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002148 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002149 vma = vmacache_find(mm, addr);
2150 if (likely(vma))
2151 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002153 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002155 while (rb_node) {
2156 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002158 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002159
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002160 if (tmp->vm_end > addr) {
2161 vma = tmp;
2162 if (tmp->vm_start <= addr)
2163 break;
2164 rb_node = rb_node->rb_left;
2165 } else
2166 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002168
2169 if (vma)
2170 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 return vma;
2172}
2173
2174EXPORT_SYMBOL(find_vma);
2175
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002176/*
2177 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002178 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179struct vm_area_struct *
2180find_vma_prev(struct mm_struct *mm, unsigned long addr,
2181 struct vm_area_struct **pprev)
2182{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002183 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002185 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002186 if (vma) {
2187 *pprev = vma->vm_prev;
2188 } else {
2189 struct rb_node *rb_node = mm->mm_rb.rb_node;
2190 *pprev = NULL;
2191 while (rb_node) {
2192 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2193 rb_node = rb_node->rb_right;
2194 }
2195 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002196 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197}
2198
2199/*
2200 * Verify that the stack growth is acceptable and
2201 * update accounting. This is shared with both the
2202 * grow-up and grow-down cases.
2203 */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002204static int acct_stack_growth(struct vm_area_struct *vma,
2205 unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206{
2207 struct mm_struct *mm = vma->vm_mm;
2208 struct rlimit *rlim = current->signal->rlim;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002209 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
2211 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002212 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 return -ENOMEM;
2214
2215 /* Stack limit test */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002216 if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 return -ENOMEM;
2218
2219 /* mlock limit tests */
2220 if (vma->vm_flags & VM_LOCKED) {
2221 unsigned long locked;
2222 unsigned long limit;
2223 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002224 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002225 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 if (locked > limit && !capable(CAP_IPC_LOCK))
2227 return -ENOMEM;
2228 }
2229
Adam Litke0d59a012007-01-30 14:35:39 -08002230 /* Check to ensure the stack will not grow into a hugetlb-only region */
2231 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2232 vma->vm_end - size;
2233 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2234 return -EFAULT;
2235
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 /*
2237 * Overcommit.. This must be the final test, as it will
2238 * update security statistics.
2239 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002240 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 return -ENOMEM;
2242
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 return 0;
2244}
2245
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002246#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002248 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2249 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002251int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252{
Oleg Nesterov09357812015-11-05 18:48:17 -08002253 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002254 struct vm_area_struct *next;
2255 unsigned long gap_addr;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002256 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257
2258 if (!(vma->vm_flags & VM_GROWSUP))
2259 return -EFAULT;
2260
Helge Deller5d10ad622017-06-19 17:34:05 +02002261 /* Guard against exceeding limits of the address space. */
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002262 address &= PAGE_MASK;
Helge Deller38dfd2e2017-07-14 14:49:38 -07002263 if (address >= (TASK_SIZE & PAGE_MASK))
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002264 return -ENOMEM;
Helge Deller5d10ad622017-06-19 17:34:05 +02002265 address += PAGE_SIZE;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002266
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002267 /* Enforce stack_guard_gap */
2268 gap_addr = address + stack_guard_gap;
Helge Deller5d10ad622017-06-19 17:34:05 +02002269
2270 /* Guard against overflow */
2271 if (gap_addr < address || gap_addr > TASK_SIZE)
2272 gap_addr = TASK_SIZE;
2273
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002274 next = vma->vm_next;
Michal Hockoc57664b2017-07-10 15:49:51 -07002275 if (next && next->vm_start < gap_addr &&
2276 (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002277 if (!(next->vm_flags & VM_GROWSUP))
2278 return -ENOMEM;
2279 /* Check that both stack segments have the same anon_vma? */
2280 }
2281
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002282 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 if (unlikely(anon_vma_prepare(vma)))
2284 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
2286 /*
2287 * vma->vm_start/vm_end cannot change under us because the caller
2288 * is required to hold the mmap_sem in read mode. We need the
2289 * anon_vma lock to serialize against concurrent expand_stacks.
2290 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002291 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292
2293 /* Somebody else might have raced and expanded it already */
2294 if (address > vma->vm_end) {
2295 unsigned long size, grow;
2296
2297 size = address - vma->vm_start;
2298 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2299
Hugh Dickins42c36f62011-05-09 17:44:42 -07002300 error = -ENOMEM;
2301 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2302 error = acct_stack_growth(vma, size, grow);
2303 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002304 /*
2305 * vma_gap_update() doesn't support concurrent
2306 * updates, but we only hold a shared mmap_sem
2307 * lock here, so we need to protect against
2308 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002309 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002310 * we don't guarantee that all growable vmas
2311 * in a mm share the same root anon vma.
2312 * So, we reuse mm->page_table_lock to guard
2313 * against concurrent vma expansions.
2314 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002315 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002316 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002317 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002318 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002319 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002320 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002321 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002322 if (vma->vm_next)
2323 vma_gap_update(vma->vm_next);
2324 else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002325 mm->highest_vm_end = vm_end_gap(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002326 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002327
Hugh Dickins42c36f62011-05-09 17:44:42 -07002328 perf_event_mmap(vma);
2329 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002332 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002333 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002334 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 return error;
2336}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002337#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2338
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339/*
2340 * vma is the first one with address < vma->vm_start. Have to extend vma.
2341 */
Michal Hockod05f3162011-05-24 17:11:44 -07002342int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002343 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344{
Oleg Nesterov09357812015-11-05 18:48:17 -08002345 struct mm_struct *mm = vma->vm_mm;
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002346 struct vm_area_struct *prev;
2347 unsigned long gap_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 int error;
2349
Eric Paris88694772007-11-26 18:47:26 -05002350 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002351 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002352 if (error)
2353 return error;
2354
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002355 /* Enforce stack_guard_gap */
2356 gap_addr = address - stack_guard_gap;
2357 if (gap_addr > address)
2358 return -ENOMEM;
2359 prev = vma->vm_prev;
Michal Hockoc57664b2017-07-10 15:49:51 -07002360 if (prev && prev->vm_end > gap_addr &&
2361 (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002362 if (!(prev->vm_flags & VM_GROWSDOWN))
2363 return -ENOMEM;
2364 /* Check that both stack segments have the same anon_vma? */
2365 }
2366
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002367 /* We must make sure the anon_vma is allocated. */
2368 if (unlikely(anon_vma_prepare(vma)))
2369 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370
2371 /*
2372 * vma->vm_start/vm_end cannot change under us because the caller
2373 * is required to hold the mmap_sem in read mode. We need the
2374 * anon_vma lock to serialize against concurrent expand_stacks.
2375 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002376 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
2378 /* Somebody else might have raced and expanded it already */
2379 if (address < vma->vm_start) {
2380 unsigned long size, grow;
2381
2382 size = vma->vm_end - address;
2383 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2384
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002385 error = -ENOMEM;
2386 if (grow <= vma->vm_pgoff) {
2387 error = acct_stack_growth(vma, size, grow);
2388 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002389 /*
2390 * vma_gap_update() doesn't support concurrent
2391 * updates, but we only hold a shared mmap_sem
2392 * lock here, so we need to protect against
2393 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002394 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002395 * we don't guarantee that all growable vmas
2396 * in a mm share the same root anon vma.
2397 * So, we reuse mm->page_table_lock to guard
2398 * against concurrent vma expansions.
2399 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002400 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002401 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002402 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002403 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002404 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002405 vma->vm_start = address;
2406 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002407 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002408 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002409 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002410
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002411 perf_event_mmap(vma);
2412 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
2414 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002415 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002416 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002417 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 return error;
2419}
2420
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002421/* enforced gap between the expanding stack and other mappings. */
2422unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2423
2424static int __init cmdline_parse_stack_guard_gap(char *p)
2425{
2426 unsigned long val;
2427 char *endptr;
2428
2429 val = simple_strtoul(p, &endptr, 10);
2430 if (!*endptr)
2431 stack_guard_gap = val << PAGE_SHIFT;
2432
2433 return 0;
2434}
2435__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2436
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002437#ifdef CONFIG_STACK_GROWSUP
2438int expand_stack(struct vm_area_struct *vma, unsigned long address)
2439{
2440 return expand_upwards(vma, address);
2441}
2442
2443struct vm_area_struct *
2444find_extend_vma(struct mm_struct *mm, unsigned long addr)
2445{
2446 struct vm_area_struct *vma, *prev;
2447
2448 addr &= PAGE_MASK;
2449 vma = find_vma_prev(mm, addr, &prev);
2450 if (vma && (vma->vm_start <= addr))
2451 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002452 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002453 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002454 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002455 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002456 return prev;
2457}
2458#else
2459int expand_stack(struct vm_area_struct *vma, unsigned long address)
2460{
2461 return expand_downwards(vma, address);
2462}
2463
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002465find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
vishnu.pscc71aba2014-10-09 15:26:29 -07002467 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 unsigned long start;
2469
2470 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002471 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 if (!vma)
2473 return NULL;
2474 if (vma->vm_start <= addr)
2475 return vma;
2476 if (!(vma->vm_flags & VM_GROWSDOWN))
2477 return NULL;
2478 start = vma->vm_start;
2479 if (expand_stack(vma, addr))
2480 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002481 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002482 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 return vma;
2484}
2485#endif
2486
Jesse Barnese1d6d012014-12-12 16:55:27 -08002487EXPORT_SYMBOL_GPL(find_extend_vma);
2488
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002490 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002492 *
2493 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002495static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002497 unsigned long nr_accounted = 0;
2498
Hugh Dickins365e9c872005-10-29 18:16:18 -07002499 /* Update high watermark before we lower total_vm */
2500 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002502 long nrpages = vma_pages(vma);
2503
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002504 if (vma->vm_flags & VM_ACCOUNT)
2505 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002506 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002507 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002508 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002509 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 validate_mm(mm);
2511}
2512
2513/*
2514 * Get rid of page table information in the indicated region.
2515 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002516 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 */
2518static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002519 struct vm_area_struct *vma, struct vm_area_struct *prev,
2520 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521{
vishnu.pscc71aba2014-10-09 15:26:29 -07002522 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002523 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
2525 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002526 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002527 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002528 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002529 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002530 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002531 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532}
2533
2534/*
2535 * Create a list of vma's touched by the unmap, removing them from the mm's
2536 * vma list as we go..
2537 */
2538static void
2539detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2540 struct vm_area_struct *prev, unsigned long end)
2541{
2542 struct vm_area_struct **insertion_point;
2543 struct vm_area_struct *tail_vma = NULL;
2544
2545 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002546 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002548 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 mm->map_count--;
2550 tail_vma = vma;
2551 vma = vma->vm_next;
2552 } while (vma && vma->vm_start < end);
2553 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002554 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002555 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002556 vma_gap_update(vma);
2557 } else
Hugh Dickinscfc0eb402017-06-19 04:03:24 -07002558 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002560
2561 /* Kill the cache */
2562 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563}
2564
2565/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002566 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2567 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002569static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 unsigned long addr, int new_below)
2571{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002573 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574
Dan Williamscebe1392017-11-29 16:10:28 -08002575 if (vma->vm_ops && vma->vm_ops->split) {
2576 err = vma->vm_ops->split(vma, addr);
2577 if (err)
2578 return err;
2579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
Christoph Lametere94b1762006-12-06 20:33:17 -08002581 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002583 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
2585 /* most fields are the same, copy all, and then fixup */
2586 *new = *vma;
2587
Rik van Riel5beb4932010-03-05 13:42:07 -08002588 INIT_LIST_HEAD(&new->anon_vma_chain);
2589
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 if (new_below)
2591 new->vm_end = addr;
2592 else {
2593 new->vm_start = addr;
2594 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2595 }
2596
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002597 err = vma_dup_policy(vma, new);
2598 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002599 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002601 err = anon_vma_clone(new, vma);
2602 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002603 goto out_free_mpol;
2604
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002605 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 get_file(new->vm_file);
2607
2608 if (new->vm_ops && new->vm_ops->open)
2609 new->vm_ops->open(new);
2610
2611 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002612 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2614 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002615 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Rik van Riel5beb4932010-03-05 13:42:07 -08002617 /* Success. */
2618 if (!err)
2619 return 0;
2620
2621 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002622 if (new->vm_ops && new->vm_ops->close)
2623 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002624 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002625 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002626 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002627 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002628 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002629 out_free_vma:
2630 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002631 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632}
2633
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002634/*
2635 * Split a vma into two pieces at address 'addr', a new vma is allocated
2636 * either for the first part or the tail.
2637 */
2638int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2639 unsigned long addr, int new_below)
2640{
2641 if (mm->map_count >= sysctl_max_map_count)
2642 return -ENOMEM;
2643
2644 return __split_vma(mm, vma, addr, new_below);
2645}
2646
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647/* Munmap is split into 2 main parts -- this part which finds
2648 * what needs doing, and the areas themselves, which do the
2649 * work. This now handles partial unmappings.
2650 * Jeremy Fitzhardinge <jeremy@goop.org>
2651 */
2652int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2653{
2654 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002655 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002657 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 return -EINVAL;
2659
vishnu.pscc71aba2014-10-09 15:26:29 -07002660 len = PAGE_ALIGN(len);
2661 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 return -EINVAL;
2663
2664 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002665 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002666 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002668 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002669 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
2671 /* if it doesn't overlap, we have nothing.. */
2672 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002673 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return 0;
2675
2676 /*
2677 * If we need to split any vma, do it now to save pain later.
2678 *
2679 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2680 * unmapped vm_area_struct will remain in use: so lower split_vma
2681 * places tmp vma above, and higher split_vma places tmp vma below.
2682 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002683 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002684 int error;
2685
2686 /*
2687 * Make sure that map_count on return from munmap() will
2688 * not exceed its limit; but let map_count go just above
2689 * its limit temporarily, to help free resources as expected.
2690 */
2691 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2692 return -ENOMEM;
2693
2694 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 if (error)
2696 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002697 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 }
2699
2700 /* Does it split the last one? */
2701 last = find_vma(mm, end);
2702 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002703 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 if (error)
2705 return error;
2706 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002707 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
2709 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002710 * unlock any mlock()ed ranges before detaching vmas
2711 */
2712 if (mm->locked_vm) {
2713 struct vm_area_struct *tmp = vma;
2714 while (tmp && tmp->vm_start < end) {
2715 if (tmp->vm_flags & VM_LOCKED) {
2716 mm->locked_vm -= vma_pages(tmp);
2717 munlock_vma_pages_all(tmp);
2718 }
2719 tmp = tmp->vm_next;
2720 }
2721 }
2722
2723 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 * Remove the vma's, and unmap the actual pages
2725 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002726 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2727 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
Dave Hansen1de4fa12014-11-14 07:18:31 -08002729 arch_unmap(mm, vma, start, end);
2730
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002732 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733
2734 return 0;
2735}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736
Al Virobfce2812012-04-20 21:57:04 -04002737int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002738{
2739 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002740 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002741
Michal Hockoae798782016-05-23 16:25:33 -07002742 if (down_write_killable(&mm->mmap_sem))
2743 return -EINTR;
2744
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002745 ret = do_munmap(mm, start, len);
2746 up_write(&mm->mmap_sem);
2747 return ret;
2748}
2749EXPORT_SYMBOL(vm_munmap);
2750
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002751SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752{
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002753 int ret;
2754 struct mm_struct *mm = current->mm;
2755
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 profile_munmap(addr);
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002757 if (down_write_killable(&mm->mmap_sem))
2758 return -EINTR;
2759 ret = do_munmap(mm, addr, len);
2760 up_write(&mm->mmap_sem);
2761 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762}
2763
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002764
2765/*
2766 * Emulation of deprecated remap_file_pages() syscall.
2767 */
2768SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2769 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2770{
2771
2772 struct mm_struct *mm = current->mm;
2773 struct vm_area_struct *vma;
2774 unsigned long populate = 0;
2775 unsigned long ret = -EINVAL;
2776 struct file *file;
2777
Joe Perches756a025f02016-03-17 14:19:47 -07002778 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2779 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002780
2781 if (prot)
2782 return ret;
2783 start = start & PAGE_MASK;
2784 size = size & PAGE_MASK;
2785
2786 if (start + size <= start)
2787 return ret;
2788
2789 /* Does pgoff wrap? */
2790 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2791 return ret;
2792
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002793 if (down_write_killable(&mm->mmap_sem))
2794 return -EINTR;
2795
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002796 vma = find_vma(mm, start);
2797
2798 if (!vma || !(vma->vm_flags & VM_SHARED))
2799 goto out;
2800
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002801 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002802 goto out;
2803
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002804 if (start + size > vma->vm_end) {
2805 struct vm_area_struct *next;
2806
2807 for (next = vma->vm_next; next; next = next->vm_next) {
2808 /* hole between vmas ? */
2809 if (next->vm_start != next->vm_prev->vm_end)
2810 goto out;
2811
2812 if (next->vm_file != vma->vm_file)
2813 goto out;
2814
2815 if (next->vm_flags != vma->vm_flags)
2816 goto out;
2817
2818 if (start + size <= next->vm_end)
2819 break;
2820 }
2821
2822 if (!next)
2823 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002824 }
2825
2826 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2827 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2828 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2829
2830 flags &= MAP_NONBLOCK;
2831 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2832 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002833 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002834 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002835
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002836 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002837 for (tmp = vma; tmp->vm_start >= start + size;
2838 tmp = tmp->vm_next) {
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07002839 /*
2840 * Split pmd and munlock page on the border
2841 * of the range.
2842 */
2843 vma_adjust_trans_huge(tmp, start, start + size, 0);
2844
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002845 munlock_vma_pages_range(tmp,
2846 max(tmp->vm_start, start),
2847 min(tmp->vm_end, start + size));
2848 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002849 }
2850
2851 file = get_file(vma->vm_file);
2852 ret = do_mmap_pgoff(vma->vm_file, start, size,
2853 prot, flags, pgoff, &populate);
2854 fput(file);
2855out:
2856 up_write(&mm->mmap_sem);
2857 if (populate)
2858 mm_populate(ret, populate);
2859 if (!IS_ERR_VALUE(ret))
2860 ret = 0;
2861 return ret;
2862}
2863
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864static inline void verify_mm_writelocked(struct mm_struct *mm)
2865{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002866#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2868 WARN_ON(1);
2869 up_read(&mm->mmap_sem);
2870 }
2871#endif
2872}
2873
2874/*
2875 * this is really a simplified "do_mmap". it only handles
2876 * anonymous maps. eventually we may be able to do some
2877 * brk-specific accounting here.
2878 */
Kees Cookba093a62016-08-02 14:04:54 -07002879static int do_brk(unsigned long addr, unsigned long request)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880{
vishnu.pscc71aba2014-10-09 15:26:29 -07002881 struct mm_struct *mm = current->mm;
2882 struct vm_area_struct *vma, *prev;
Kees Cookba093a62016-08-02 14:04:54 -07002883 unsigned long flags, len;
vishnu.pscc71aba2014-10-09 15:26:29 -07002884 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002886 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887
Kees Cookba093a62016-08-02 14:04:54 -07002888 len = PAGE_ALIGN(request);
2889 if (len < request)
2890 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002892 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893
Kirill Korotaev3a459752006-09-07 14:17:04 +04002894 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2895
Al Viro2c6a1012009-12-03 19:40:46 -05002896 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002897 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002898 return error;
2899
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002900 error = mlock_future_check(mm, mm->def_flags, len);
2901 if (error)
2902 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903
2904 /*
2905 * mm->mmap_sem is required to protect against another thread
2906 * changing the mappings in case we sleep.
2907 */
2908 verify_mm_writelocked(mm);
2909
2910 /*
2911 * Clear old maps. this also does some error checking for us
2912 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002913 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2914 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 if (do_munmap(mm, addr, len))
2916 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 }
2918
2919 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002920 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 return -ENOMEM;
2922
2923 if (mm->map_count > sysctl_max_map_count)
2924 return -ENOMEM;
2925
Al Viro191c5422012-02-13 03:58:52 +00002926 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 return -ENOMEM;
2928
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002930 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002931 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002932 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 goto out;
2934
2935 /*
2936 * create a vma struct for an anonymous mapping
2937 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002938 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 if (!vma) {
2940 vm_unacct_memory(len >> PAGE_SHIFT);
2941 return -ENOMEM;
2942 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943
Rik van Riel5beb4932010-03-05 13:42:07 -08002944 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 vma->vm_mm = mm;
2946 vma->vm_start = addr;
2947 vma->vm_end = addr + len;
2948 vma->vm_pgoff = pgoff;
2949 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002950 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 vma_link(mm, vma, prev, rb_link, rb_parent);
2952out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002953 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002955 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002956 if (flags & VM_LOCKED)
2957 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002958 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002959 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960}
2961
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002962int vm_brk(unsigned long addr, unsigned long len)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002963{
2964 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002965 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002966 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002967
Michal Hocko2d6c9282016-05-23 16:25:42 -07002968 if (down_write_killable(&mm->mmap_sem))
2969 return -EINTR;
2970
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002971 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002972 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002973 up_write(&mm->mmap_sem);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002974 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002975 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002976 return ret;
2977}
2978EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979
2980/* Release all mmaps. */
2981void exit_mmap(struct mm_struct *mm)
2982{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002983 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002984 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 unsigned long nr_accounted = 0;
2986
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002987 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002988 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002989
Rik van Rielba470de2008-10-18 20:26:50 -07002990 if (mm->locked_vm) {
2991 vma = mm->mmap;
2992 while (vma) {
2993 if (vma->vm_flags & VM_LOCKED)
2994 munlock_vma_pages_all(vma);
2995 vma = vma->vm_next;
2996 }
2997 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002998
2999 arch_exit_mmap(mm);
3000
Rik van Rielba470de2008-10-18 20:26:50 -07003001 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08003002 if (!vma) /* Can happen if dup_mmap() received an OOM */
3003 return;
3004
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07003007 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08003008 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07003009 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003010 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07003011
Hugh Dickins6ee86302013-04-29 15:07:44 -07003012 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05003013 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07003016 * Walk the list again, actually closing and freeing it,
3017 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003019 while (vma) {
3020 if (vma->vm_flags & VM_ACCOUNT)
3021 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07003022 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003023 }
3024 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025}
3026
3027/* Insert vm structure into process list sorted by address
3028 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003029 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 */
Hugh Dickins6597d782012-10-08 16:29:07 -07003031int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032{
Hugh Dickins6597d782012-10-08 16:29:07 -07003033 struct vm_area_struct *prev;
3034 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035
Chen Gangc9d13f52015-09-08 15:04:08 -07003036 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3037 &prev, &rb_link, &rb_parent))
3038 return -ENOMEM;
3039 if ((vma->vm_flags & VM_ACCOUNT) &&
3040 security_vm_enough_memory_mm(mm, vma_pages(vma)))
3041 return -ENOMEM;
3042
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 /*
3044 * The vm_pgoff of a purely anonymous vma should be irrelevant
3045 * until its first write fault, when page's anon_vma and index
3046 * are set. But now set the vm_pgoff it will almost certainly
3047 * end up with (unless mremap moves it elsewhere before that
3048 * first wfault), so /proc/pid/maps tells a consistent story.
3049 *
3050 * By setting it to reflect the virtual start address of the
3051 * vma, merges and splits can happen in a seamless way, just
3052 * using the existing file pgoff checks and manipulations.
3053 * Similarly in do_mmap_pgoff and in do_brk.
3054 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003055 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 BUG_ON(vma->anon_vma);
3057 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3058 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05303059
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 vma_link(mm, vma, prev, rb_link, rb_parent);
3061 return 0;
3062}
3063
3064/*
3065 * Copy the vma structure to a new location in the same mm,
3066 * prior to moving page table entries, to effect an mremap move.
3067 */
3068struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07003069 unsigned long addr, unsigned long len, pgoff_t pgoff,
3070 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071{
3072 struct vm_area_struct *vma = *vmap;
3073 unsigned long vma_start = vma->vm_start;
3074 struct mm_struct *mm = vma->vm_mm;
3075 struct vm_area_struct *new_vma, *prev;
3076 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003077 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078
3079 /*
3080 * If anonymous vma has not yet been faulted, update new pgoff
3081 * to match new location, to increase its chance of merging.
3082 */
Oleg Nesterovce757992015-09-08 14:58:34 -07003083 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003085 faulted_in_anon_vma = false;
3086 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087
Hugh Dickins6597d782012-10-08 16:29:07 -07003088 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3089 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07003091 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3092 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 if (new_vma) {
3094 /*
3095 * Source vma may have been merged into new_vma
3096 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003097 if (unlikely(vma_start >= new_vma->vm_start &&
3098 vma_start < new_vma->vm_end)) {
3099 /*
3100 * The only way we can get a vma_merge with
3101 * self during an mremap is if the vma hasn't
3102 * been faulted in yet and we were allowed to
3103 * reset the dst vma->vm_pgoff to the
3104 * destination address of the mremap to allow
3105 * the merge to happen. mremap must change the
3106 * vm_pgoff linearity between src and dst vmas
3107 * (in turn preventing a vma_merge) to be
3108 * safe. It is only safe to keep the vm_pgoff
3109 * linear if there are no pages mapped yet.
3110 */
Sasha Levin81d1b092014-10-09 15:28:10 -07003111 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07003112 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07003113 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07003114 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08003116 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07003117 if (!new_vma)
3118 goto out;
3119 *new_vma = *vma;
3120 new_vma->vm_start = addr;
3121 new_vma->vm_end = addr + len;
3122 new_vma->vm_pgoff = pgoff;
3123 if (vma_dup_policy(vma, new_vma))
3124 goto out_free_vma;
3125 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3126 if (anon_vma_clone(new_vma, vma))
3127 goto out_free_mempol;
3128 if (new_vma->vm_file)
3129 get_file(new_vma->vm_file);
3130 if (new_vma->vm_ops && new_vma->vm_ops->open)
3131 new_vma->vm_ops->open(new_vma);
3132 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3133 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 }
3135 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003136
Chen Gange3975892015-09-08 15:03:38 -07003137out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07003138 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07003139out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08003140 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07003141out:
Rik van Riel5beb4932010-03-05 13:42:07 -08003142 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003144
3145/*
3146 * Return true if the calling process may expand its vm space by the passed
3147 * number of pages
3148 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003149bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003150{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003151 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3152 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003153
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003154 if (is_data_mapping(flags) &&
3155 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003156 /* Workaround for Valgrind */
3157 if (rlimit(RLIMIT_DATA) == 0 &&
3158 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3159 return true;
3160 if (!ignore_rlimit_data) {
3161 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 -08003162 current->comm, current->pid,
3163 (mm->data_vm + npages) << PAGE_SHIFT,
3164 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003165 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003166 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003167 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07003168
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003169 return true;
3170}
3171
3172void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3173{
3174 mm->total_vm += npages;
3175
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003176 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003177 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003178 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003179 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003180 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003181 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003182}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003183
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003184static int special_mapping_fault(struct vm_area_struct *vma,
3185 struct vm_fault *vmf);
3186
3187/*
3188 * Having a close hook prevents vma merging regardless of flags.
3189 */
3190static void special_mapping_close(struct vm_area_struct *vma)
3191{
3192}
3193
3194static const char *special_mapping_name(struct vm_area_struct *vma)
3195{
3196 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3197}
3198
Dmitry Safonovb059a452016-06-28 14:35:38 +03003199static int special_mapping_mremap(struct vm_area_struct *new_vma)
3200{
3201 struct vm_special_mapping *sm = new_vma->vm_private_data;
3202
3203 if (sm->mremap)
3204 return sm->mremap(sm, new_vma);
3205 return 0;
3206}
3207
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003208static const struct vm_operations_struct special_mapping_vmops = {
3209 .close = special_mapping_close,
3210 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03003211 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003212 .name = special_mapping_name,
3213};
3214
3215static const struct vm_operations_struct legacy_special_mapping_vmops = {
3216 .close = special_mapping_close,
3217 .fault = special_mapping_fault,
3218};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003219
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003220static int special_mapping_fault(struct vm_area_struct *vma,
3221 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003222{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003223 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003224 struct page **pages;
3225
Andy Lutomirskif872f542015-12-29 20:12:19 -08003226 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003227 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08003228 } else {
3229 struct vm_special_mapping *sm = vma->vm_private_data;
3230
3231 if (sm->fault)
3232 return sm->fault(sm, vma, vmf);
3233
3234 pages = sm->pages;
3235 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003236
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003237 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003238 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003239
3240 if (*pages) {
3241 struct page *page = *pages;
3242 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003243 vmf->page = page;
3244 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003245 }
3246
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003247 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003248}
3249
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003250static struct vm_area_struct *__install_special_mapping(
3251 struct mm_struct *mm,
3252 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003253 unsigned long vm_flags, void *priv,
3254 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003255{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003256 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003257 struct vm_area_struct *vma;
3258
3259 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3260 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003261 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003262
Rik van Riel5beb4932010-03-05 13:42:07 -08003263 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003264 vma->vm_mm = mm;
3265 vma->vm_start = addr;
3266 vma->vm_end = addr + len;
3267
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003268 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003269 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003270
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003271 vma->vm_ops = ops;
3272 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003273
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003274 ret = insert_vm_struct(mm, vma);
3275 if (ret)
3276 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003277
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003278 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003279
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003280 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003281
Stefani Seibold3935ed62014-03-17 23:22:02 +01003282 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003283
3284out:
3285 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003286 return ERR_PTR(ret);
3287}
3288
Dmitry Safonov2eefd872016-09-05 16:33:05 +03003289bool vma_is_special_mapping(const struct vm_area_struct *vma,
3290 const struct vm_special_mapping *sm)
3291{
3292 return vma->vm_private_data == sm &&
3293 (vma->vm_ops == &special_mapping_vmops ||
3294 vma->vm_ops == &legacy_special_mapping_vmops);
3295}
3296
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003297/*
3298 * Called with mm->mmap_sem held for writing.
3299 * Insert a new vma covering the given region, with the given flags.
3300 * Its pages are supplied by the given array of struct page *.
3301 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3302 * The region past the last page supplied will always produce SIGBUS.
3303 * The array pointer and the pages it points to are assumed to stay alive
3304 * for as long as this mapping might exist.
3305 */
3306struct vm_area_struct *_install_special_mapping(
3307 struct mm_struct *mm,
3308 unsigned long addr, unsigned long len,
3309 unsigned long vm_flags, const struct vm_special_mapping *spec)
3310{
Chen Gang27f28b92015-11-05 18:48:41 -08003311 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3312 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003313}
3314
Stefani Seibold3935ed62014-03-17 23:22:02 +01003315int install_special_mapping(struct mm_struct *mm,
3316 unsigned long addr, unsigned long len,
3317 unsigned long vm_flags, struct page **pages)
3318{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003319 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003320 mm, addr, len, vm_flags, (void *)pages,
3321 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003322
Duan Jiong14bd5b42014-06-04 16:07:05 -07003323 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003324}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003325
3326static DEFINE_MUTEX(mm_all_locks_mutex);
3327
Peter Zijlstra454ed842008-08-11 09:30:25 +02003328static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003329{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003330 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003331 /*
3332 * The LSB of head.next can't change from under us
3333 * because we hold the mm_all_locks_mutex.
3334 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003335 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003336 /*
3337 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003338 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003339 * the same anon_vma we won't take it again.
3340 *
3341 * No need of atomic instructions here, head.next
3342 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003343 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003344 */
3345 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003346 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003347 BUG();
3348 }
3349}
3350
Peter Zijlstra454ed842008-08-11 09:30:25 +02003351static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003352{
3353 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3354 /*
3355 * AS_MM_ALL_LOCKS can't change from under us because
3356 * we hold the mm_all_locks_mutex.
3357 *
3358 * Operations on ->flags have to be atomic because
3359 * even if AS_MM_ALL_LOCKS is stable thanks to the
3360 * mm_all_locks_mutex, there may be other cpus
3361 * changing other bitflags in parallel to us.
3362 */
3363 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3364 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003365 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003366 }
3367}
3368
3369/*
3370 * This operation locks against the VM for all pte/vma/mm related
3371 * operations that could ever happen on a certain mm. This includes
3372 * vmtruncate, try_to_unmap, and all page faults.
3373 *
3374 * The caller must take the mmap_sem in write mode before calling
3375 * mm_take_all_locks(). The caller isn't allowed to release the
3376 * mmap_sem until mm_drop_all_locks() returns.
3377 *
3378 * mmap_sem in write mode is required in order to block all operations
3379 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003380 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003381 * anon_vmas to be associated with existing vmas.
3382 *
3383 * A single task can't take more than one mm_take_all_locks() in a row
3384 * or it would deadlock.
3385 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003386 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003387 * mapping->flags avoid to take the same lock twice, if more than one
3388 * vma in this mm is backed by the same anon_vma or address_space.
3389 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003390 * We take locks in following order, accordingly to comment at beginning
3391 * of mm/rmap.c:
3392 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3393 * hugetlb mapping);
3394 * - all i_mmap_rwsem locks;
3395 * - all anon_vma->rwseml
3396 *
3397 * We can take all locks within these types randomly because the VM code
3398 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3399 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003400 *
3401 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3402 * that may have to take thousand of locks.
3403 *
3404 * mm_take_all_locks() can fail if it's interrupted by signals.
3405 */
3406int mm_take_all_locks(struct mm_struct *mm)
3407{
3408 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003409 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003410
3411 BUG_ON(down_read_trylock(&mm->mmap_sem));
3412
3413 mutex_lock(&mm_all_locks_mutex);
3414
3415 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3416 if (signal_pending(current))
3417 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003418 if (vma->vm_file && vma->vm_file->f_mapping &&
3419 is_vm_hugetlb_page(vma))
3420 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3421 }
3422
3423 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3424 if (signal_pending(current))
3425 goto out_unlock;
3426 if (vma->vm_file && vma->vm_file->f_mapping &&
3427 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003428 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003429 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003430
3431 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3432 if (signal_pending(current))
3433 goto out_unlock;
3434 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003435 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3436 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003437 }
3438
Kautuk Consul584cff52011-10-31 17:08:59 -07003439 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003440
3441out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003442 mm_drop_all_locks(mm);
3443 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003444}
3445
3446static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3447{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003448 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003449 /*
3450 * The LSB of head.next can't change to 0 from under
3451 * us because we hold the mm_all_locks_mutex.
3452 *
3453 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003454 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003455 * never see our bitflag.
3456 *
3457 * No need of atomic instructions here, head.next
3458 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003459 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003460 */
3461 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003462 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003463 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003464 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003465 }
3466}
3467
3468static void vm_unlock_mapping(struct address_space *mapping)
3469{
3470 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3471 /*
3472 * AS_MM_ALL_LOCKS can't change to 0 from under us
3473 * because we hold the mm_all_locks_mutex.
3474 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003475 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003476 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3477 &mapping->flags))
3478 BUG();
3479 }
3480}
3481
3482/*
3483 * The mmap_sem cannot be released by the caller until
3484 * mm_drop_all_locks() returns.
3485 */
3486void mm_drop_all_locks(struct mm_struct *mm)
3487{
3488 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003489 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003490
3491 BUG_ON(down_read_trylock(&mm->mmap_sem));
3492 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3493
3494 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3495 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003496 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3497 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003498 if (vma->vm_file && vma->vm_file->f_mapping)
3499 vm_unlock_mapping(vma->vm_file->f_mapping);
3500 }
3501
3502 mutex_unlock(&mm_all_locks_mutex);
3503}
David Howells8feae132009-01-08 12:04:47 +00003504
3505/*
3506 * initialise the VMA slab
3507 */
3508void __init mmap_init(void)
3509{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003510 int ret;
3511
Tejun Heo908c7f12014-09-08 09:51:29 +09003512 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003513 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003514}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003515
3516/*
3517 * Initialise sysctl_user_reserve_kbytes.
3518 *
3519 * This is intended to prevent a user from starting a single memory hogging
3520 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3521 * mode.
3522 *
3523 * The default value is min(3% of free memory, 128MB)
3524 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3525 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003526static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003527{
3528 unsigned long free_kbytes;
3529
3530 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3531
3532 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3533 return 0;
3534}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003535subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003536
3537/*
3538 * Initialise sysctl_admin_reserve_kbytes.
3539 *
3540 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3541 * to log in and kill a memory hogging process.
3542 *
3543 * Systems with more than 256MB will reserve 8MB, enough to recover
3544 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3545 * only reserve 3% of free pages by default.
3546 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003547static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003548{
3549 unsigned long free_kbytes;
3550
3551 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3552
3553 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3554 return 0;
3555}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003556subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003557
3558/*
3559 * Reinititalise user and admin reserves if memory is added or removed.
3560 *
3561 * The default user reserve max is 128MB, and the default max for the
3562 * admin reserve is 8MB. These are usually, but not always, enough to
3563 * enable recovery from a memory hogging process using login/sshd, a shell,
3564 * and tools like top. It may make sense to increase or even disable the
3565 * reserve depending on the existence of swap or variations in the recovery
3566 * tools. So, the admin may have changed them.
3567 *
3568 * If memory is added and the reserves have been eliminated or increased above
3569 * the default max, then we'll trust the admin.
3570 *
3571 * If memory is removed and there isn't enough free memory, then we
3572 * need to reset the reserves.
3573 *
3574 * Otherwise keep the reserve set by the admin.
3575 */
3576static int reserve_mem_notifier(struct notifier_block *nb,
3577 unsigned long action, void *data)
3578{
3579 unsigned long tmp, free_kbytes;
3580
3581 switch (action) {
3582 case MEM_ONLINE:
3583 /* Default max is 128MB. Leave alone if modified by operator. */
3584 tmp = sysctl_user_reserve_kbytes;
3585 if (0 < tmp && tmp < (1UL << 17))
3586 init_user_reserve();
3587
3588 /* Default max is 8MB. Leave alone if modified by operator. */
3589 tmp = sysctl_admin_reserve_kbytes;
3590 if (0 < tmp && tmp < (1UL << 13))
3591 init_admin_reserve();
3592
3593 break;
3594 case MEM_OFFLINE:
3595 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3596
3597 if (sysctl_user_reserve_kbytes > free_kbytes) {
3598 init_user_reserve();
3599 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3600 sysctl_user_reserve_kbytes);
3601 }
3602
3603 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3604 init_admin_reserve();
3605 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3606 sysctl_admin_reserve_kbytes);
3607 }
3608 break;
3609 default:
3610 break;
3611 }
3612 return NOTIFY_OK;
3613}
3614
3615static struct notifier_block reserve_mem_nb = {
3616 .notifier_call = reserve_mem_notifier,
3617};
3618
3619static int __meminit init_reserve_notifier(void)
3620{
3621 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003622 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003623
3624 return 0;
3625}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003626subsys_initcall(init_reserve_notifier);