blob: d1902872414f8679ebf8e1a7a10ee2fdef8e36b1 [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
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080048#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#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 */
Daniel Micayac34cea2017-07-06 15:36:47 -070097pgprot_t protection_map[16] __ro_after_init = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 __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
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800179static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf);
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 Dickins1be71072017-06-19 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;
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800189 LIST_HEAD(uf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700191 if (down_write_killable(&mm->mmap_sem))
192 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700194#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800195 /*
196 * CONFIG_COMPAT_BRK can still be overridden by setting
197 * randomize_va_space to 2, which will still cause mm->start_brk
198 * to be arbitrarily shifted
199 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700200 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800201 min_brk = mm->start_brk;
202 else
203 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700204#else
205 min_brk = mm->start_brk;
206#endif
207 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700209
210 /*
211 * Check against rlimit here. If this check is done later after the test
212 * of oldbrk with newbrk then it can escape the test and let the data
213 * segment grow beyond its set limit the in case where the limit is
214 * not page aligned -Ram Gupta
215 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700216 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
217 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700218 goto out;
219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 newbrk = PAGE_ALIGN(brk);
221 oldbrk = PAGE_ALIGN(mm->brk);
222 if (oldbrk == newbrk)
223 goto set_brk;
224
225 /* Always allow shrinking brk. */
226 if (brk <= mm->brk) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800227 if (!do_munmap(mm, newbrk, oldbrk-newbrk, &uf))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 goto set_brk;
229 goto out;
230 }
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 /* Check against existing mmap mappings. */
Hugh Dickins1be71072017-06-19 04:03:24 -0700233 next = find_vma(mm, oldbrk);
234 if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 goto out;
236
237 /* Ok, looks good - let it rip. */
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800238 if (do_brk(oldbrk, newbrk-oldbrk, &uf) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241set_brk:
242 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800243 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
244 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800245 userfaultfd_unmap_complete(mm, &uf);
Michel Lespinasse128557f2013-02-22 16:32:40 -0800246 if (populate)
247 mm_populate(oldbrk, newbrk - oldbrk);
248 return brk;
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250out:
251 retval = mm->brk;
252 up_write(&mm->mmap_sem);
253 return retval;
254}
255
Michel Lespinassed3737182012-12-11 16:01:38 -0800256static long vma_compute_subtree_gap(struct vm_area_struct *vma)
257{
Hugh Dickins1be71072017-06-19 04:03:24 -0700258 unsigned long max, prev_end, subtree_gap;
259
260 /*
261 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
262 * allow two stack_guard_gaps between them here, and when choosing
263 * an unmapped area; whereas when expanding we only require one.
264 * That's a little inconsistent, but keeps the code here simpler.
265 */
266 max = vm_start_gap(vma);
267 if (vma->vm_prev) {
268 prev_end = vm_end_gap(vma->vm_prev);
269 if (max > prev_end)
270 max -= prev_end;
271 else
272 max = 0;
273 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800274 if (vma->vm_rb.rb_left) {
275 subtree_gap = rb_entry(vma->vm_rb.rb_left,
276 struct vm_area_struct, vm_rb)->rb_subtree_gap;
277 if (subtree_gap > max)
278 max = subtree_gap;
279 }
280 if (vma->vm_rb.rb_right) {
281 subtree_gap = rb_entry(vma->vm_rb.rb_right,
282 struct vm_area_struct, vm_rb)->rb_subtree_gap;
283 if (subtree_gap > max)
284 max = subtree_gap;
285 }
286 return max;
287}
288
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700289#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800290static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800292 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800293 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 struct rb_node *nd, *pn = NULL;
295 unsigned long prev = 0, pend = 0;
296
297 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
298 struct vm_area_struct *vma;
299 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800300 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700301 pr_emerg("vm_start %lx < prev %lx\n",
302 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800303 bug = 1;
304 }
305 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700306 pr_emerg("vm_start %lx < pend %lx\n",
307 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800308 bug = 1;
309 }
310 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700311 pr_emerg("vm_start %lx > vm_end %lx\n",
312 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800313 bug = 1;
314 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800315 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800316 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700317 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800318 vma->rb_subtree_gap,
319 vma_compute_subtree_gap(vma));
320 bug = 1;
321 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800322 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 i++;
324 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800325 prev = vma->vm_start;
326 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
328 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800329 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800331 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700332 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800333 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800335 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336}
337
Michel Lespinassed3737182012-12-11 16:01:38 -0800338static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
339{
340 struct rb_node *nd;
341
342 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
343 struct vm_area_struct *vma;
344 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700345 VM_BUG_ON_VMA(vma != ignore &&
346 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
347 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349}
350
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700351static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
353 int bug = 0;
354 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800355 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700356 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700357
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700358 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800359 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700360 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700361
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800362 if (anon_vma) {
363 anon_vma_lock_read(anon_vma);
364 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
365 anon_vma_interval_tree_verify(avc);
366 anon_vma_unlock_read(anon_vma);
367 }
368
Hugh Dickins1be71072017-06-19 04:03:24 -0700369 highest_address = vm_end_gap(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700370 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 i++;
372 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800373 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700374 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800375 bug = 1;
376 }
377 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700378 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700379 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800380 bug = 1;
381 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800382 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800383 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700384 if (i != -1)
385 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800386 bug = 1;
387 }
Sasha Levin96dad672014-10-09 15:28:39 -0700388 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
390#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800391#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392#define validate_mm(mm) do { } while (0)
393#endif
394
Michel Lespinassed3737182012-12-11 16:01:38 -0800395RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
396 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
397
398/*
399 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
400 * vma->vm_prev->vm_end values changed, without modifying the vma's position
401 * in the rbtree.
402 */
403static void vma_gap_update(struct vm_area_struct *vma)
404{
405 /*
406 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
407 * function that does exacltly what we want.
408 */
409 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
410}
411
412static inline void vma_rb_insert(struct vm_area_struct *vma,
413 struct rb_root *root)
414{
415 /* All rb_subtree_gap values must be consistent prior to insertion */
416 validate_mm_rb(root, NULL);
417
418 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
419}
420
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700421static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
422{
423 /*
424 * Note rb_erase_augmented is a fairly large inline function,
425 * so make sure we instantiate it only once with our desired
426 * augmented rbtree callbacks.
427 */
428 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
429}
430
431static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
432 struct rb_root *root,
433 struct vm_area_struct *ignore)
434{
435 /*
436 * All rb_subtree_gap values must be consistent prior to erase,
437 * with the possible exception of the "next" vma being erased if
438 * next->vm_start was reduced.
439 */
440 validate_mm_rb(root, ignore);
441
442 __vma_rb_erase(vma, root);
443}
444
445static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
446 struct rb_root *root)
Michel Lespinassed3737182012-12-11 16:01:38 -0800447{
448 /*
449 * All rb_subtree_gap values must be consistent prior to erase,
450 * with the possible exception of the vma being erased.
451 */
452 validate_mm_rb(root, vma);
453
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700454 __vma_rb_erase(vma, root);
Michel Lespinassed3737182012-12-11 16:01:38 -0800455}
456
Michel Lespinassebf181b92012-10-08 16:31:39 -0700457/*
458 * vma has some anon_vma assigned, and is already inserted on that
459 * anon_vma's interval trees.
460 *
461 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
462 * vma must be removed from the anon_vma's interval trees using
463 * anon_vma_interval_tree_pre_update_vma().
464 *
465 * After the update, the vma will be reinserted using
466 * anon_vma_interval_tree_post_update_vma().
467 *
468 * The entire update must be protected by exclusive mmap_sem and by
469 * the root anon_vma's mutex.
470 */
471static inline void
472anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
473{
474 struct anon_vma_chain *avc;
475
476 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
477 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
478}
479
480static inline void
481anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
482{
483 struct anon_vma_chain *avc;
484
485 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
486 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
487}
488
Hugh Dickins6597d782012-10-08 16:29:07 -0700489static int find_vma_links(struct mm_struct *mm, unsigned long addr,
490 unsigned long end, struct vm_area_struct **pprev,
491 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
Hugh Dickins6597d782012-10-08 16:29:07 -0700493 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
495 __rb_link = &mm->mm_rb.rb_node;
496 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498 while (*__rb_link) {
499 struct vm_area_struct *vma_tmp;
500
501 __rb_parent = *__rb_link;
502 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
503
504 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700505 /* Fail if an existing vma overlaps the area */
506 if (vma_tmp->vm_start < end)
507 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 __rb_link = &__rb_parent->rb_left;
509 } else {
510 rb_prev = __rb_parent;
511 __rb_link = &__rb_parent->rb_right;
512 }
513 }
514
515 *pprev = NULL;
516 if (rb_prev)
517 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
518 *rb_link = __rb_link;
519 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700520 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521}
522
Cyril Hrubise8420a82013-04-29 15:08:33 -0700523static unsigned long count_vma_pages_range(struct mm_struct *mm,
524 unsigned long addr, unsigned long end)
525{
526 unsigned long nr_pages = 0;
527 struct vm_area_struct *vma;
528
529 /* Find first overlaping mapping */
530 vma = find_vma_intersection(mm, addr, end);
531 if (!vma)
532 return 0;
533
534 nr_pages = (min(end, vma->vm_end) -
535 max(addr, vma->vm_start)) >> PAGE_SHIFT;
536
537 /* Iterate over the rest of the overlaps */
538 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
539 unsigned long overlap_len;
540
541 if (vma->vm_start > end)
542 break;
543
544 overlap_len = min(end, vma->vm_end) - vma->vm_start;
545 nr_pages += overlap_len >> PAGE_SHIFT;
546 }
547
548 return nr_pages;
549}
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
552 struct rb_node **rb_link, struct rb_node *rb_parent)
553{
Michel Lespinassed3737182012-12-11 16:01:38 -0800554 /* Update tracking information for the gap following the new vma. */
555 if (vma->vm_next)
556 vma_gap_update(vma->vm_next);
557 else
Hugh Dickins1be71072017-06-19 04:03:24 -0700558 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800559
560 /*
561 * vma->vm_prev wasn't known when we followed the rbtree to find the
562 * correct insertion point for that vma. As a result, we could not
563 * update the vma vm_rb parents rb_subtree_gap values on the way down.
564 * So, we first insert the vma with a zero rb_subtree_gap value
565 * (to be consistent with what we did on the way down), and then
566 * immediately update the gap to the correct value. Finally we
567 * rebalance the rbtree after all augmented values have been set.
568 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800570 vma->rb_subtree_gap = 0;
571 vma_gap_update(vma);
572 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573}
574
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700575static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
ZhenwenXu48aae422009-01-06 14:40:21 -0800577 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579 file = vma->vm_file;
580 if (file) {
581 struct address_space *mapping = file->f_mapping;
582
583 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500584 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700586 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800589 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 flush_dcache_mmap_unlock(mapping);
591 }
592}
593
594static void
595__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
596 struct vm_area_struct *prev, struct rb_node **rb_link,
597 struct rb_node *rb_parent)
598{
599 __vma_link_list(mm, vma, prev, rb_parent);
600 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
603static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
604 struct vm_area_struct *prev, struct rb_node **rb_link,
605 struct rb_node *rb_parent)
606{
607 struct address_space *mapping = NULL;
608
Huang Shijie64ac4942014-06-04 16:07:33 -0700609 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800611 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614 __vma_link(mm, vma, prev, rb_link, rb_parent);
615 __vma_link_file(vma);
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800618 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
620 mm->map_count++;
621 validate_mm(mm);
622}
623
624/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700625 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700626 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800628static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Hugh Dickins6597d782012-10-08 16:29:07 -0700630 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800631 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
Hugh Dickins6597d782012-10-08 16:29:07 -0700633 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
634 &prev, &rb_link, &rb_parent))
635 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 __vma_link(mm, vma, prev, rb_link, rb_parent);
637 mm->map_count++;
638}
639
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700640static __always_inline void __vma_unlink_common(struct mm_struct *mm,
641 struct vm_area_struct *vma,
642 struct vm_area_struct *prev,
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700643 bool has_prev,
644 struct vm_area_struct *ignore)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Michel Lespinassed3737182012-12-11 16:01:38 -0800646 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700647
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700648 vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700649 next = vma->vm_next;
650 if (has_prev)
651 prev->vm_next = next;
652 else {
653 prev = vma->vm_prev;
654 if (prev)
655 prev->vm_next = next;
656 else
657 mm->mmap = next;
658 }
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700659 if (next)
660 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700661
662 /* Kill the cache */
663 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700666static inline void __vma_unlink_prev(struct mm_struct *mm,
667 struct vm_area_struct *vma,
668 struct vm_area_struct *prev)
669{
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700670 __vma_unlink_common(mm, vma, prev, true, vma);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700671}
672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673/*
674 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
675 * is already present in an i_mmap tree without adjusting the tree.
676 * The following helper function should be used when such adjustments
677 * are necessary. The "insert" vma (if any) is to be inserted
678 * before we drop the necessary locks.
679 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700680int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
681 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
682 struct vm_area_struct *expand)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 struct mm_struct *mm = vma->vm_mm;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700685 struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700687 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700688 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800690 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 long adjust_next = 0;
692 int remove_next = 0;
693
694 if (next && !insert) {
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700695 struct vm_area_struct *exporter = NULL, *importer = NULL;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700696
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (end >= next->vm_end) {
698 /*
699 * vma expands, overlapping all the next, and
700 * perhaps the one after too (mprotect case 6).
Andrea Arcangeli86d12e42016-10-07 17:01:34 -0700701 * The only other cases that gets here are
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700702 * case 1, case 7 and case 8.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700704 if (next == expand) {
705 /*
706 * The only case where we don't expand "vma"
707 * and we expand "next" instead is case 8.
708 */
709 VM_WARN_ON(end != next->vm_end);
710 /*
711 * remove_next == 3 means we're
712 * removing "vma" and that to do so we
713 * swapped "vma" and "next".
714 */
715 remove_next = 3;
716 VM_WARN_ON(file != next->vm_file);
717 swap(vma, next);
718 } else {
719 VM_WARN_ON(expand != vma);
720 /*
721 * case 1, 6, 7, remove_next == 2 is case 6,
722 * remove_next == 1 is case 1 or 7.
723 */
724 remove_next = 1 + (end > next->vm_end);
725 VM_WARN_ON(remove_next == 2 &&
726 end != next->vm_next->vm_end);
727 VM_WARN_ON(remove_next == 1 &&
728 end != next->vm_end);
729 /* trim end to next, for case 6 first pass */
730 end = next->vm_end;
731 }
732
Linus Torvalds287d97a2010-04-10 15:22:30 -0700733 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 importer = vma;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700735
736 /*
737 * If next doesn't have anon_vma, import from vma after
738 * next, if the vma overlaps with it.
739 */
Andrea Arcangeli97a42cd2016-10-07 17:01:31 -0700740 if (remove_next == 2 && !next->anon_vma)
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700741 exporter = next->vm_next;
742
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 } else if (end > next->vm_start) {
744 /*
745 * vma expands, overlapping part of the next:
746 * mprotect case 5 shifting the boundary up.
747 */
748 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700749 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 importer = vma;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700751 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 } else if (end < vma->vm_end) {
753 /*
754 * vma shrinks, and !insert tells it's not
755 * split_vma inserting another: so it must be
756 * mprotect case 4 shifting the boundary down.
757 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700758 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700759 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 importer = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700761 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Rik van Riel5beb4932010-03-05 13:42:07 -0800764 /*
765 * Easily overlooked: when mprotect shifts the boundary,
766 * make sure the expanding vma has anon_vma set if the
767 * shrinking vma had, to cover any anon pages imported.
768 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700769 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800770 int error;
771
Linus Torvalds287d97a2010-04-10 15:22:30 -0700772 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300773 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700774 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300775 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800776 }
777 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700778again:
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700779 vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
Kirill A. Shutemov37f9f552016-07-26 15:25:48 -0700780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 if (file) {
782 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800783 root = &mapping->i_mmap;
784 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530785
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800786 if (adjust_next)
787 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530788
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800789 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700792 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 * are visible to arm/parisc __flush_dcache_page
794 * throughout; but we cannot insert into address
795 * space until vma start or end is updated.
796 */
797 __vma_link_file(insert);
798 }
799 }
800
Michel Lespinassebf181b92012-10-08 16:31:39 -0700801 anon_vma = vma->anon_vma;
802 if (!anon_vma && adjust_next)
803 anon_vma = next->anon_vma;
804 if (anon_vma) {
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700805 VM_WARN_ON(adjust_next && next->anon_vma &&
806 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000807 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700808 anon_vma_interval_tree_pre_update_vma(vma);
809 if (adjust_next)
810 anon_vma_interval_tree_pre_update_vma(next);
811 }
Rik van Riel012f18002010-08-09 17:18:40 -0700812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 if (root) {
814 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700815 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700817 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 }
819
Michel Lespinassed3737182012-12-11 16:01:38 -0800820 if (start != vma->vm_start) {
821 vma->vm_start = start;
822 start_changed = true;
823 }
824 if (end != vma->vm_end) {
825 vma->vm_end = end;
826 end_changed = true;
827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 vma->vm_pgoff = pgoff;
829 if (adjust_next) {
830 next->vm_start += adjust_next << PAGE_SHIFT;
831 next->vm_pgoff += adjust_next;
832 }
833
834 if (root) {
835 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700836 vma_interval_tree_insert(next, root);
837 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 flush_dcache_mmap_unlock(mapping);
839 }
840
841 if (remove_next) {
842 /*
843 * vma_merge has merged next into vma, and needs
844 * us to remove next before dropping the locks.
845 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700846 if (remove_next != 3)
847 __vma_unlink_prev(mm, next, vma);
848 else
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700849 /*
850 * vma is not before next if they've been
851 * swapped.
852 *
853 * pre-swap() next->vm_start was reduced so
854 * tell validate_mm_rb to ignore pre-swap()
855 * "next" (which is stored in post-swap()
856 * "vma").
857 */
858 __vma_unlink_common(mm, next, NULL, false, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 if (file)
860 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 } else if (insert) {
862 /*
863 * split_vma has split insert from vma, and needs
864 * us to insert it before dropping the locks
865 * (it may either follow vma or precede it).
866 */
867 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800868 } else {
869 if (start_changed)
870 vma_gap_update(vma);
871 if (end_changed) {
872 if (!next)
Hugh Dickins1be71072017-06-19 04:03:24 -0700873 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800874 else if (!adjust_next)
875 vma_gap_update(next);
876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
878
Michel Lespinassebf181b92012-10-08 16:31:39 -0700879 if (anon_vma) {
880 anon_vma_interval_tree_post_update_vma(vma);
881 if (adjust_next)
882 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800883 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800886 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530888 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100889 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530890
891 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100892 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530893 }
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700896 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530897 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700899 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800900 if (next->anon_vma)
901 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700903 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 kmem_cache_free(vm_area_cachep, next);
905 /*
906 * In mprotect's case 6 (see comments on vma_merge),
907 * we must remove another next too. It would clutter
908 * up the code too much to do both in one go.
909 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700910 if (remove_next != 3) {
911 /*
912 * If "next" was removed and vma->vm_end was
913 * expanded (up) over it, in turn
914 * "next->vm_prev->vm_end" changed and the
915 * "vma->vm_next" gap must be updated.
916 */
917 next = vma->vm_next;
918 } else {
919 /*
920 * For the scope of the comment "next" and
921 * "vma" considered pre-swap(): if "vma" was
922 * removed, next->vm_start was expanded (down)
923 * over it and the "next" gap must be updated.
924 * Because of the swap() the post-swap() "vma"
925 * actually points to pre-swap() "next"
926 * (post-swap() "next" as opposed is now a
927 * dangling pointer).
928 */
929 next = vma;
930 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700931 if (remove_next == 2) {
932 remove_next = 1;
933 end = next->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 goto again;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700935 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800936 else if (next)
937 vma_gap_update(next);
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700938 else {
939 /*
940 * If remove_next == 2 we obviously can't
941 * reach this path.
942 *
943 * If remove_next == 3 we can't reach this
944 * path because pre-swap() next is always not
945 * NULL. pre-swap() "next" is not being
946 * removed and its next->vm_end is not altered
947 * (and furthermore "end" already matches
948 * next->vm_end in remove_next == 3).
949 *
950 * We reach this only in the remove_next == 1
951 * case if the "next" vma that was removed was
952 * the highest vma of the mm. However in such
953 * case next->vm_end == "end" and the extended
954 * "vma" has vma->vm_end == next->vm_end so
955 * mm->highest_vm_end doesn't need any update
956 * in remove_next == 1 case.
957 */
Hugh Dickins1be71072017-06-19 04:03:24 -0700958 VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530961 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100962 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
964 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800965
966 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967}
968
969/*
970 * If the vma has a ->close operation then the driver probably needs to release
971 * per-vma resources, so we don't attempt to merge those.
972 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700974 struct file *file, unsigned long vm_flags,
975 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800977 /*
978 * VM_SOFTDIRTY should not prevent from VMA merging, if we
979 * match the flags but dirty bit -- the caller should mark
980 * merged VMA as dirty. If dirty bit won't be excluded from
981 * comparison, we increase pressue on the memory system forcing
982 * the kernel to generate new VMAs when old one could be
983 * extended instead.
984 */
985 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 return 0;
987 if (vma->vm_file != file)
988 return 0;
989 if (vma->vm_ops && vma->vm_ops->close)
990 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700991 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
992 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 return 1;
994}
995
996static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700997 struct anon_vma *anon_vma2,
998 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
Shaohua Li965f55d2011-05-24 17:11:20 -07001000 /*
1001 * The list_is_singular() test is to avoid merging VMA cloned from
1002 * parents. This can improve scalability caused by anon_vma lock.
1003 */
1004 if ((!anon_vma1 || !anon_vma2) && (!vma ||
1005 list_is_singular(&vma->anon_vma_chain)))
1006 return 1;
1007 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008}
1009
1010/*
1011 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1012 * in front of (at a lower virtual address and file offset than) the vma.
1013 *
1014 * We cannot merge two vmas if they have differently assigned (non-NULL)
1015 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1016 *
1017 * We don't check here for the merged mmap wrapping around the end of pagecache
1018 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1019 * wrap, nor mmaps which cover the final page at index -1UL.
1020 */
1021static int
1022can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001023 struct anon_vma *anon_vma, struct file *file,
1024 pgoff_t vm_pgoff,
1025 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001027 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001028 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 if (vma->vm_pgoff == vm_pgoff)
1030 return 1;
1031 }
1032 return 0;
1033}
1034
1035/*
1036 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1037 * beyond (at a higher virtual address and file offset than) the vma.
1038 *
1039 * We cannot merge two vmas if they have differently assigned (non-NULL)
1040 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1041 */
1042static int
1043can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001044 struct anon_vma *anon_vma, struct file *file,
1045 pgoff_t vm_pgoff,
1046 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001048 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001049 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001051 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1053 return 1;
1054 }
1055 return 0;
1056}
1057
1058/*
1059 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1060 * whether that can be merged with its predecessor or its successor.
1061 * Or both (it neatly fills a hole).
1062 *
1063 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1064 * certain not to be mapped by the time vma_merge is called; but when
1065 * called for mprotect, it is certain to be already mapped (either at
1066 * an offset within prev, or at the start of next), and the flags of
1067 * this area are about to be changed to vm_flags - and the no-change
1068 * case has already been eliminated.
1069 *
1070 * The following mprotect cases have to be considered, where AAAA is
1071 * the area passed down from mprotect_fixup, never extending beyond one
1072 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1073 *
1074 * AAAA AAAA AAAA AAAA
1075 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1076 * cannot merge might become might become might become
1077 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1078 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001079 * mremap move: PPPPXXXXXXXX 8
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 * AAAA
1081 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1082 * might become case 1 below case 2 below case 3 below
1083 *
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001084 * It is important for case 8 that the the vma NNNN overlapping the
1085 * region AAAA is never going to extended over XXXX. Instead XXXX must
1086 * be extended in region AAAA and NNNN must be removed. This way in
1087 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1088 * rmap_locks, the properties of the merged vma will be already
1089 * correct for the whole merged range. Some of those properties like
1090 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1091 * be correct for the whole merged range immediately after the
1092 * rmap_locks are released. Otherwise if XXXX would be removed and
1093 * NNNN would be extended over the XXXX range, remove_migration_ptes
1094 * or other rmap walkers (if working on addresses beyond the "end"
1095 * parameter) may establish ptes with the wrong permissions of NNNN
1096 * instead of the right permissions of XXXX.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 */
1098struct vm_area_struct *vma_merge(struct mm_struct *mm,
1099 struct vm_area_struct *prev, unsigned long addr,
1100 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001101 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001102 pgoff_t pgoff, struct mempolicy *policy,
1103 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104{
1105 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1106 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001107 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109 /*
1110 * We later require that vma->vm_flags == vm_flags,
1111 * so this tests vma->vm_flags & VM_SPECIAL, too.
1112 */
1113 if (vm_flags & VM_SPECIAL)
1114 return NULL;
1115
1116 if (prev)
1117 next = prev->vm_next;
1118 else
1119 next = mm->mmap;
1120 area = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001121 if (area && area->vm_end == end) /* cases 6, 7, 8 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 next = next->vm_next;
1123
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001124 /* verify some invariant that must be enforced by the caller */
1125 VM_WARN_ON(prev && addr <= prev->vm_start);
1126 VM_WARN_ON(area && end > area->vm_end);
1127 VM_WARN_ON(addr >= end);
1128
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 /*
1130 * Can it merge with the predecessor?
1131 */
1132 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001133 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001135 anon_vma, file, pgoff,
1136 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 /*
1138 * OK, it can. Can we now merge in the successor as well?
1139 */
1140 if (next && end == next->vm_start &&
1141 mpol_equal(policy, vma_policy(next)) &&
1142 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001143 anon_vma, file,
1144 pgoff+pglen,
1145 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001147 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 /* cases 1, 6 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001149 err = __vma_adjust(prev, prev->vm_start,
1150 next->vm_end, prev->vm_pgoff, NULL,
1151 prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 } else /* cases 2, 5, 7 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001153 err = __vma_adjust(prev, prev->vm_start,
1154 end, prev->vm_pgoff, NULL, prev);
Rik van Riel5beb4932010-03-05 13:42:07 -08001155 if (err)
1156 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001157 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 return prev;
1159 }
1160
1161 /*
1162 * Can this new request be merged in front of next?
1163 */
1164 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001165 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001167 anon_vma, file, pgoff+pglen,
1168 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 if (prev && addr < prev->vm_end) /* case 4 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001170 err = __vma_adjust(prev, prev->vm_start,
1171 addr, prev->vm_pgoff, NULL, next);
1172 else { /* cases 3, 8 */
1173 err = __vma_adjust(area, addr, next->vm_end,
1174 next->vm_pgoff - pglen, NULL, next);
1175 /*
1176 * In case 3 area is already equal to next and
1177 * this is a noop, but in case 8 "area" has
1178 * been removed and next was expanded over it.
1179 */
1180 area = next;
1181 }
Rik van Riel5beb4932010-03-05 13:42:07 -08001182 if (err)
1183 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001184 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return area;
1186 }
1187
1188 return NULL;
1189}
1190
1191/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001192 * Rough compatbility check to quickly see if it's even worth looking
1193 * at sharing an anon_vma.
1194 *
1195 * They need to have the same vm_file, and the flags can only differ
1196 * in things that mprotect may change.
1197 *
1198 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1199 * we can merge the two vma's. For example, we refuse to merge a vma if
1200 * there is a vm_ops->close() function, because that indicates that the
1201 * driver is doing some kind of reference counting. But that doesn't
1202 * really matter for the anon_vma sharing case.
1203 */
1204static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1205{
1206 return a->vm_end == b->vm_start &&
1207 mpol_equal(vma_policy(a), vma_policy(b)) &&
1208 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001209 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001210 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1211}
1212
1213/*
1214 * Do some basic sanity checking to see if we can re-use the anon_vma
1215 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1216 * the same as 'old', the other will be the new one that is trying
1217 * to share the anon_vma.
1218 *
1219 * NOTE! This runs with mm_sem held for reading, so it is possible that
1220 * the anon_vma of 'old' is concurrently in the process of being set up
1221 * by another page fault trying to merge _that_. But that's ok: if it
1222 * is being set up, that automatically means that it will be a singleton
1223 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001224 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001225 *
1226 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1227 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1228 * is to return an anon_vma that is "complex" due to having gone through
1229 * a fork).
1230 *
1231 * We also make sure that the two vma's are compatible (adjacent,
1232 * and with the same memory policies). That's all stable, even with just
1233 * a read lock on the mm_sem.
1234 */
1235static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1236{
1237 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001238 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001239
1240 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1241 return anon_vma;
1242 }
1243 return NULL;
1244}
1245
1246/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1248 * neighbouring vmas for a suitable anon_vma, before it goes off
1249 * to allocate a new anon_vma. It checks because a repetitive
1250 * sequence of mprotects and faults may otherwise lead to distinct
1251 * anon_vmas being allocated, preventing vma merge in subsequent
1252 * mprotect.
1253 */
1254struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1255{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001256 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 near = vma->vm_next;
1260 if (!near)
1261 goto try_prev;
1262
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001263 anon_vma = reusable_anon_vma(near, vma, near);
1264 if (anon_vma)
1265 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001267 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 if (!near)
1269 goto none;
1270
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001271 anon_vma = reusable_anon_vma(near, near, vma);
1272 if (anon_vma)
1273 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274none:
1275 /*
1276 * There's no absolute need to look only at touching neighbours:
1277 * we could search further afield for "compatible" anon_vmas.
1278 * But it would probably just be a waste of time searching,
1279 * or lead to too many vmas hanging off the same anon_vma.
1280 * We're trying to allow mprotect remerging later on,
1281 * not trying to minimize memory used for anon_vmas.
1282 */
1283 return NULL;
1284}
1285
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286/*
Al Viro40401532012-02-13 03:58:52 +00001287 * If a hint addr is less than mmap_min_addr change hint to be as
1288 * low as possible but still greater than mmap_min_addr
1289 */
1290static inline unsigned long round_hint_to_min(unsigned long hint)
1291{
1292 hint &= PAGE_MASK;
1293 if (((void *)hint != NULL) &&
1294 (hint < mmap_min_addr))
1295 return PAGE_ALIGN(mmap_min_addr);
1296 return hint;
1297}
1298
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001299static inline int mlock_future_check(struct mm_struct *mm,
1300 unsigned long flags,
1301 unsigned long len)
1302{
1303 unsigned long locked, lock_limit;
1304
1305 /* mlock MCL_FUTURE? */
1306 if (flags & VM_LOCKED) {
1307 locked = len >> PAGE_SHIFT;
1308 locked += mm->locked_vm;
1309 lock_limit = rlimit(RLIMIT_MEMLOCK);
1310 lock_limit >>= PAGE_SHIFT;
1311 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1312 return -EAGAIN;
1313 }
1314 return 0;
1315}
1316
Al Viro40401532012-02-13 03:58:52 +00001317/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001318 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001320unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001322 unsigned long flags, vm_flags_t vm_flags,
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001323 unsigned long pgoff, unsigned long *populate,
1324 struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
vishnu.pscc71aba2014-10-09 15:26:29 -07001326 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001327 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Michel Lespinasse41badc12013-02-22 16:32:47 -08001329 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001330
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001331 if (!len)
1332 return -EINVAL;
1333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 /*
1335 * Does the application expect PROT_READ to imply PROT_EXEC?
1336 *
1337 * (the exception is when the underlying filesystem is noexec
1338 * mounted, in which case we dont add PROT_EXEC.)
1339 */
1340 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001341 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 prot |= PROT_EXEC;
1343
Eric Paris7cd94142007-11-26 18:47:40 -05001344 if (!(flags & MAP_FIXED))
1345 addr = round_hint_to_min(addr);
1346
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 /* Careful about overflows.. */
1348 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001349 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 return -ENOMEM;
1351
1352 /* offset overflow? */
1353 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001354 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
1356 /* Too many mappings? */
1357 if (mm->map_count > sysctl_max_map_count)
1358 return -ENOMEM;
1359
1360 /* Obtain the address to map to. we verify (or select) it and ensure
1361 * that it represents a valid section of the address space.
1362 */
1363 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001364 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 return addr;
1366
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001367 if (prot == PROT_EXEC) {
1368 pkey = execute_only_pkey(mm);
1369 if (pkey < 0)
1370 pkey = 0;
1371 }
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 /* Do simple checking here so the lower-level routines won't have
1374 * to. we assume access permissions have been handled by the open
1375 * of the memory object, so we don't do any here.
1376 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001377 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1379
Huang Shijiecdf7b342009-09-21 17:03:36 -07001380 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 if (!can_do_mlock())
1382 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001383
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001384 if (mlock_future_check(mm, vm_flags, len))
1385 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001388 struct inode *inode = file_inode(file);
1389
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 switch (flags & MAP_TYPE) {
1391 case MAP_SHARED:
1392 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1393 return -EACCES;
1394
1395 /*
1396 * Make sure we don't allow writing to an append-only
1397 * file..
1398 */
1399 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1400 return -EACCES;
1401
1402 /*
1403 * Make sure there are no mandatory locks on the file.
1404 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001405 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 return -EAGAIN;
1407
1408 vm_flags |= VM_SHARED | VM_MAYSHARE;
1409 if (!(file->f_mode & FMODE_WRITE))
1410 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1411
1412 /* fall through */
1413 case MAP_PRIVATE:
1414 if (!(file->f_mode & FMODE_READ))
1415 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001416 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001417 if (vm_flags & VM_EXEC)
1418 return -EPERM;
1419 vm_flags &= ~VM_MAYEXEC;
1420 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001421
Al Viro72c2d532013-09-22 16:27:52 -04001422 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001423 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001424 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1425 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 break;
1427
1428 default:
1429 return -EINVAL;
1430 }
1431 } else {
1432 switch (flags & MAP_TYPE) {
1433 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001434 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1435 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001436 /*
1437 * Ignore pgoff.
1438 */
1439 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 vm_flags |= VM_SHARED | VM_MAYSHARE;
1441 break;
1442 case MAP_PRIVATE:
1443 /*
1444 * Set pgoff according to addr for anon_vma.
1445 */
1446 pgoff = addr >> PAGE_SHIFT;
1447 break;
1448 default:
1449 return -EINVAL;
1450 }
1451 }
1452
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001453 /*
1454 * Set 'VM_NORESERVE' if we should not account for the
1455 * memory use of this mapping.
1456 */
1457 if (flags & MAP_NORESERVE) {
1458 /* We honor MAP_NORESERVE if allowed to overcommit */
1459 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1460 vm_flags |= VM_NORESERVE;
1461
1462 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1463 if (file && is_file_hugepages(file))
1464 vm_flags |= VM_NORESERVE;
1465 }
1466
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001467 addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001468 if (!IS_ERR_VALUE(addr) &&
1469 ((vm_flags & VM_LOCKED) ||
1470 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001471 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001472 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001473}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001474
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001475SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1476 unsigned long, prot, unsigned long, flags,
1477 unsigned long, fd, unsigned long, pgoff)
1478{
1479 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001480 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001481
1482 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001483 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001484 file = fget(fd);
1485 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001486 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001487 if (is_file_hugepages(file))
1488 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001489 retval = -EINVAL;
1490 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1491 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001492 } else if (flags & MAP_HUGETLB) {
1493 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001494 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001495
Anshuman Khandual20ac2892017-05-03 14:55:00 -07001496 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001497 if (!hs)
1498 return -EINVAL;
1499
1500 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001501 /*
1502 * VM_NORESERVE is used because the reservations will be
1503 * taken when vm_ops->mmap() is called
1504 * A dummy user value is used because we are not locking
1505 * memory so no accounting is necessary
1506 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001507 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001508 VM_NORESERVE,
1509 &user, HUGETLB_ANONHUGE_INODE,
1510 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001511 if (IS_ERR(file))
1512 return PTR_ERR(file);
1513 }
1514
1515 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1516
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001517 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001518out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001519 if (file)
1520 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001521 return retval;
1522}
1523
Christoph Hellwiga4679372010-03-10 15:21:15 -08001524#ifdef __ARCH_WANT_SYS_OLD_MMAP
1525struct mmap_arg_struct {
1526 unsigned long addr;
1527 unsigned long len;
1528 unsigned long prot;
1529 unsigned long flags;
1530 unsigned long fd;
1531 unsigned long offset;
1532};
1533
1534SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1535{
1536 struct mmap_arg_struct a;
1537
1538 if (copy_from_user(&a, arg, sizeof(a)))
1539 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001540 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001541 return -EINVAL;
1542
1543 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1544 a.offset >> PAGE_SHIFT);
1545}
1546#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1547
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001548/*
1549 * Some shared mappigns will want the pages marked read-only
1550 * to track write events. If so, we'll downgrade vm_page_prot
1551 * to the private version (using protection_map[] without the
1552 * VM_SHARED bit).
1553 */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001554int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001555{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001556 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001557 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001558
1559 /* If it was private or non-writable, the write bit is already clear */
1560 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1561 return 0;
1562
1563 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001564 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001565 return 1;
1566
Peter Feiner64e45502014-10-13 15:55:46 -07001567 /* The open routine did something to the protections that pgprot_modify
1568 * won't preserve? */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001569 if (pgprot_val(vm_page_prot) !=
1570 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001571 return 0;
1572
Peter Feiner64e45502014-10-13 15:55:46 -07001573 /* Do we need to track softdirty? */
1574 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1575 return 1;
1576
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001577 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001578 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001579 return 0;
1580
1581 /* Can the mapping track the dirty pages? */
1582 return vma->vm_file && vma->vm_file->f_mapping &&
1583 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1584}
1585
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001586/*
1587 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001588 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001589 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001590static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001591{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001592 /*
1593 * hugetlb has its own accounting separate from the core VM
1594 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1595 */
1596 if (file && is_file_hugepages(file))
1597 return 0;
1598
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001599 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1600}
1601
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001602unsigned long mmap_region(struct file *file, unsigned long addr,
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001603 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
1604 struct list_head *uf)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001605{
1606 struct mm_struct *mm = current->mm;
1607 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001608 int error;
1609 struct rb_node **rb_link, *rb_parent;
1610 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001611
Cyril Hrubise8420a82013-04-29 15:08:33 -07001612 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001613 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001614 unsigned long nr_pages;
1615
1616 /*
1617 * MAP_FIXED may remove pages of mappings that intersects with
1618 * requested mapping. Account for the pages it would unmap.
1619 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001620 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1621
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001622 if (!may_expand_vm(mm, vm_flags,
1623 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001624 return -ENOMEM;
1625 }
1626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001628 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1629 &rb_parent)) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001630 if (do_munmap(mm, addr, len, uf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 }
1633
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001634 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001635 * Private writable mapping: check memory availability
1636 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001637 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001638 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001639 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001640 return -ENOMEM;
1641 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 }
1643
1644 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001645 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001647 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1648 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001649 if (vma)
1650 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652 /*
1653 * Determine the object being mapped and call the appropriate
1654 * specific mapper. the address has already been validated, but
1655 * not unmapped, but the maps are removed from the list.
1656 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001657 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 if (!vma) {
1659 error = -ENOMEM;
1660 goto unacct_error;
1661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
1663 vma->vm_mm = mm;
1664 vma->vm_start = addr;
1665 vma->vm_end = addr + len;
1666 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001667 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001669 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
1671 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 if (vm_flags & VM_DENYWRITE) {
1673 error = deny_write_access(file);
1674 if (error)
1675 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001677 if (vm_flags & VM_SHARED) {
1678 error = mapping_map_writable(file->f_mapping);
1679 if (error)
1680 goto allow_write_and_free_vma;
1681 }
1682
1683 /* ->mmap() can change vma->vm_file, but must guarantee that
1684 * vma_link() below can deny write-access if VM_DENYWRITE is set
1685 * and map writably if VM_SHARED is set. This usually means the
1686 * new file must not have been exposed to user-space, yet.
1687 */
Al Virocb0942b2012-08-27 14:48:26 -04001688 vma->vm_file = get_file(file);
Miklos Szeredif74ac012017-02-20 16:51:23 +01001689 error = call_mmap(file, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 if (error)
1691 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001692
1693 /* Can addr have changed??
1694 *
1695 * Answer: Yes, several device drivers can do it in their
1696 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001697 * Bug: If addr is changed, prev, rb_link, rb_parent should
1698 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001699 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001700 WARN_ON_ONCE(addr != vma->vm_start);
1701
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001702 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001703 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 } else if (vm_flags & VM_SHARED) {
1705 error = shmem_zero_setup(vma);
1706 if (error)
1707 goto free_vma;
1708 }
1709
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001710 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001711 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001712 if (file) {
1713 if (vm_flags & VM_SHARED)
1714 mapping_unmap_writable(file->f_mapping);
1715 if (vm_flags & VM_DENYWRITE)
1716 allow_write_access(file);
1717 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001718 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001719out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001720 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001721
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001722 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001724 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1725 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001726 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001727 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001728 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001729 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301730
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001731 if (file)
1732 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301733
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001734 /*
1735 * New (or expanded) vma always get soft dirty status.
1736 * Otherwise user-space soft-dirty page tracker won't
1737 * be able to distinguish situation when vma area unmapped,
1738 * then new mapped in-place (which must be aimed as
1739 * a completely new data area).
1740 */
1741 vma->vm_flags |= VM_SOFTDIRTY;
1742
Peter Feiner64e45502014-10-13 15:55:46 -07001743 vma_set_page_prot(vma);
1744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 return addr;
1746
1747unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 vma->vm_file = NULL;
1749 fput(file);
1750
1751 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001752 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1753 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001754 if (vm_flags & VM_SHARED)
1755 mapping_unmap_writable(file->f_mapping);
1756allow_write_and_free_vma:
1757 if (vm_flags & VM_DENYWRITE)
1758 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759free_vma:
1760 kmem_cache_free(vm_area_cachep, vma);
1761unacct_error:
1762 if (charged)
1763 vm_unacct_memory(charged);
1764 return error;
1765}
1766
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001767unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1768{
1769 /*
1770 * We implement the search by looking for an rbtree node that
1771 * immediately follows a suitable gap. That is,
1772 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1773 * - gap_end = vma->vm_start >= info->low_limit + length;
1774 * - gap_end - gap_start >= length
1775 */
1776
1777 struct mm_struct *mm = current->mm;
1778 struct vm_area_struct *vma;
1779 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1780
1781 /* Adjust search length to account for worst case alignment overhead */
1782 length = info->length + info->align_mask;
1783 if (length < info->length)
1784 return -ENOMEM;
1785
1786 /* Adjust search limits by the desired length */
1787 if (info->high_limit < length)
1788 return -ENOMEM;
1789 high_limit = info->high_limit - length;
1790
1791 if (info->low_limit > high_limit)
1792 return -ENOMEM;
1793 low_limit = info->low_limit + length;
1794
1795 /* Check if rbtree root looks promising */
1796 if (RB_EMPTY_ROOT(&mm->mm_rb))
1797 goto check_highest;
1798 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1799 if (vma->rb_subtree_gap < length)
1800 goto check_highest;
1801
1802 while (true) {
1803 /* Visit left subtree if it looks promising */
Hugh Dickins1be71072017-06-19 04:03:24 -07001804 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001805 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1806 struct vm_area_struct *left =
1807 rb_entry(vma->vm_rb.rb_left,
1808 struct vm_area_struct, vm_rb);
1809 if (left->rb_subtree_gap >= length) {
1810 vma = left;
1811 continue;
1812 }
1813 }
1814
Hugh Dickins1be71072017-06-19 04:03:24 -07001815 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001816check_current:
1817 /* Check if current node has a suitable gap */
1818 if (gap_start > high_limit)
1819 return -ENOMEM;
Hugh Dickinsf4cb7672017-06-20 02:10:44 -07001820 if (gap_end >= low_limit &&
1821 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001822 goto found;
1823
1824 /* Visit right subtree if it looks promising */
1825 if (vma->vm_rb.rb_right) {
1826 struct vm_area_struct *right =
1827 rb_entry(vma->vm_rb.rb_right,
1828 struct vm_area_struct, vm_rb);
1829 if (right->rb_subtree_gap >= length) {
1830 vma = right;
1831 continue;
1832 }
1833 }
1834
1835 /* Go back up the rbtree to find next candidate node */
1836 while (true) {
1837 struct rb_node *prev = &vma->vm_rb;
1838 if (!rb_parent(prev))
1839 goto check_highest;
1840 vma = rb_entry(rb_parent(prev),
1841 struct vm_area_struct, vm_rb);
1842 if (prev == vma->vm_rb.rb_left) {
Hugh Dickins1be71072017-06-19 04:03:24 -07001843 gap_start = vm_end_gap(vma->vm_prev);
1844 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001845 goto check_current;
1846 }
1847 }
1848 }
1849
1850check_highest:
1851 /* Check highest gap, which does not precede any rbtree node */
1852 gap_start = mm->highest_vm_end;
1853 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1854 if (gap_start > high_limit)
1855 return -ENOMEM;
1856
1857found:
1858 /* We found a suitable gap. Clip it with the original low_limit. */
1859 if (gap_start < info->low_limit)
1860 gap_start = info->low_limit;
1861
1862 /* Adjust gap address to the desired alignment */
1863 gap_start += (info->align_offset - gap_start) & info->align_mask;
1864
1865 VM_BUG_ON(gap_start + info->length > info->high_limit);
1866 VM_BUG_ON(gap_start + info->length > gap_end);
1867 return gap_start;
1868}
1869
1870unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1871{
1872 struct mm_struct *mm = current->mm;
1873 struct vm_area_struct *vma;
1874 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1875
1876 /* Adjust search length to account for worst case alignment overhead */
1877 length = info->length + info->align_mask;
1878 if (length < info->length)
1879 return -ENOMEM;
1880
1881 /*
1882 * Adjust search limits by the desired length.
1883 * See implementation comment at top of unmapped_area().
1884 */
1885 gap_end = info->high_limit;
1886 if (gap_end < length)
1887 return -ENOMEM;
1888 high_limit = gap_end - length;
1889
1890 if (info->low_limit > high_limit)
1891 return -ENOMEM;
1892 low_limit = info->low_limit + length;
1893
1894 /* Check highest gap, which does not precede any rbtree node */
1895 gap_start = mm->highest_vm_end;
1896 if (gap_start <= high_limit)
1897 goto found_highest;
1898
1899 /* Check if rbtree root looks promising */
1900 if (RB_EMPTY_ROOT(&mm->mm_rb))
1901 return -ENOMEM;
1902 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1903 if (vma->rb_subtree_gap < length)
1904 return -ENOMEM;
1905
1906 while (true) {
1907 /* Visit right subtree if it looks promising */
Hugh Dickins1be71072017-06-19 04:03:24 -07001908 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001909 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1910 struct vm_area_struct *right =
1911 rb_entry(vma->vm_rb.rb_right,
1912 struct vm_area_struct, vm_rb);
1913 if (right->rb_subtree_gap >= length) {
1914 vma = right;
1915 continue;
1916 }
1917 }
1918
1919check_current:
1920 /* Check if current node has a suitable gap */
Hugh Dickins1be71072017-06-19 04:03:24 -07001921 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001922 if (gap_end < low_limit)
1923 return -ENOMEM;
Hugh Dickinsf4cb7672017-06-20 02:10:44 -07001924 if (gap_start <= high_limit &&
1925 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001926 goto found;
1927
1928 /* Visit left subtree if it looks promising */
1929 if (vma->vm_rb.rb_left) {
1930 struct vm_area_struct *left =
1931 rb_entry(vma->vm_rb.rb_left,
1932 struct vm_area_struct, vm_rb);
1933 if (left->rb_subtree_gap >= length) {
1934 vma = left;
1935 continue;
1936 }
1937 }
1938
1939 /* Go back up the rbtree to find next candidate node */
1940 while (true) {
1941 struct rb_node *prev = &vma->vm_rb;
1942 if (!rb_parent(prev))
1943 return -ENOMEM;
1944 vma = rb_entry(rb_parent(prev),
1945 struct vm_area_struct, vm_rb);
1946 if (prev == vma->vm_rb.rb_right) {
1947 gap_start = vma->vm_prev ?
Hugh Dickins1be71072017-06-19 04:03:24 -07001948 vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001949 goto check_current;
1950 }
1951 }
1952 }
1953
1954found:
1955 /* We found a suitable gap. Clip it with the original high_limit. */
1956 if (gap_end > info->high_limit)
1957 gap_end = info->high_limit;
1958
1959found_highest:
1960 /* Compute highest gap address at the desired alignment */
1961 gap_end -= info->length;
1962 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1963
1964 VM_BUG_ON(gap_end < info->low_limit);
1965 VM_BUG_ON(gap_end < gap_start);
1966 return gap_end;
1967}
1968
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969/* Get an address range which is currently unmapped.
1970 * For shmat() with addr=0.
1971 *
1972 * Ugly calling convention alert:
1973 * Return value with the low bits set means error value,
1974 * ie
1975 * if (ret & ~PAGE_MASK)
1976 * error = ret;
1977 *
1978 * This function "knows" that -ENOMEM has the bits set.
1979 */
1980#ifndef HAVE_ARCH_UNMAPPED_AREA
1981unsigned long
1982arch_get_unmapped_area(struct file *filp, unsigned long addr,
1983 unsigned long len, unsigned long pgoff, unsigned long flags)
1984{
1985 struct mm_struct *mm = current->mm;
Hugh Dickins1be71072017-06-19 04:03:24 -07001986 struct vm_area_struct *vma, *prev;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001987 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001989 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 return -ENOMEM;
1991
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001992 if (flags & MAP_FIXED)
1993 return addr;
1994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 if (addr) {
1996 addr = PAGE_ALIGN(addr);
Hugh Dickins1be71072017-06-19 04:03:24 -07001997 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001998 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickins1be71072017-06-19 04:03:24 -07001999 (!vma || addr + len <= vm_start_gap(vma)) &&
2000 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 return addr;
2002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002004 info.flags = 0;
2005 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08002006 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002007 info.high_limit = TASK_SIZE;
2008 info.align_mask = 0;
2009 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010}
vishnu.pscc71aba2014-10-09 15:26:29 -07002011#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013/*
2014 * This mmap-allocator allocates new areas top-down from below the
2015 * stack's low limit (the base):
2016 */
2017#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2018unsigned long
2019arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
2020 const unsigned long len, const unsigned long pgoff,
2021 const unsigned long flags)
2022{
Hugh Dickins1be71072017-06-19 04:03:24 -07002023 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002025 unsigned long addr = addr0;
2026 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
2028 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002029 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 return -ENOMEM;
2031
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002032 if (flags & MAP_FIXED)
2033 return addr;
2034
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 /* requesting a specific address */
2036 if (addr) {
2037 addr = PAGE_ALIGN(addr);
Hugh Dickins1be71072017-06-19 04:03:24 -07002038 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002039 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickins1be71072017-06-19 04:03:24 -07002040 (!vma || addr + len <= vm_start_gap(vma)) &&
2041 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 return addr;
2043 }
2044
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002045 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2046 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002047 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002048 info.high_limit = mm->mmap_base;
2049 info.align_mask = 0;
2050 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07002051
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 /*
2053 * A failed mmap() very likely causes application failure,
2054 * so fall back to the bottom-up function here. This scenario
2055 * can happen with large stack limits and large mmap()
2056 * allocations.
2057 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002058 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002059 VM_BUG_ON(addr != -ENOMEM);
2060 info.flags = 0;
2061 info.low_limit = TASK_UNMAPPED_BASE;
2062 info.high_limit = TASK_SIZE;
2063 addr = vm_unmapped_area(&info);
2064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
2066 return addr;
2067}
2068#endif
2069
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070unsigned long
2071get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2072 unsigned long pgoff, unsigned long flags)
2073{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002074 unsigned long (*get_area)(struct file *, unsigned long,
2075 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
Al Viro9206de92009-12-03 15:23:11 -05002077 unsigned long error = arch_mmap_check(addr, len, flags);
2078 if (error)
2079 return error;
2080
2081 /* Careful about overflows.. */
2082 if (len > TASK_SIZE)
2083 return -ENOMEM;
2084
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002085 get_area = current->mm->get_unmapped_area;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002086 if (file) {
2087 if (file->f_op->get_unmapped_area)
2088 get_area = file->f_op->get_unmapped_area;
2089 } else if (flags & MAP_SHARED) {
2090 /*
2091 * mmap_region() will call shmem_zero_setup() to create a file,
2092 * so use shmem's get_unmapped_area in case it can be huge.
2093 * do_mmap_pgoff() will clear pgoff, so match alignment.
2094 */
2095 pgoff = 0;
2096 get_area = shmem_get_unmapped_area;
2097 }
2098
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002099 addr = get_area(file, addr, len, pgoff, flags);
2100 if (IS_ERR_VALUE(addr))
2101 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002103 if (addr > TASK_SIZE - len)
2104 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002105 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002106 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002107
Al Viro9ac4ed42012-05-30 17:13:15 -04002108 error = security_mmap_addr(addr);
2109 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110}
2111
2112EXPORT_SYMBOL(get_unmapped_area);
2113
2114/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002115struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002117 struct rb_node *rb_node;
2118 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002120 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002121 vma = vmacache_find(mm, addr);
2122 if (likely(vma))
2123 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002125 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002127 while (rb_node) {
2128 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002130 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002131
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002132 if (tmp->vm_end > addr) {
2133 vma = tmp;
2134 if (tmp->vm_start <= addr)
2135 break;
2136 rb_node = rb_node->rb_left;
2137 } else
2138 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002140
2141 if (vma)
2142 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 return vma;
2144}
2145
2146EXPORT_SYMBOL(find_vma);
2147
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002148/*
2149 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002150 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151struct vm_area_struct *
2152find_vma_prev(struct mm_struct *mm, unsigned long addr,
2153 struct vm_area_struct **pprev)
2154{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002155 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002157 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002158 if (vma) {
2159 *pprev = vma->vm_prev;
2160 } else {
2161 struct rb_node *rb_node = mm->mm_rb.rb_node;
2162 *pprev = NULL;
2163 while (rb_node) {
2164 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2165 rb_node = rb_node->rb_right;
2166 }
2167 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002168 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169}
2170
2171/*
2172 * Verify that the stack growth is acceptable and
2173 * update accounting. This is shared with both the
2174 * grow-up and grow-down cases.
2175 */
Hugh Dickins1be71072017-06-19 04:03:24 -07002176static int acct_stack_growth(struct vm_area_struct *vma,
2177 unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178{
2179 struct mm_struct *mm = vma->vm_mm;
2180 struct rlimit *rlim = current->signal->rlim;
Hugh Dickins1be71072017-06-19 04:03:24 -07002181 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182
2183 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002184 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 return -ENOMEM;
2186
2187 /* Stack limit test */
Hugh Dickins1be71072017-06-19 04:03:24 -07002188 if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 return -ENOMEM;
2190
2191 /* mlock limit tests */
2192 if (vma->vm_flags & VM_LOCKED) {
2193 unsigned long locked;
2194 unsigned long limit;
2195 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002196 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002197 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 if (locked > limit && !capable(CAP_IPC_LOCK))
2199 return -ENOMEM;
2200 }
2201
Adam Litke0d59a012007-01-30 14:35:39 -08002202 /* Check to ensure the stack will not grow into a hugetlb-only region */
2203 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2204 vma->vm_end - size;
2205 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2206 return -EFAULT;
2207
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 /*
2209 * Overcommit.. This must be the final test, as it will
2210 * update security statistics.
2211 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002212 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 return -ENOMEM;
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 return 0;
2216}
2217
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002218#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002220 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2221 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002223int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224{
Oleg Nesterov09357812015-11-05 18:48:17 -08002225 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins1be71072017-06-19 04:03:24 -07002226 struct vm_area_struct *next;
2227 unsigned long gap_addr;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002228 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
2230 if (!(vma->vm_flags & VM_GROWSUP))
2231 return -EFAULT;
2232
Helge Dellerbd726c92017-06-19 17:34:05 +02002233 /* Guard against exceeding limits of the address space. */
Hugh Dickins1be71072017-06-19 04:03:24 -07002234 address &= PAGE_MASK;
Helge Dellerbd726c92017-06-19 17:34:05 +02002235 if (address >= TASK_SIZE)
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002236 return -ENOMEM;
Helge Dellerbd726c92017-06-19 17:34:05 +02002237 address += PAGE_SIZE;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002238
Hugh Dickins1be71072017-06-19 04:03:24 -07002239 /* Enforce stack_guard_gap */
2240 gap_addr = address + stack_guard_gap;
Helge Dellerbd726c92017-06-19 17:34:05 +02002241
2242 /* Guard against overflow */
2243 if (gap_addr < address || gap_addr > TASK_SIZE)
2244 gap_addr = TASK_SIZE;
2245
Hugh Dickins1be71072017-06-19 04:03:24 -07002246 next = vma->vm_next;
Michal Hocko561b5e02017-07-10 15:49:51 -07002247 if (next && next->vm_start < gap_addr &&
2248 (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Hugh Dickins1be71072017-06-19 04:03:24 -07002249 if (!(next->vm_flags & VM_GROWSUP))
2250 return -ENOMEM;
2251 /* Check that both stack segments have the same anon_vma? */
2252 }
2253
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002254 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 if (unlikely(anon_vma_prepare(vma)))
2256 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257
2258 /*
2259 * vma->vm_start/vm_end cannot change under us because the caller
2260 * is required to hold the mmap_sem in read mode. We need the
2261 * anon_vma lock to serialize against concurrent expand_stacks.
2262 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002263 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
2265 /* Somebody else might have raced and expanded it already */
2266 if (address > vma->vm_end) {
2267 unsigned long size, grow;
2268
2269 size = address - vma->vm_start;
2270 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2271
Hugh Dickins42c36f62011-05-09 17:44:42 -07002272 error = -ENOMEM;
2273 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2274 error = acct_stack_growth(vma, size, grow);
2275 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002276 /*
2277 * vma_gap_update() doesn't support concurrent
2278 * updates, but we only hold a shared mmap_sem
2279 * lock here, so we need to protect against
2280 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002281 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002282 * we don't guarantee that all growable vmas
2283 * in a mm share the same root anon vma.
2284 * So, we reuse mm->page_table_lock to guard
2285 * against concurrent vma expansions.
2286 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002287 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002288 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002289 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002290 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002291 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002292 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002293 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002294 if (vma->vm_next)
2295 vma_gap_update(vma->vm_next);
2296 else
Hugh Dickins1be71072017-06-19 04:03:24 -07002297 mm->highest_vm_end = vm_end_gap(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002298 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002299
Hugh Dickins42c36f62011-05-09 17:44:42 -07002300 perf_event_mmap(vma);
2301 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002304 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002305 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002306 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 return error;
2308}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002309#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311/*
2312 * vma is the first one with address < vma->vm_start. Have to extend vma.
2313 */
Michal Hockod05f3162011-05-24 17:11:44 -07002314int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002315 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316{
Oleg Nesterov09357812015-11-05 18:48:17 -08002317 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins1be71072017-06-19 04:03:24 -07002318 struct vm_area_struct *prev;
2319 unsigned long gap_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 int error;
2321
Eric Paris88694772007-11-26 18:47:26 -05002322 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002323 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002324 if (error)
2325 return error;
2326
Hugh Dickins1be71072017-06-19 04:03:24 -07002327 /* Enforce stack_guard_gap */
2328 gap_addr = address - stack_guard_gap;
2329 if (gap_addr > address)
2330 return -ENOMEM;
2331 prev = vma->vm_prev;
Michal Hocko561b5e02017-07-10 15:49:51 -07002332 if (prev && prev->vm_end > gap_addr &&
2333 (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Hugh Dickins1be71072017-06-19 04:03:24 -07002334 if (!(prev->vm_flags & VM_GROWSDOWN))
2335 return -ENOMEM;
2336 /* Check that both stack segments have the same anon_vma? */
2337 }
2338
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002339 /* We must make sure the anon_vma is allocated. */
2340 if (unlikely(anon_vma_prepare(vma)))
2341 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
2343 /*
2344 * vma->vm_start/vm_end cannot change under us because the caller
2345 * is required to hold the mmap_sem in read mode. We need the
2346 * anon_vma lock to serialize against concurrent expand_stacks.
2347 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002348 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349
2350 /* Somebody else might have raced and expanded it already */
2351 if (address < vma->vm_start) {
2352 unsigned long size, grow;
2353
2354 size = vma->vm_end - address;
2355 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2356
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002357 error = -ENOMEM;
2358 if (grow <= vma->vm_pgoff) {
2359 error = acct_stack_growth(vma, size, grow);
2360 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002361 /*
2362 * vma_gap_update() doesn't support concurrent
2363 * updates, but we only hold a shared mmap_sem
2364 * lock here, so we need to protect against
2365 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002366 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002367 * we don't guarantee that all growable vmas
2368 * in a mm share the same root anon vma.
2369 * So, we reuse mm->page_table_lock to guard
2370 * against concurrent vma expansions.
2371 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002372 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002373 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002374 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002375 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002376 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002377 vma->vm_start = address;
2378 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002379 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002380 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002381 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002382
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002383 perf_event_mmap(vma);
2384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 }
2386 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002387 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002388 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002389 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 return error;
2391}
2392
Hugh Dickins1be71072017-06-19 04:03:24 -07002393/* enforced gap between the expanding stack and other mappings. */
2394unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2395
2396static int __init cmdline_parse_stack_guard_gap(char *p)
2397{
2398 unsigned long val;
2399 char *endptr;
2400
2401 val = simple_strtoul(p, &endptr, 10);
2402 if (!*endptr)
2403 stack_guard_gap = val << PAGE_SHIFT;
2404
2405 return 0;
2406}
2407__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2408
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002409#ifdef CONFIG_STACK_GROWSUP
2410int expand_stack(struct vm_area_struct *vma, unsigned long address)
2411{
2412 return expand_upwards(vma, address);
2413}
2414
2415struct vm_area_struct *
2416find_extend_vma(struct mm_struct *mm, unsigned long addr)
2417{
2418 struct vm_area_struct *vma, *prev;
2419
2420 addr &= PAGE_MASK;
2421 vma = find_vma_prev(mm, addr, &prev);
2422 if (vma && (vma->vm_start <= addr))
2423 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002424 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002425 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002426 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002427 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002428 return prev;
2429}
2430#else
2431int expand_stack(struct vm_area_struct *vma, unsigned long address)
2432{
2433 return expand_downwards(vma, address);
2434}
2435
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002437find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438{
vishnu.pscc71aba2014-10-09 15:26:29 -07002439 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 unsigned long start;
2441
2442 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002443 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 if (!vma)
2445 return NULL;
2446 if (vma->vm_start <= addr)
2447 return vma;
2448 if (!(vma->vm_flags & VM_GROWSDOWN))
2449 return NULL;
2450 start = vma->vm_start;
2451 if (expand_stack(vma, addr))
2452 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002453 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002454 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 return vma;
2456}
2457#endif
2458
Jesse Barnese1d6d012014-12-12 16:55:27 -08002459EXPORT_SYMBOL_GPL(find_extend_vma);
2460
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002462 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002464 *
2465 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002467static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002469 unsigned long nr_accounted = 0;
2470
Hugh Dickins365e9c872005-10-29 18:16:18 -07002471 /* Update high watermark before we lower total_vm */
2472 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002474 long nrpages = vma_pages(vma);
2475
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002476 if (vma->vm_flags & VM_ACCOUNT)
2477 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002478 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002479 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002480 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002481 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 validate_mm(mm);
2483}
2484
2485/*
2486 * Get rid of page table information in the indicated region.
2487 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002488 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 */
2490static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002491 struct vm_area_struct *vma, struct vm_area_struct *prev,
2492 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493{
vishnu.pscc71aba2014-10-09 15:26:29 -07002494 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002495 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496
2497 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002498 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002499 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002500 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002501 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002502 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002503 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504}
2505
2506/*
2507 * Create a list of vma's touched by the unmap, removing them from the mm's
2508 * vma list as we go..
2509 */
2510static void
2511detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2512 struct vm_area_struct *prev, unsigned long end)
2513{
2514 struct vm_area_struct **insertion_point;
2515 struct vm_area_struct *tail_vma = NULL;
2516
2517 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002518 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002520 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 mm->map_count--;
2522 tail_vma = vma;
2523 vma = vma->vm_next;
2524 } while (vma && vma->vm_start < end);
2525 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002526 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002527 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002528 vma_gap_update(vma);
2529 } else
Hugh Dickins1be71072017-06-19 04:03:24 -07002530 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002532
2533 /* Kill the cache */
2534 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535}
2536
2537/*
David Rientjesdef5efe2017-02-24 14:58:47 -08002538 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2539 * has already been checked or doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 */
David Rientjesdef5efe2017-02-24 14:58:47 -08002541int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2542 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002545 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Andi Kleena5516432008-07-23 21:27:41 -07002547 if (is_vm_hugetlb_page(vma) && (addr &
2548 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 return -EINVAL;
2550
Christoph Lametere94b1762006-12-06 20:33:17 -08002551 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002553 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554
2555 /* most fields are the same, copy all, and then fixup */
2556 *new = *vma;
2557
Rik van Riel5beb4932010-03-05 13:42:07 -08002558 INIT_LIST_HEAD(&new->anon_vma_chain);
2559
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 if (new_below)
2561 new->vm_end = addr;
2562 else {
2563 new->vm_start = addr;
2564 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2565 }
2566
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002567 err = vma_dup_policy(vma, new);
2568 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002569 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002571 err = anon_vma_clone(new, vma);
2572 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002573 goto out_free_mpol;
2574
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002575 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 get_file(new->vm_file);
2577
2578 if (new->vm_ops && new->vm_ops->open)
2579 new->vm_ops->open(new);
2580
2581 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002582 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2584 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002585 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
Rik van Riel5beb4932010-03-05 13:42:07 -08002587 /* Success. */
2588 if (!err)
2589 return 0;
2590
2591 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002592 if (new->vm_ops && new->vm_ops->close)
2593 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002594 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002595 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002596 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002597 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002598 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002599 out_free_vma:
2600 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002601 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602}
2603
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002604/*
2605 * Split a vma into two pieces at address 'addr', a new vma is allocated
2606 * either for the first part or the tail.
2607 */
2608int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2609 unsigned long addr, int new_below)
2610{
2611 if (mm->map_count >= sysctl_max_map_count)
2612 return -ENOMEM;
2613
2614 return __split_vma(mm, vma, addr, new_below);
2615}
2616
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617/* Munmap is split into 2 main parts -- this part which finds
2618 * what needs doing, and the areas themselves, which do the
2619 * work. This now handles partial unmappings.
2620 * Jeremy Fitzhardinge <jeremy@goop.org>
2621 */
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002622int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2623 struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624{
2625 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002626 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002628 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 return -EINVAL;
2630
vishnu.pscc71aba2014-10-09 15:26:29 -07002631 len = PAGE_ALIGN(len);
2632 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 return -EINVAL;
2634
2635 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002636 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002637 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002639 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002640 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
2642 /* if it doesn't overlap, we have nothing.. */
2643 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002644 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 return 0;
2646
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002647 if (uf) {
2648 int error = userfaultfd_unmap_prep(vma, start, end, uf);
2649
2650 if (error)
2651 return error;
2652 }
2653
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 /*
2655 * If we need to split any vma, do it now to save pain later.
2656 *
2657 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2658 * unmapped vm_area_struct will remain in use: so lower split_vma
2659 * places tmp vma above, and higher split_vma places tmp vma below.
2660 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002661 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002662 int error;
2663
2664 /*
2665 * Make sure that map_count on return from munmap() will
2666 * not exceed its limit; but let map_count go just above
2667 * its limit temporarily, to help free resources as expected.
2668 */
2669 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2670 return -ENOMEM;
2671
2672 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 if (error)
2674 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002675 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 }
2677
2678 /* Does it split the last one? */
2679 last = find_vma(mm, end);
2680 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002681 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 if (error)
2683 return error;
2684 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002685 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686
2687 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002688 * unlock any mlock()ed ranges before detaching vmas
2689 */
2690 if (mm->locked_vm) {
2691 struct vm_area_struct *tmp = vma;
2692 while (tmp && tmp->vm_start < end) {
2693 if (tmp->vm_flags & VM_LOCKED) {
2694 mm->locked_vm -= vma_pages(tmp);
2695 munlock_vma_pages_all(tmp);
2696 }
2697 tmp = tmp->vm_next;
2698 }
2699 }
2700
2701 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 * Remove the vma's, and unmap the actual pages
2703 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002704 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2705 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Dave Hansen1de4fa12014-11-14 07:18:31 -08002707 arch_unmap(mm, vma, start, end);
2708
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002710 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
2712 return 0;
2713}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Al Virobfce2812012-04-20 21:57:04 -04002715int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002716{
2717 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002718 struct mm_struct *mm = current->mm;
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002719 LIST_HEAD(uf);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002720
Michal Hockoae798782016-05-23 16:25:33 -07002721 if (down_write_killable(&mm->mmap_sem))
2722 return -EINTR;
2723
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002724 ret = do_munmap(mm, start, len, &uf);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002725 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002726 userfaultfd_unmap_complete(mm, &uf);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002727 return ret;
2728}
2729EXPORT_SYMBOL(vm_munmap);
2730
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002731SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 profile_munmap(addr);
Mike Rapoport846b1a02017-02-24 14:58:19 -08002734 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735}
2736
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002737
2738/*
2739 * Emulation of deprecated remap_file_pages() syscall.
2740 */
2741SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2742 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2743{
2744
2745 struct mm_struct *mm = current->mm;
2746 struct vm_area_struct *vma;
2747 unsigned long populate = 0;
2748 unsigned long ret = -EINVAL;
2749 struct file *file;
2750
Joe Perches756a0252016-03-17 14:19:47 -07002751 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2752 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002753
2754 if (prot)
2755 return ret;
2756 start = start & PAGE_MASK;
2757 size = size & PAGE_MASK;
2758
2759 if (start + size <= start)
2760 return ret;
2761
2762 /* Does pgoff wrap? */
2763 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2764 return ret;
2765
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002766 if (down_write_killable(&mm->mmap_sem))
2767 return -EINTR;
2768
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002769 vma = find_vma(mm, start);
2770
2771 if (!vma || !(vma->vm_flags & VM_SHARED))
2772 goto out;
2773
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002774 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002775 goto out;
2776
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002777 if (start + size > vma->vm_end) {
2778 struct vm_area_struct *next;
2779
2780 for (next = vma->vm_next; next; next = next->vm_next) {
2781 /* hole between vmas ? */
2782 if (next->vm_start != next->vm_prev->vm_end)
2783 goto out;
2784
2785 if (next->vm_file != vma->vm_file)
2786 goto out;
2787
2788 if (next->vm_flags != vma->vm_flags)
2789 goto out;
2790
2791 if (start + size <= next->vm_end)
2792 break;
2793 }
2794
2795 if (!next)
2796 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002797 }
2798
2799 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2800 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2801 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2802
2803 flags &= MAP_NONBLOCK;
2804 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2805 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002806 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002807 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002808
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002809 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002810 for (tmp = vma; tmp->vm_start >= start + size;
2811 tmp = tmp->vm_next) {
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07002812 /*
2813 * Split pmd and munlock page on the border
2814 * of the range.
2815 */
2816 vma_adjust_trans_huge(tmp, start, start + size, 0);
2817
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002818 munlock_vma_pages_range(tmp,
2819 max(tmp->vm_start, start),
2820 min(tmp->vm_end, start + size));
2821 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002822 }
2823
2824 file = get_file(vma->vm_file);
2825 ret = do_mmap_pgoff(vma->vm_file, start, size,
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002826 prot, flags, pgoff, &populate, NULL);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002827 fput(file);
2828out:
2829 up_write(&mm->mmap_sem);
2830 if (populate)
2831 mm_populate(ret, populate);
2832 if (!IS_ERR_VALUE(ret))
2833 ret = 0;
2834 return ret;
2835}
2836
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837static inline void verify_mm_writelocked(struct mm_struct *mm)
2838{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002839#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2841 WARN_ON(1);
2842 up_read(&mm->mmap_sem);
2843 }
2844#endif
2845}
2846
2847/*
2848 * this is really a simplified "do_mmap". it only handles
2849 * anonymous maps. eventually we may be able to do some
2850 * brk-specific accounting here.
2851 */
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002852static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853{
vishnu.pscc71aba2014-10-09 15:26:29 -07002854 struct mm_struct *mm = current->mm;
2855 struct vm_area_struct *vma, *prev;
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002856 unsigned long len;
vishnu.pscc71aba2014-10-09 15:26:29 -07002857 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002859 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860
Kees Cookba093a62016-08-02 14:04:54 -07002861 len = PAGE_ALIGN(request);
2862 if (len < request)
2863 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002865 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002867 /* Until we need other flags, refuse anything except VM_EXEC. */
2868 if ((flags & (~VM_EXEC)) != 0)
2869 return -EINVAL;
2870 flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002871
Al Viro2c6a1012009-12-03 19:40:46 -05002872 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002873 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002874 return error;
2875
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002876 error = mlock_future_check(mm, mm->def_flags, len);
2877 if (error)
2878 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879
2880 /*
2881 * mm->mmap_sem is required to protect against another thread
2882 * changing the mappings in case we sleep.
2883 */
2884 verify_mm_writelocked(mm);
2885
2886 /*
2887 * Clear old maps. this also does some error checking for us
2888 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002889 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2890 &rb_parent)) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002891 if (do_munmap(mm, addr, len, uf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 }
2894
2895 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002896 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 return -ENOMEM;
2898
2899 if (mm->map_count > sysctl_max_map_count)
2900 return -ENOMEM;
2901
Al Viro191c5422012-02-13 03:58:52 +00002902 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 return -ENOMEM;
2904
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002906 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002907 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002908 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 goto out;
2910
2911 /*
2912 * create a vma struct for an anonymous mapping
2913 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002914 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 if (!vma) {
2916 vm_unacct_memory(len >> PAGE_SHIFT);
2917 return -ENOMEM;
2918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
Rik van Riel5beb4932010-03-05 13:42:07 -08002920 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 vma->vm_mm = mm;
2922 vma->vm_start = addr;
2923 vma->vm_end = addr + len;
2924 vma->vm_pgoff = pgoff;
2925 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002926 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 vma_link(mm, vma, prev, rb_link, rb_parent);
2928out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002929 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002931 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002932 if (flags & VM_LOCKED)
2933 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002934 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002935 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936}
2937
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002938static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf)
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002939{
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002940 return do_brk_flags(addr, len, 0, uf);
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002941}
2942
2943int vm_brk_flags(unsigned long addr, unsigned long len, unsigned long flags)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002944{
2945 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002946 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002947 bool populate;
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002948 LIST_HEAD(uf);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002949
Michal Hocko2d6c9282016-05-23 16:25:42 -07002950 if (down_write_killable(&mm->mmap_sem))
2951 return -EINTR;
2952
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002953 ret = do_brk_flags(addr, len, flags, &uf);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002954 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002955 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002956 userfaultfd_unmap_complete(mm, &uf);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002957 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002958 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002959 return ret;
2960}
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002961EXPORT_SYMBOL(vm_brk_flags);
2962
2963int vm_brk(unsigned long addr, unsigned long len)
2964{
2965 return vm_brk_flags(addr, len, 0);
2966}
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002967EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968
2969/* Release all mmaps. */
2970void exit_mmap(struct mm_struct *mm)
2971{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002972 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002973 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 unsigned long nr_accounted = 0;
2975
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002976 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002977 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002978
Rik van Rielba470de2008-10-18 20:26:50 -07002979 if (mm->locked_vm) {
2980 vma = mm->mmap;
2981 while (vma) {
2982 if (vma->vm_flags & VM_LOCKED)
2983 munlock_vma_pages_all(vma);
2984 vma = vma->vm_next;
2985 }
2986 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002987
2988 arch_exit_mmap(mm);
2989
Rik van Rielba470de2008-10-18 20:26:50 -07002990 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002991 if (!vma) /* Can happen if dup_mmap() received an OOM */
2992 return;
2993
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002996 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002997 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002998 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002999 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07003000
Hugh Dickins6ee86302013-04-29 15:07:44 -07003001 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05003002 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07003005 * Walk the list again, actually closing and freeing it,
3006 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003008 while (vma) {
3009 if (vma->vm_flags & VM_ACCOUNT)
3010 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07003011 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003012 }
3013 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014}
3015
3016/* Insert vm structure into process list sorted by address
3017 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003018 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 */
Hugh Dickins6597d782012-10-08 16:29:07 -07003020int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021{
Hugh Dickins6597d782012-10-08 16:29:07 -07003022 struct vm_area_struct *prev;
3023 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024
Chen Gangc9d13f52015-09-08 15:04:08 -07003025 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3026 &prev, &rb_link, &rb_parent))
3027 return -ENOMEM;
3028 if ((vma->vm_flags & VM_ACCOUNT) &&
3029 security_vm_enough_memory_mm(mm, vma_pages(vma)))
3030 return -ENOMEM;
3031
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 /*
3033 * The vm_pgoff of a purely anonymous vma should be irrelevant
3034 * until its first write fault, when page's anon_vma and index
3035 * are set. But now set the vm_pgoff it will almost certainly
3036 * end up with (unless mremap moves it elsewhere before that
3037 * first wfault), so /proc/pid/maps tells a consistent story.
3038 *
3039 * By setting it to reflect the virtual start address of the
3040 * vma, merges and splits can happen in a seamless way, just
3041 * using the existing file pgoff checks and manipulations.
3042 * Similarly in do_mmap_pgoff and in do_brk.
3043 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003044 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 BUG_ON(vma->anon_vma);
3046 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3047 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05303048
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 vma_link(mm, vma, prev, rb_link, rb_parent);
3050 return 0;
3051}
3052
3053/*
3054 * Copy the vma structure to a new location in the same mm,
3055 * prior to moving page table entries, to effect an mremap move.
3056 */
3057struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07003058 unsigned long addr, unsigned long len, pgoff_t pgoff,
3059 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060{
3061 struct vm_area_struct *vma = *vmap;
3062 unsigned long vma_start = vma->vm_start;
3063 struct mm_struct *mm = vma->vm_mm;
3064 struct vm_area_struct *new_vma, *prev;
3065 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003066 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
3068 /*
3069 * If anonymous vma has not yet been faulted, update new pgoff
3070 * to match new location, to increase its chance of merging.
3071 */
Oleg Nesterovce757992015-09-08 14:58:34 -07003072 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003074 faulted_in_anon_vma = false;
3075 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076
Hugh Dickins6597d782012-10-08 16:29:07 -07003077 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3078 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07003080 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3081 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 if (new_vma) {
3083 /*
3084 * Source vma may have been merged into new_vma
3085 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003086 if (unlikely(vma_start >= new_vma->vm_start &&
3087 vma_start < new_vma->vm_end)) {
3088 /*
3089 * The only way we can get a vma_merge with
3090 * self during an mremap is if the vma hasn't
3091 * been faulted in yet and we were allowed to
3092 * reset the dst vma->vm_pgoff to the
3093 * destination address of the mremap to allow
3094 * the merge to happen. mremap must change the
3095 * vm_pgoff linearity between src and dst vmas
3096 * (in turn preventing a vma_merge) to be
3097 * safe. It is only safe to keep the vm_pgoff
3098 * linear if there are no pages mapped yet.
3099 */
Sasha Levin81d1b092014-10-09 15:28:10 -07003100 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07003101 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07003102 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07003103 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08003105 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07003106 if (!new_vma)
3107 goto out;
3108 *new_vma = *vma;
3109 new_vma->vm_start = addr;
3110 new_vma->vm_end = addr + len;
3111 new_vma->vm_pgoff = pgoff;
3112 if (vma_dup_policy(vma, new_vma))
3113 goto out_free_vma;
3114 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3115 if (anon_vma_clone(new_vma, vma))
3116 goto out_free_mempol;
3117 if (new_vma->vm_file)
3118 get_file(new_vma->vm_file);
3119 if (new_vma->vm_ops && new_vma->vm_ops->open)
3120 new_vma->vm_ops->open(new_vma);
3121 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3122 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 }
3124 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003125
Chen Gange3975892015-09-08 15:03:38 -07003126out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07003127 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07003128out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08003129 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07003130out:
Rik van Riel5beb4932010-03-05 13:42:07 -08003131 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003133
3134/*
3135 * Return true if the calling process may expand its vm space by the passed
3136 * number of pages
3137 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003138bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003139{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003140 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3141 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003142
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003143 if (is_data_mapping(flags) &&
3144 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003145 /* Workaround for Valgrind */
3146 if (rlimit(RLIMIT_DATA) == 0 &&
3147 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3148 return true;
3149 if (!ignore_rlimit_data) {
3150 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 -08003151 current->comm, current->pid,
3152 (mm->data_vm + npages) << PAGE_SHIFT,
3153 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003154 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003155 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003156 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07003157
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003158 return true;
3159}
3160
3161void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3162{
3163 mm->total_vm += npages;
3164
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003165 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003166 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003167 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003168 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003169 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003170 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003171}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003172
Dave Jiang11bac802017-02-24 14:56:41 -08003173static int special_mapping_fault(struct vm_fault *vmf);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003174
3175/*
3176 * Having a close hook prevents vma merging regardless of flags.
3177 */
3178static void special_mapping_close(struct vm_area_struct *vma)
3179{
3180}
3181
3182static const char *special_mapping_name(struct vm_area_struct *vma)
3183{
3184 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3185}
3186
Dmitry Safonovb059a452016-06-28 14:35:38 +03003187static int special_mapping_mremap(struct vm_area_struct *new_vma)
3188{
3189 struct vm_special_mapping *sm = new_vma->vm_private_data;
3190
Dmitry Safonov280e87e2017-06-19 17:32:42 +01003191 if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3192 return -EFAULT;
3193
Dmitry Safonovb059a452016-06-28 14:35:38 +03003194 if (sm->mremap)
3195 return sm->mremap(sm, new_vma);
Dmitry Safonov280e87e2017-06-19 17:32:42 +01003196
Dmitry Safonovb059a452016-06-28 14:35:38 +03003197 return 0;
3198}
3199
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003200static const struct vm_operations_struct special_mapping_vmops = {
3201 .close = special_mapping_close,
3202 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03003203 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003204 .name = special_mapping_name,
3205};
3206
3207static const struct vm_operations_struct legacy_special_mapping_vmops = {
3208 .close = special_mapping_close,
3209 .fault = special_mapping_fault,
3210};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003211
Dave Jiang11bac802017-02-24 14:56:41 -08003212static int special_mapping_fault(struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003213{
Dave Jiang11bac802017-02-24 14:56:41 -08003214 struct vm_area_struct *vma = vmf->vma;
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003215 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003216 struct page **pages;
3217
Andy Lutomirskif872f542015-12-29 20:12:19 -08003218 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003219 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08003220 } else {
3221 struct vm_special_mapping *sm = vma->vm_private_data;
3222
3223 if (sm->fault)
Dave Jiang11bac802017-02-24 14:56:41 -08003224 return sm->fault(sm, vmf->vma, vmf);
Andy Lutomirskif872f542015-12-29 20:12:19 -08003225
3226 pages = sm->pages;
3227 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003228
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003229 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003230 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003231
3232 if (*pages) {
3233 struct page *page = *pages;
3234 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003235 vmf->page = page;
3236 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003237 }
3238
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003239 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003240}
3241
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003242static struct vm_area_struct *__install_special_mapping(
3243 struct mm_struct *mm,
3244 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003245 unsigned long vm_flags, void *priv,
3246 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003247{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003248 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003249 struct vm_area_struct *vma;
3250
3251 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3252 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003253 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003254
Rik van Riel5beb4932010-03-05 13:42:07 -08003255 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003256 vma->vm_mm = mm;
3257 vma->vm_start = addr;
3258 vma->vm_end = addr + len;
3259
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003260 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003261 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003262
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003263 vma->vm_ops = ops;
3264 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003265
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003266 ret = insert_vm_struct(mm, vma);
3267 if (ret)
3268 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003269
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003270 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003271
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003272 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003273
Stefani Seibold3935ed62014-03-17 23:22:02 +01003274 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003275
3276out:
3277 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003278 return ERR_PTR(ret);
3279}
3280
Dmitry Safonov2eefd872016-09-05 16:33:05 +03003281bool vma_is_special_mapping(const struct vm_area_struct *vma,
3282 const struct vm_special_mapping *sm)
3283{
3284 return vma->vm_private_data == sm &&
3285 (vma->vm_ops == &special_mapping_vmops ||
3286 vma->vm_ops == &legacy_special_mapping_vmops);
3287}
3288
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003289/*
3290 * Called with mm->mmap_sem held for writing.
3291 * Insert a new vma covering the given region, with the given flags.
3292 * Its pages are supplied by the given array of struct page *.
3293 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3294 * The region past the last page supplied will always produce SIGBUS.
3295 * The array pointer and the pages it points to are assumed to stay alive
3296 * for as long as this mapping might exist.
3297 */
3298struct vm_area_struct *_install_special_mapping(
3299 struct mm_struct *mm,
3300 unsigned long addr, unsigned long len,
3301 unsigned long vm_flags, const struct vm_special_mapping *spec)
3302{
Chen Gang27f28b92015-11-05 18:48:41 -08003303 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3304 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003305}
3306
Stefani Seibold3935ed62014-03-17 23:22:02 +01003307int install_special_mapping(struct mm_struct *mm,
3308 unsigned long addr, unsigned long len,
3309 unsigned long vm_flags, struct page **pages)
3310{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003311 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003312 mm, addr, len, vm_flags, (void *)pages,
3313 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003314
Duan Jiong14bd5b42014-06-04 16:07:05 -07003315 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003316}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003317
3318static DEFINE_MUTEX(mm_all_locks_mutex);
3319
Peter Zijlstra454ed842008-08-11 09:30:25 +02003320static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003321{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003322 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003323 /*
3324 * The LSB of head.next can't change from under us
3325 * because we hold the mm_all_locks_mutex.
3326 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003327 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003328 /*
3329 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003330 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003331 * the same anon_vma we won't take it again.
3332 *
3333 * No need of atomic instructions here, head.next
3334 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003335 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003336 */
3337 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003338 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003339 BUG();
3340 }
3341}
3342
Peter Zijlstra454ed842008-08-11 09:30:25 +02003343static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003344{
3345 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3346 /*
3347 * AS_MM_ALL_LOCKS can't change from under us because
3348 * we hold the mm_all_locks_mutex.
3349 *
3350 * Operations on ->flags have to be atomic because
3351 * even if AS_MM_ALL_LOCKS is stable thanks to the
3352 * mm_all_locks_mutex, there may be other cpus
3353 * changing other bitflags in parallel to us.
3354 */
3355 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3356 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003357 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003358 }
3359}
3360
3361/*
3362 * This operation locks against the VM for all pte/vma/mm related
3363 * operations that could ever happen on a certain mm. This includes
3364 * vmtruncate, try_to_unmap, and all page faults.
3365 *
3366 * The caller must take the mmap_sem in write mode before calling
3367 * mm_take_all_locks(). The caller isn't allowed to release the
3368 * mmap_sem until mm_drop_all_locks() returns.
3369 *
3370 * mmap_sem in write mode is required in order to block all operations
3371 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003372 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003373 * anon_vmas to be associated with existing vmas.
3374 *
3375 * A single task can't take more than one mm_take_all_locks() in a row
3376 * or it would deadlock.
3377 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003378 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003379 * mapping->flags avoid to take the same lock twice, if more than one
3380 * vma in this mm is backed by the same anon_vma or address_space.
3381 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003382 * We take locks in following order, accordingly to comment at beginning
3383 * of mm/rmap.c:
3384 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3385 * hugetlb mapping);
3386 * - all i_mmap_rwsem locks;
3387 * - all anon_vma->rwseml
3388 *
3389 * We can take all locks within these types randomly because the VM code
3390 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3391 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003392 *
3393 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3394 * that may have to take thousand of locks.
3395 *
3396 * mm_take_all_locks() can fail if it's interrupted by signals.
3397 */
3398int mm_take_all_locks(struct mm_struct *mm)
3399{
3400 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003401 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003402
3403 BUG_ON(down_read_trylock(&mm->mmap_sem));
3404
3405 mutex_lock(&mm_all_locks_mutex);
3406
3407 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3408 if (signal_pending(current))
3409 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003410 if (vma->vm_file && vma->vm_file->f_mapping &&
3411 is_vm_hugetlb_page(vma))
3412 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3413 }
3414
3415 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3416 if (signal_pending(current))
3417 goto out_unlock;
3418 if (vma->vm_file && vma->vm_file->f_mapping &&
3419 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003420 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003421 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003422
3423 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3424 if (signal_pending(current))
3425 goto out_unlock;
3426 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003427 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3428 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003429 }
3430
Kautuk Consul584cff52011-10-31 17:08:59 -07003431 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003432
3433out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003434 mm_drop_all_locks(mm);
3435 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003436}
3437
3438static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3439{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003440 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003441 /*
3442 * The LSB of head.next can't change to 0 from under
3443 * us because we hold the mm_all_locks_mutex.
3444 *
3445 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003446 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003447 * never see our bitflag.
3448 *
3449 * No need of atomic instructions here, head.next
3450 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003451 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003452 */
3453 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003454 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003455 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003456 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003457 }
3458}
3459
3460static void vm_unlock_mapping(struct address_space *mapping)
3461{
3462 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3463 /*
3464 * AS_MM_ALL_LOCKS can't change to 0 from under us
3465 * because we hold the mm_all_locks_mutex.
3466 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003467 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003468 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3469 &mapping->flags))
3470 BUG();
3471 }
3472}
3473
3474/*
3475 * The mmap_sem cannot be released by the caller until
3476 * mm_drop_all_locks() returns.
3477 */
3478void mm_drop_all_locks(struct mm_struct *mm)
3479{
3480 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003481 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003482
3483 BUG_ON(down_read_trylock(&mm->mmap_sem));
3484 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3485
3486 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3487 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003488 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3489 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003490 if (vma->vm_file && vma->vm_file->f_mapping)
3491 vm_unlock_mapping(vma->vm_file->f_mapping);
3492 }
3493
3494 mutex_unlock(&mm_all_locks_mutex);
3495}
David Howells8feae132009-01-08 12:04:47 +00003496
3497/*
seokhoon.yoon3edf41d2017-02-24 14:56:44 -08003498 * initialise the percpu counter for VM
David Howells8feae132009-01-08 12:04:47 +00003499 */
3500void __init mmap_init(void)
3501{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003502 int ret;
3503
Tejun Heo908c7f12014-09-08 09:51:29 +09003504 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003505 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003506}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003507
3508/*
3509 * Initialise sysctl_user_reserve_kbytes.
3510 *
3511 * This is intended to prevent a user from starting a single memory hogging
3512 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3513 * mode.
3514 *
3515 * The default value is min(3% of free memory, 128MB)
3516 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3517 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003518static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003519{
3520 unsigned long free_kbytes;
3521
3522 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3523
3524 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3525 return 0;
3526}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003527subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003528
3529/*
3530 * Initialise sysctl_admin_reserve_kbytes.
3531 *
3532 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3533 * to log in and kill a memory hogging process.
3534 *
3535 * Systems with more than 256MB will reserve 8MB, enough to recover
3536 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3537 * only reserve 3% of free pages by default.
3538 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003539static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003540{
3541 unsigned long free_kbytes;
3542
3543 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3544
3545 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3546 return 0;
3547}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003548subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003549
3550/*
3551 * Reinititalise user and admin reserves if memory is added or removed.
3552 *
3553 * The default user reserve max is 128MB, and the default max for the
3554 * admin reserve is 8MB. These are usually, but not always, enough to
3555 * enable recovery from a memory hogging process using login/sshd, a shell,
3556 * and tools like top. It may make sense to increase or even disable the
3557 * reserve depending on the existence of swap or variations in the recovery
3558 * tools. So, the admin may have changed them.
3559 *
3560 * If memory is added and the reserves have been eliminated or increased above
3561 * the default max, then we'll trust the admin.
3562 *
3563 * If memory is removed and there isn't enough free memory, then we
3564 * need to reset the reserves.
3565 *
3566 * Otherwise keep the reserve set by the admin.
3567 */
3568static int reserve_mem_notifier(struct notifier_block *nb,
3569 unsigned long action, void *data)
3570{
3571 unsigned long tmp, free_kbytes;
3572
3573 switch (action) {
3574 case MEM_ONLINE:
3575 /* Default max is 128MB. Leave alone if modified by operator. */
3576 tmp = sysctl_user_reserve_kbytes;
3577 if (0 < tmp && tmp < (1UL << 17))
3578 init_user_reserve();
3579
3580 /* Default max is 8MB. Leave alone if modified by operator. */
3581 tmp = sysctl_admin_reserve_kbytes;
3582 if (0 < tmp && tmp < (1UL << 13))
3583 init_admin_reserve();
3584
3585 break;
3586 case MEM_OFFLINE:
3587 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3588
3589 if (sysctl_user_reserve_kbytes > free_kbytes) {
3590 init_user_reserve();
3591 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3592 sysctl_user_reserve_kbytes);
3593 }
3594
3595 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3596 init_admin_reserve();
3597 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3598 sysctl_admin_reserve_kbytes);
3599 }
3600 break;
3601 default:
3602 break;
3603 }
3604 return NOTIFY_OK;
3605}
3606
3607static struct notifier_block reserve_mem_nb = {
3608 .notifier_call = reserve_mem_notifier,
3609};
3610
3611static int __meminit init_reserve_notifier(void)
3612{
3613 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003614 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003615
3616 return 0;
3617}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003618subsys_initcall(init_reserve_notifier);