blob: dc4291dcc99b8f245fe38aeb52bb5aa5fbe3243d [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 */
97pgprot_t protection_map[16] = {
98 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
99 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
100};
101
Hugh Dickins804af2c2006-07-26 21:39:49 +0100102pgprot_t vm_get_page_prot(unsigned long vm_flags)
103{
Dave Kleikampb845f312008-07-08 00:28:51 +1000104 return __pgprot(pgprot_val(protection_map[vm_flags &
105 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
106 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100107}
108EXPORT_SYMBOL(vm_get_page_prot);
109
Peter Feiner64e45502014-10-13 15:55:46 -0700110static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
111{
112 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
113}
114
115/* Update vma->vm_page_prot to reflect vma->vm_flags. */
116void vma_set_page_prot(struct vm_area_struct *vma)
117{
118 unsigned long vm_flags = vma->vm_flags;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700119 pgprot_t vm_page_prot;
Peter Feiner64e45502014-10-13 15:55:46 -0700120
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700121 vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
122 if (vma_wants_writenotify(vma, vm_page_prot)) {
Peter Feiner64e45502014-10-13 15:55:46 -0700123 vm_flags &= ~VM_SHARED;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700124 vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
Peter Feiner64e45502014-10-13 15:55:46 -0700125 }
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700126 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
127 WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
Peter Feiner64e45502014-10-13 15:55:46 -0700128}
129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800131 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 */
133static void __remove_shared_vm_struct(struct vm_area_struct *vma,
134 struct file *file, struct address_space *mapping)
135{
136 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500137 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700139 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
141 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800142 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 flush_dcache_mmap_unlock(mapping);
144}
145
146/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700147 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700148 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700150void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
152 struct file *file = vma->vm_file;
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 if (file) {
155 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800156 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800158 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700160}
161
162/*
163 * Close a vm structure and free it, returning the next.
164 */
165static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
166{
167 struct vm_area_struct *next = vma->vm_next;
168
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700169 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 if (vma->vm_ops && vma->vm_ops->close)
171 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700172 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700173 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700174 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700176 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700179static int do_brk(unsigned long addr, unsigned long len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700180
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100181SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700183 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 unsigned long newbrk, oldbrk;
185 struct mm_struct *mm = current->mm;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700186 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800187 bool populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700189 if (down_write_killable(&mm->mmap_sem))
190 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700192#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800193 /*
194 * CONFIG_COMPAT_BRK can still be overridden by setting
195 * randomize_va_space to 2, which will still cause mm->start_brk
196 * to be arbitrarily shifted
197 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700198 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800199 min_brk = mm->start_brk;
200 else
201 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700202#else
203 min_brk = mm->start_brk;
204#endif
205 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700207
208 /*
209 * Check against rlimit here. If this check is done later after the test
210 * of oldbrk with newbrk then it can escape the test and let the data
211 * segment grow beyond its set limit the in case where the limit is
212 * not page aligned -Ram Gupta
213 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700214 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
215 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700216 goto out;
217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 newbrk = PAGE_ALIGN(brk);
219 oldbrk = PAGE_ALIGN(mm->brk);
220 if (oldbrk == newbrk)
221 goto set_brk;
222
223 /* Always allow shrinking brk. */
224 if (brk <= mm->brk) {
225 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
226 goto set_brk;
227 goto out;
228 }
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 /* Check against existing mmap mappings. */
231 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
232 goto out;
233
234 /* Ok, looks good - let it rip. */
Linus Torvalds5d22fc22016-05-27 15:57:31 -0700235 if (do_brk(oldbrk, newbrk-oldbrk) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238set_brk:
239 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800240 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
241 up_write(&mm->mmap_sem);
242 if (populate)
243 mm_populate(oldbrk, newbrk - oldbrk);
244 return brk;
245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246out:
247 retval = mm->brk;
248 up_write(&mm->mmap_sem);
249 return retval;
250}
251
Michel Lespinassed3737182012-12-11 16:01:38 -0800252static long vma_compute_subtree_gap(struct vm_area_struct *vma)
253{
254 unsigned long max, subtree_gap;
255 max = vma->vm_start;
256 if (vma->vm_prev)
257 max -= vma->vm_prev->vm_end;
258 if (vma->vm_rb.rb_left) {
259 subtree_gap = rb_entry(vma->vm_rb.rb_left,
260 struct vm_area_struct, vm_rb)->rb_subtree_gap;
261 if (subtree_gap > max)
262 max = subtree_gap;
263 }
264 if (vma->vm_rb.rb_right) {
265 subtree_gap = rb_entry(vma->vm_rb.rb_right,
266 struct vm_area_struct, vm_rb)->rb_subtree_gap;
267 if (subtree_gap > max)
268 max = subtree_gap;
269 }
270 return max;
271}
272
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700273#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800274static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800276 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800277 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 struct rb_node *nd, *pn = NULL;
279 unsigned long prev = 0, pend = 0;
280
281 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
282 struct vm_area_struct *vma;
283 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800284 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700285 pr_emerg("vm_start %lx < prev %lx\n",
286 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800287 bug = 1;
288 }
289 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700290 pr_emerg("vm_start %lx < pend %lx\n",
291 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800292 bug = 1;
293 }
294 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700295 pr_emerg("vm_start %lx > vm_end %lx\n",
296 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800297 bug = 1;
298 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800299 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800300 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700301 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800302 vma->rb_subtree_gap,
303 vma_compute_subtree_gap(vma));
304 bug = 1;
305 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800306 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 i++;
308 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800309 prev = vma->vm_start;
310 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
312 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800313 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800315 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700316 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800317 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800319 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320}
321
Michel Lespinassed3737182012-12-11 16:01:38 -0800322static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
323{
324 struct rb_node *nd;
325
326 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
327 struct vm_area_struct *vma;
328 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700329 VM_BUG_ON_VMA(vma != ignore &&
330 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
331 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700335static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
337 int bug = 0;
338 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800339 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700340 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700341
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700342 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800343 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700344 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700345
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800346 if (anon_vma) {
347 anon_vma_lock_read(anon_vma);
348 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
349 anon_vma_interval_tree_verify(avc);
350 anon_vma_unlock_read(anon_vma);
351 }
352
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800353 highest_address = vma->vm_end;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700354 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 i++;
356 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800357 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700358 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800359 bug = 1;
360 }
361 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700362 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700363 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800364 bug = 1;
365 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800366 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800367 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700368 if (i != -1)
369 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800370 bug = 1;
371 }
Sasha Levin96dad672014-10-09 15:28:39 -0700372 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800375#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376#define validate_mm(mm) do { } while (0)
377#endif
378
Michel Lespinassed3737182012-12-11 16:01:38 -0800379RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
380 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
381
382/*
383 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
384 * vma->vm_prev->vm_end values changed, without modifying the vma's position
385 * in the rbtree.
386 */
387static void vma_gap_update(struct vm_area_struct *vma)
388{
389 /*
390 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
391 * function that does exacltly what we want.
392 */
393 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
394}
395
396static inline void vma_rb_insert(struct vm_area_struct *vma,
397 struct rb_root *root)
398{
399 /* All rb_subtree_gap values must be consistent prior to insertion */
400 validate_mm_rb(root, NULL);
401
402 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
403}
404
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700405static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
406{
407 /*
408 * Note rb_erase_augmented is a fairly large inline function,
409 * so make sure we instantiate it only once with our desired
410 * augmented rbtree callbacks.
411 */
412 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
413}
414
415static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
416 struct rb_root *root,
417 struct vm_area_struct *ignore)
418{
419 /*
420 * All rb_subtree_gap values must be consistent prior to erase,
421 * with the possible exception of the "next" vma being erased if
422 * next->vm_start was reduced.
423 */
424 validate_mm_rb(root, ignore);
425
426 __vma_rb_erase(vma, root);
427}
428
429static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
430 struct rb_root *root)
Michel Lespinassed3737182012-12-11 16:01:38 -0800431{
432 /*
433 * All rb_subtree_gap values must be consistent prior to erase,
434 * with the possible exception of the vma being erased.
435 */
436 validate_mm_rb(root, vma);
437
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700438 __vma_rb_erase(vma, root);
Michel Lespinassed3737182012-12-11 16:01:38 -0800439}
440
Michel Lespinassebf181b92012-10-08 16:31:39 -0700441/*
442 * vma has some anon_vma assigned, and is already inserted on that
443 * anon_vma's interval trees.
444 *
445 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
446 * vma must be removed from the anon_vma's interval trees using
447 * anon_vma_interval_tree_pre_update_vma().
448 *
449 * After the update, the vma will be reinserted using
450 * anon_vma_interval_tree_post_update_vma().
451 *
452 * The entire update must be protected by exclusive mmap_sem and by
453 * the root anon_vma's mutex.
454 */
455static inline void
456anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
457{
458 struct anon_vma_chain *avc;
459
460 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
461 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
462}
463
464static inline void
465anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
466{
467 struct anon_vma_chain *avc;
468
469 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
470 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
471}
472
Hugh Dickins6597d782012-10-08 16:29:07 -0700473static int find_vma_links(struct mm_struct *mm, unsigned long addr,
474 unsigned long end, struct vm_area_struct **pprev,
475 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
Hugh Dickins6597d782012-10-08 16:29:07 -0700477 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 __rb_link = &mm->mm_rb.rb_node;
480 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 while (*__rb_link) {
483 struct vm_area_struct *vma_tmp;
484
485 __rb_parent = *__rb_link;
486 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
487
488 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700489 /* Fail if an existing vma overlaps the area */
490 if (vma_tmp->vm_start < end)
491 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 __rb_link = &__rb_parent->rb_left;
493 } else {
494 rb_prev = __rb_parent;
495 __rb_link = &__rb_parent->rb_right;
496 }
497 }
498
499 *pprev = NULL;
500 if (rb_prev)
501 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
502 *rb_link = __rb_link;
503 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700504 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
506
Cyril Hrubise8420a82013-04-29 15:08:33 -0700507static unsigned long count_vma_pages_range(struct mm_struct *mm,
508 unsigned long addr, unsigned long end)
509{
510 unsigned long nr_pages = 0;
511 struct vm_area_struct *vma;
512
513 /* Find first overlaping mapping */
514 vma = find_vma_intersection(mm, addr, end);
515 if (!vma)
516 return 0;
517
518 nr_pages = (min(end, vma->vm_end) -
519 max(addr, vma->vm_start)) >> PAGE_SHIFT;
520
521 /* Iterate over the rest of the overlaps */
522 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
523 unsigned long overlap_len;
524
525 if (vma->vm_start > end)
526 break;
527
528 overlap_len = min(end, vma->vm_end) - vma->vm_start;
529 nr_pages += overlap_len >> PAGE_SHIFT;
530 }
531
532 return nr_pages;
533}
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
536 struct rb_node **rb_link, struct rb_node *rb_parent)
537{
Michel Lespinassed3737182012-12-11 16:01:38 -0800538 /* Update tracking information for the gap following the new vma. */
539 if (vma->vm_next)
540 vma_gap_update(vma->vm_next);
541 else
542 mm->highest_vm_end = vma->vm_end;
543
544 /*
545 * vma->vm_prev wasn't known when we followed the rbtree to find the
546 * correct insertion point for that vma. As a result, we could not
547 * update the vma vm_rb parents rb_subtree_gap values on the way down.
548 * So, we first insert the vma with a zero rb_subtree_gap value
549 * (to be consistent with what we did on the way down), and then
550 * immediately update the gap to the correct value. Finally we
551 * rebalance the rbtree after all augmented values have been set.
552 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800554 vma->rb_subtree_gap = 0;
555 vma_gap_update(vma);
556 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700559static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
ZhenwenXu48aae422009-01-06 14:40:21 -0800561 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
563 file = vma->vm_file;
564 if (file) {
565 struct address_space *mapping = file->f_mapping;
566
567 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500568 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700570 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800573 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 flush_dcache_mmap_unlock(mapping);
575 }
576}
577
578static void
579__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
580 struct vm_area_struct *prev, struct rb_node **rb_link,
581 struct rb_node *rb_parent)
582{
583 __vma_link_list(mm, vma, prev, rb_parent);
584 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
587static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
588 struct vm_area_struct *prev, struct rb_node **rb_link,
589 struct rb_node *rb_parent)
590{
591 struct address_space *mapping = NULL;
592
Huang Shijie64ac4942014-06-04 16:07:33 -0700593 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800595 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598 __vma_link(mm, vma, prev, rb_link, rb_parent);
599 __vma_link_file(vma);
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800602 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604 mm->map_count++;
605 validate_mm(mm);
606}
607
608/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700609 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700610 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800612static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
Hugh Dickins6597d782012-10-08 16:29:07 -0700614 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800615 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Hugh Dickins6597d782012-10-08 16:29:07 -0700617 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
618 &prev, &rb_link, &rb_parent))
619 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 __vma_link(mm, vma, prev, rb_link, rb_parent);
621 mm->map_count++;
622}
623
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700624static __always_inline void __vma_unlink_common(struct mm_struct *mm,
625 struct vm_area_struct *vma,
626 struct vm_area_struct *prev,
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700627 bool has_prev,
628 struct vm_area_struct *ignore)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Michel Lespinassed3737182012-12-11 16:01:38 -0800630 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700631
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700632 vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700633 next = vma->vm_next;
634 if (has_prev)
635 prev->vm_next = next;
636 else {
637 prev = vma->vm_prev;
638 if (prev)
639 prev->vm_next = next;
640 else
641 mm->mmap = next;
642 }
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700643 if (next)
644 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700645
646 /* Kill the cache */
647 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
649
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700650static inline void __vma_unlink_prev(struct mm_struct *mm,
651 struct vm_area_struct *vma,
652 struct vm_area_struct *prev)
653{
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700654 __vma_unlink_common(mm, vma, prev, true, vma);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700655}
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657/*
658 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
659 * is already present in an i_mmap tree without adjusting the tree.
660 * The following helper function should be used when such adjustments
661 * are necessary. The "insert" vma (if any) is to be inserted
662 * before we drop the necessary locks.
663 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700664int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
665 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
666 struct vm_area_struct *expand)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
668 struct mm_struct *mm = vma->vm_mm;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700669 struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 struct address_space *mapping = NULL;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700671 struct rb_root *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700672 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800674 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 long adjust_next = 0;
676 int remove_next = 0;
677
678 if (next && !insert) {
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700679 struct vm_area_struct *exporter = NULL, *importer = NULL;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 if (end >= next->vm_end) {
682 /*
683 * vma expands, overlapping all the next, and
684 * perhaps the one after too (mprotect case 6).
Andrea Arcangeli86d12e42016-10-07 17:01:34 -0700685 * The only other cases that gets here are
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700686 * case 1, case 7 and case 8.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700688 if (next == expand) {
689 /*
690 * The only case where we don't expand "vma"
691 * and we expand "next" instead is case 8.
692 */
693 VM_WARN_ON(end != next->vm_end);
694 /*
695 * remove_next == 3 means we're
696 * removing "vma" and that to do so we
697 * swapped "vma" and "next".
698 */
699 remove_next = 3;
700 VM_WARN_ON(file != next->vm_file);
701 swap(vma, next);
702 } else {
703 VM_WARN_ON(expand != vma);
704 /*
705 * case 1, 6, 7, remove_next == 2 is case 6,
706 * remove_next == 1 is case 1 or 7.
707 */
708 remove_next = 1 + (end > next->vm_end);
709 VM_WARN_ON(remove_next == 2 &&
710 end != next->vm_next->vm_end);
711 VM_WARN_ON(remove_next == 1 &&
712 end != next->vm_end);
713 /* trim end to next, for case 6 first pass */
714 end = next->vm_end;
715 }
716
Linus Torvalds287d97a2010-04-10 15:22:30 -0700717 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 importer = vma;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700719
720 /*
721 * If next doesn't have anon_vma, import from vma after
722 * next, if the vma overlaps with it.
723 */
Andrea Arcangeli97a42cd2016-10-07 17:01:31 -0700724 if (remove_next == 2 && !next->anon_vma)
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700725 exporter = next->vm_next;
726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 } else if (end > next->vm_start) {
728 /*
729 * vma expands, overlapping part of the next:
730 * mprotect case 5 shifting the boundary up.
731 */
732 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700733 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 importer = vma;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700735 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 } else if (end < vma->vm_end) {
737 /*
738 * vma shrinks, and !insert tells it's not
739 * split_vma inserting another: so it must be
740 * mprotect case 4 shifting the boundary down.
741 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700742 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700743 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 importer = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700745 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Rik van Riel5beb4932010-03-05 13:42:07 -0800748 /*
749 * Easily overlooked: when mprotect shifts the boundary,
750 * make sure the expanding vma has anon_vma set if the
751 * shrinking vma had, to cover any anon pages imported.
752 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700753 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800754 int error;
755
Linus Torvalds287d97a2010-04-10 15:22:30 -0700756 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300757 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700758 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300759 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800760 }
761 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700762again:
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700763 vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
Kirill A. Shutemov37f9f552016-07-26 15:25:48 -0700764
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (file) {
766 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800767 root = &mapping->i_mmap;
768 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530769
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800770 if (adjust_next)
771 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530772
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800773 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700776 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 * are visible to arm/parisc __flush_dcache_page
778 * throughout; but we cannot insert into address
779 * space until vma start or end is updated.
780 */
781 __vma_link_file(insert);
782 }
783 }
784
Michel Lespinassebf181b92012-10-08 16:31:39 -0700785 anon_vma = vma->anon_vma;
786 if (!anon_vma && adjust_next)
787 anon_vma = next->anon_vma;
788 if (anon_vma) {
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700789 VM_WARN_ON(adjust_next && next->anon_vma &&
790 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000791 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700792 anon_vma_interval_tree_pre_update_vma(vma);
793 if (adjust_next)
794 anon_vma_interval_tree_pre_update_vma(next);
795 }
Rik van Riel012f18002010-08-09 17:18:40 -0700796
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 if (root) {
798 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700799 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700801 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 }
803
Michel Lespinassed3737182012-12-11 16:01:38 -0800804 if (start != vma->vm_start) {
805 vma->vm_start = start;
806 start_changed = true;
807 }
808 if (end != vma->vm_end) {
809 vma->vm_end = end;
810 end_changed = true;
811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 vma->vm_pgoff = pgoff;
813 if (adjust_next) {
814 next->vm_start += adjust_next << PAGE_SHIFT;
815 next->vm_pgoff += adjust_next;
816 }
817
818 if (root) {
819 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700820 vma_interval_tree_insert(next, root);
821 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 flush_dcache_mmap_unlock(mapping);
823 }
824
825 if (remove_next) {
826 /*
827 * vma_merge has merged next into vma, and needs
828 * us to remove next before dropping the locks.
829 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700830 if (remove_next != 3)
831 __vma_unlink_prev(mm, next, vma);
832 else
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700833 /*
834 * vma is not before next if they've been
835 * swapped.
836 *
837 * pre-swap() next->vm_start was reduced so
838 * tell validate_mm_rb to ignore pre-swap()
839 * "next" (which is stored in post-swap()
840 * "vma").
841 */
842 __vma_unlink_common(mm, next, NULL, false, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 if (file)
844 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 } else if (insert) {
846 /*
847 * split_vma has split insert from vma, and needs
848 * us to insert it before dropping the locks
849 * (it may either follow vma or precede it).
850 */
851 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800852 } else {
853 if (start_changed)
854 vma_gap_update(vma);
855 if (end_changed) {
856 if (!next)
857 mm->highest_vm_end = end;
858 else if (!adjust_next)
859 vma_gap_update(next);
860 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
862
Michel Lespinassebf181b92012-10-08 16:31:39 -0700863 if (anon_vma) {
864 anon_vma_interval_tree_post_update_vma(vma);
865 if (adjust_next)
866 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800867 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800870 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530872 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100873 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530874
875 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100876 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530877 }
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700880 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530881 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700883 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800884 if (next->anon_vma)
885 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700887 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 kmem_cache_free(vm_area_cachep, next);
889 /*
890 * In mprotect's case 6 (see comments on vma_merge),
891 * we must remove another next too. It would clutter
892 * up the code too much to do both in one go.
893 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700894 if (remove_next != 3) {
895 /*
896 * If "next" was removed and vma->vm_end was
897 * expanded (up) over it, in turn
898 * "next->vm_prev->vm_end" changed and the
899 * "vma->vm_next" gap must be updated.
900 */
901 next = vma->vm_next;
902 } else {
903 /*
904 * For the scope of the comment "next" and
905 * "vma" considered pre-swap(): if "vma" was
906 * removed, next->vm_start was expanded (down)
907 * over it and the "next" gap must be updated.
908 * Because of the swap() the post-swap() "vma"
909 * actually points to pre-swap() "next"
910 * (post-swap() "next" as opposed is now a
911 * dangling pointer).
912 */
913 next = vma;
914 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700915 if (remove_next == 2) {
916 remove_next = 1;
917 end = next->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 goto again;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700919 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800920 else if (next)
921 vma_gap_update(next);
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700922 else {
923 /*
924 * If remove_next == 2 we obviously can't
925 * reach this path.
926 *
927 * If remove_next == 3 we can't reach this
928 * path because pre-swap() next is always not
929 * NULL. pre-swap() "next" is not being
930 * removed and its next->vm_end is not altered
931 * (and furthermore "end" already matches
932 * next->vm_end in remove_next == 3).
933 *
934 * We reach this only in the remove_next == 1
935 * case if the "next" vma that was removed was
936 * the highest vma of the mm. However in such
937 * case next->vm_end == "end" and the extended
938 * "vma" has vma->vm_end == next->vm_end so
939 * mm->highest_vm_end doesn't need any update
940 * in remove_next == 1 case.
941 */
942 VM_WARN_ON(mm->highest_vm_end != end);
943 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530945 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100946 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800949
950 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
952
953/*
954 * If the vma has a ->close operation then the driver probably needs to release
955 * per-vma resources, so we don't attempt to merge those.
956 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700958 struct file *file, unsigned long vm_flags,
959 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800961 /*
962 * VM_SOFTDIRTY should not prevent from VMA merging, if we
963 * match the flags but dirty bit -- the caller should mark
964 * merged VMA as dirty. If dirty bit won't be excluded from
965 * comparison, we increase pressue on the memory system forcing
966 * the kernel to generate new VMAs when old one could be
967 * extended instead.
968 */
969 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 return 0;
971 if (vma->vm_file != file)
972 return 0;
973 if (vma->vm_ops && vma->vm_ops->close)
974 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700975 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
976 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 return 1;
978}
979
980static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700981 struct anon_vma *anon_vma2,
982 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983{
Shaohua Li965f55d2011-05-24 17:11:20 -0700984 /*
985 * The list_is_singular() test is to avoid merging VMA cloned from
986 * parents. This can improve scalability caused by anon_vma lock.
987 */
988 if ((!anon_vma1 || !anon_vma2) && (!vma ||
989 list_is_singular(&vma->anon_vma_chain)))
990 return 1;
991 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992}
993
994/*
995 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
996 * in front of (at a lower virtual address and file offset than) the vma.
997 *
998 * We cannot merge two vmas if they have differently assigned (non-NULL)
999 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1000 *
1001 * We don't check here for the merged mmap wrapping around the end of pagecache
1002 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1003 * wrap, nor mmaps which cover the final page at index -1UL.
1004 */
1005static int
1006can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001007 struct anon_vma *anon_vma, struct file *file,
1008 pgoff_t vm_pgoff,
1009 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001011 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001012 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (vma->vm_pgoff == vm_pgoff)
1014 return 1;
1015 }
1016 return 0;
1017}
1018
1019/*
1020 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1021 * beyond (at a higher virtual address and file offset than) the vma.
1022 *
1023 * We cannot merge two vmas if they have differently assigned (non-NULL)
1024 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1025 */
1026static int
1027can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001028 struct anon_vma *anon_vma, struct file *file,
1029 pgoff_t vm_pgoff,
1030 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001032 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001033 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001035 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1037 return 1;
1038 }
1039 return 0;
1040}
1041
1042/*
1043 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1044 * whether that can be merged with its predecessor or its successor.
1045 * Or both (it neatly fills a hole).
1046 *
1047 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1048 * certain not to be mapped by the time vma_merge is called; but when
1049 * called for mprotect, it is certain to be already mapped (either at
1050 * an offset within prev, or at the start of next), and the flags of
1051 * this area are about to be changed to vm_flags - and the no-change
1052 * case has already been eliminated.
1053 *
1054 * The following mprotect cases have to be considered, where AAAA is
1055 * the area passed down from mprotect_fixup, never extending beyond one
1056 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1057 *
1058 * AAAA AAAA AAAA AAAA
1059 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1060 * cannot merge might become might become might become
1061 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1062 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001063 * mremap move: PPPPXXXXXXXX 8
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 * AAAA
1065 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1066 * might become case 1 below case 2 below case 3 below
1067 *
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001068 * It is important for case 8 that the the vma NNNN overlapping the
1069 * region AAAA is never going to extended over XXXX. Instead XXXX must
1070 * be extended in region AAAA and NNNN must be removed. This way in
1071 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1072 * rmap_locks, the properties of the merged vma will be already
1073 * correct for the whole merged range. Some of those properties like
1074 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1075 * be correct for the whole merged range immediately after the
1076 * rmap_locks are released. Otherwise if XXXX would be removed and
1077 * NNNN would be extended over the XXXX range, remove_migration_ptes
1078 * or other rmap walkers (if working on addresses beyond the "end"
1079 * parameter) may establish ptes with the wrong permissions of NNNN
1080 * instead of the right permissions of XXXX.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 */
1082struct vm_area_struct *vma_merge(struct mm_struct *mm,
1083 struct vm_area_struct *prev, unsigned long addr,
1084 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001085 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001086 pgoff_t pgoff, struct mempolicy *policy,
1087 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088{
1089 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1090 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001091 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093 /*
1094 * We later require that vma->vm_flags == vm_flags,
1095 * so this tests vma->vm_flags & VM_SPECIAL, too.
1096 */
1097 if (vm_flags & VM_SPECIAL)
1098 return NULL;
1099
1100 if (prev)
1101 next = prev->vm_next;
1102 else
1103 next = mm->mmap;
1104 area = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001105 if (area && area->vm_end == end) /* cases 6, 7, 8 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 next = next->vm_next;
1107
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001108 /* verify some invariant that must be enforced by the caller */
1109 VM_WARN_ON(prev && addr <= prev->vm_start);
1110 VM_WARN_ON(area && end > area->vm_end);
1111 VM_WARN_ON(addr >= end);
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 /*
1114 * Can it merge with the predecessor?
1115 */
1116 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001117 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001119 anon_vma, file, pgoff,
1120 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 /*
1122 * OK, it can. Can we now merge in the successor as well?
1123 */
1124 if (next && end == next->vm_start &&
1125 mpol_equal(policy, vma_policy(next)) &&
1126 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001127 anon_vma, file,
1128 pgoff+pglen,
1129 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001131 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 /* cases 1, 6 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001133 err = __vma_adjust(prev, prev->vm_start,
1134 next->vm_end, prev->vm_pgoff, NULL,
1135 prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 } else /* cases 2, 5, 7 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001137 err = __vma_adjust(prev, prev->vm_start,
1138 end, prev->vm_pgoff, NULL, prev);
Rik van Riel5beb4932010-03-05 13:42:07 -08001139 if (err)
1140 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001141 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 return prev;
1143 }
1144
1145 /*
1146 * Can this new request be merged in front of next?
1147 */
1148 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001149 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001151 anon_vma, file, pgoff+pglen,
1152 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 if (prev && addr < prev->vm_end) /* case 4 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001154 err = __vma_adjust(prev, prev->vm_start,
1155 addr, prev->vm_pgoff, NULL, next);
1156 else { /* cases 3, 8 */
1157 err = __vma_adjust(area, addr, next->vm_end,
1158 next->vm_pgoff - pglen, NULL, next);
1159 /*
1160 * In case 3 area is already equal to next and
1161 * this is a noop, but in case 8 "area" has
1162 * been removed and next was expanded over it.
1163 */
1164 area = next;
1165 }
Rik van Riel5beb4932010-03-05 13:42:07 -08001166 if (err)
1167 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001168 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 return area;
1170 }
1171
1172 return NULL;
1173}
1174
1175/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001176 * Rough compatbility check to quickly see if it's even worth looking
1177 * at sharing an anon_vma.
1178 *
1179 * They need to have the same vm_file, and the flags can only differ
1180 * in things that mprotect may change.
1181 *
1182 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1183 * we can merge the two vma's. For example, we refuse to merge a vma if
1184 * there is a vm_ops->close() function, because that indicates that the
1185 * driver is doing some kind of reference counting. But that doesn't
1186 * really matter for the anon_vma sharing case.
1187 */
1188static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1189{
1190 return a->vm_end == b->vm_start &&
1191 mpol_equal(vma_policy(a), vma_policy(b)) &&
1192 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001193 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001194 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1195}
1196
1197/*
1198 * Do some basic sanity checking to see if we can re-use the anon_vma
1199 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1200 * the same as 'old', the other will be the new one that is trying
1201 * to share the anon_vma.
1202 *
1203 * NOTE! This runs with mm_sem held for reading, so it is possible that
1204 * the anon_vma of 'old' is concurrently in the process of being set up
1205 * by another page fault trying to merge _that_. But that's ok: if it
1206 * is being set up, that automatically means that it will be a singleton
1207 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001208 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001209 *
1210 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1211 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1212 * is to return an anon_vma that is "complex" due to having gone through
1213 * a fork).
1214 *
1215 * We also make sure that the two vma's are compatible (adjacent,
1216 * and with the same memory policies). That's all stable, even with just
1217 * a read lock on the mm_sem.
1218 */
1219static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1220{
1221 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001222 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001223
1224 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1225 return anon_vma;
1226 }
1227 return NULL;
1228}
1229
1230/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1232 * neighbouring vmas for a suitable anon_vma, before it goes off
1233 * to allocate a new anon_vma. It checks because a repetitive
1234 * sequence of mprotects and faults may otherwise lead to distinct
1235 * anon_vmas being allocated, preventing vma merge in subsequent
1236 * mprotect.
1237 */
1238struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1239{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001240 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243 near = vma->vm_next;
1244 if (!near)
1245 goto try_prev;
1246
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001247 anon_vma = reusable_anon_vma(near, vma, near);
1248 if (anon_vma)
1249 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001251 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 if (!near)
1253 goto none;
1254
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001255 anon_vma = reusable_anon_vma(near, near, vma);
1256 if (anon_vma)
1257 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258none:
1259 /*
1260 * There's no absolute need to look only at touching neighbours:
1261 * we could search further afield for "compatible" anon_vmas.
1262 * But it would probably just be a waste of time searching,
1263 * or lead to too many vmas hanging off the same anon_vma.
1264 * We're trying to allow mprotect remerging later on,
1265 * not trying to minimize memory used for anon_vmas.
1266 */
1267 return NULL;
1268}
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270/*
Al Viro40401532012-02-13 03:58:52 +00001271 * If a hint addr is less than mmap_min_addr change hint to be as
1272 * low as possible but still greater than mmap_min_addr
1273 */
1274static inline unsigned long round_hint_to_min(unsigned long hint)
1275{
1276 hint &= PAGE_MASK;
1277 if (((void *)hint != NULL) &&
1278 (hint < mmap_min_addr))
1279 return PAGE_ALIGN(mmap_min_addr);
1280 return hint;
1281}
1282
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001283static inline int mlock_future_check(struct mm_struct *mm,
1284 unsigned long flags,
1285 unsigned long len)
1286{
1287 unsigned long locked, lock_limit;
1288
1289 /* mlock MCL_FUTURE? */
1290 if (flags & VM_LOCKED) {
1291 locked = len >> PAGE_SHIFT;
1292 locked += mm->locked_vm;
1293 lock_limit = rlimit(RLIMIT_MEMLOCK);
1294 lock_limit >>= PAGE_SHIFT;
1295 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1296 return -EAGAIN;
1297 }
1298 return 0;
1299}
1300
Al Viro40401532012-02-13 03:58:52 +00001301/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001302 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001304unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001306 unsigned long flags, vm_flags_t vm_flags,
1307 unsigned long pgoff, unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
vishnu.pscc71aba2014-10-09 15:26:29 -07001309 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001310 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Michel Lespinasse41badc12013-02-22 16:32:47 -08001312 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001313
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001314 if (!len)
1315 return -EINVAL;
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 /*
1318 * Does the application expect PROT_READ to imply PROT_EXEC?
1319 *
1320 * (the exception is when the underlying filesystem is noexec
1321 * mounted, in which case we dont add PROT_EXEC.)
1322 */
1323 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001324 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 prot |= PROT_EXEC;
1326
Eric Paris7cd94142007-11-26 18:47:40 -05001327 if (!(flags & MAP_FIXED))
1328 addr = round_hint_to_min(addr);
1329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 /* Careful about overflows.. */
1331 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001332 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 return -ENOMEM;
1334
1335 /* offset overflow? */
1336 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001337 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339 /* Too many mappings? */
1340 if (mm->map_count > sysctl_max_map_count)
1341 return -ENOMEM;
1342
1343 /* Obtain the address to map to. we verify (or select) it and ensure
1344 * that it represents a valid section of the address space.
1345 */
1346 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001347 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 return addr;
1349
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001350 if (prot == PROT_EXEC) {
1351 pkey = execute_only_pkey(mm);
1352 if (pkey < 0)
1353 pkey = 0;
1354 }
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 /* Do simple checking here so the lower-level routines won't have
1357 * to. we assume access permissions have been handled by the open
1358 * of the memory object, so we don't do any here.
1359 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001360 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1362
Huang Shijiecdf7b342009-09-21 17:03:36 -07001363 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 if (!can_do_mlock())
1365 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001366
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001367 if (mlock_future_check(mm, vm_flags, len))
1368 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001371 struct inode *inode = file_inode(file);
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 switch (flags & MAP_TYPE) {
1374 case MAP_SHARED:
1375 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1376 return -EACCES;
1377
1378 /*
1379 * Make sure we don't allow writing to an append-only
1380 * file..
1381 */
1382 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1383 return -EACCES;
1384
1385 /*
1386 * Make sure there are no mandatory locks on the file.
1387 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001388 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 return -EAGAIN;
1390
1391 vm_flags |= VM_SHARED | VM_MAYSHARE;
1392 if (!(file->f_mode & FMODE_WRITE))
1393 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1394
1395 /* fall through */
1396 case MAP_PRIVATE:
1397 if (!(file->f_mode & FMODE_READ))
1398 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001399 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001400 if (vm_flags & VM_EXEC)
1401 return -EPERM;
1402 vm_flags &= ~VM_MAYEXEC;
1403 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001404
Al Viro72c2d532013-09-22 16:27:52 -04001405 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001406 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001407 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1408 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 break;
1410
1411 default:
1412 return -EINVAL;
1413 }
1414 } else {
1415 switch (flags & MAP_TYPE) {
1416 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001417 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1418 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001419 /*
1420 * Ignore pgoff.
1421 */
1422 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 vm_flags |= VM_SHARED | VM_MAYSHARE;
1424 break;
1425 case MAP_PRIVATE:
1426 /*
1427 * Set pgoff according to addr for anon_vma.
1428 */
1429 pgoff = addr >> PAGE_SHIFT;
1430 break;
1431 default:
1432 return -EINVAL;
1433 }
1434 }
1435
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001436 /*
1437 * Set 'VM_NORESERVE' if we should not account for the
1438 * memory use of this mapping.
1439 */
1440 if (flags & MAP_NORESERVE) {
1441 /* We honor MAP_NORESERVE if allowed to overcommit */
1442 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1443 vm_flags |= VM_NORESERVE;
1444
1445 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1446 if (file && is_file_hugepages(file))
1447 vm_flags |= VM_NORESERVE;
1448 }
1449
1450 addr = mmap_region(file, addr, len, vm_flags, pgoff);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001451 if (!IS_ERR_VALUE(addr) &&
1452 ((vm_flags & VM_LOCKED) ||
1453 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001454 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001455 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001456}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001457
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001458SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1459 unsigned long, prot, unsigned long, flags,
1460 unsigned long, fd, unsigned long, pgoff)
1461{
1462 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001463 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001464
1465 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001466 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001467 file = fget(fd);
1468 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001469 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001470 if (is_file_hugepages(file))
1471 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001472 retval = -EINVAL;
1473 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1474 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001475 } else if (flags & MAP_HUGETLB) {
1476 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001477 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001478
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001479 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001480 if (!hs)
1481 return -EINVAL;
1482
1483 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001484 /*
1485 * VM_NORESERVE is used because the reservations will be
1486 * taken when vm_ops->mmap() is called
1487 * A dummy user value is used because we are not locking
1488 * memory so no accounting is necessary
1489 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001490 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001491 VM_NORESERVE,
1492 &user, HUGETLB_ANONHUGE_INODE,
1493 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001494 if (IS_ERR(file))
1495 return PTR_ERR(file);
1496 }
1497
1498 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1499
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001500 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001501out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001502 if (file)
1503 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001504 return retval;
1505}
1506
Christoph Hellwiga4679372010-03-10 15:21:15 -08001507#ifdef __ARCH_WANT_SYS_OLD_MMAP
1508struct mmap_arg_struct {
1509 unsigned long addr;
1510 unsigned long len;
1511 unsigned long prot;
1512 unsigned long flags;
1513 unsigned long fd;
1514 unsigned long offset;
1515};
1516
1517SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1518{
1519 struct mmap_arg_struct a;
1520
1521 if (copy_from_user(&a, arg, sizeof(a)))
1522 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001523 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001524 return -EINVAL;
1525
1526 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1527 a.offset >> PAGE_SHIFT);
1528}
1529#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1530
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001531/*
1532 * Some shared mappigns will want the pages marked read-only
1533 * to track write events. If so, we'll downgrade vm_page_prot
1534 * to the private version (using protection_map[] without the
1535 * VM_SHARED bit).
1536 */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001537int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001538{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001539 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001540 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001541
1542 /* If it was private or non-writable, the write bit is already clear */
1543 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1544 return 0;
1545
1546 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001547 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001548 return 1;
1549
Peter Feiner64e45502014-10-13 15:55:46 -07001550 /* The open routine did something to the protections that pgprot_modify
1551 * won't preserve? */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001552 if (pgprot_val(vm_page_prot) !=
1553 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001554 return 0;
1555
Peter Feiner64e45502014-10-13 15:55:46 -07001556 /* Do we need to track softdirty? */
1557 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1558 return 1;
1559
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001560 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001561 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001562 return 0;
1563
1564 /* Can the mapping track the dirty pages? */
1565 return vma->vm_file && vma->vm_file->f_mapping &&
1566 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1567}
1568
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001569/*
1570 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001571 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001572 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001573static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001574{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001575 /*
1576 * hugetlb has its own accounting separate from the core VM
1577 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1578 */
1579 if (file && is_file_hugepages(file))
1580 return 0;
1581
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001582 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1583}
1584
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001585unsigned long mmap_region(struct file *file, unsigned long addr,
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001586 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001587{
1588 struct mm_struct *mm = current->mm;
1589 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001590 int error;
1591 struct rb_node **rb_link, *rb_parent;
1592 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001593
Cyril Hrubise8420a82013-04-29 15:08:33 -07001594 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001595 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001596 unsigned long nr_pages;
1597
1598 /*
1599 * MAP_FIXED may remove pages of mappings that intersects with
1600 * requested mapping. Account for the pages it would unmap.
1601 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001602 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1603
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001604 if (!may_expand_vm(mm, vm_flags,
1605 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001606 return -ENOMEM;
1607 }
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001610 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1611 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (do_munmap(mm, addr, len))
1613 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 }
1615
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001616 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001617 * Private writable mapping: check memory availability
1618 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001619 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001620 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001621 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001622 return -ENOMEM;
1623 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 }
1625
1626 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001627 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001629 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1630 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001631 if (vma)
1632 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
1634 /*
1635 * Determine the object being mapped and call the appropriate
1636 * specific mapper. the address has already been validated, but
1637 * not unmapped, but the maps are removed from the list.
1638 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001639 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 if (!vma) {
1641 error = -ENOMEM;
1642 goto unacct_error;
1643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
1645 vma->vm_mm = mm;
1646 vma->vm_start = addr;
1647 vma->vm_end = addr + len;
1648 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001649 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001651 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
1653 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (vm_flags & VM_DENYWRITE) {
1655 error = deny_write_access(file);
1656 if (error)
1657 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001659 if (vm_flags & VM_SHARED) {
1660 error = mapping_map_writable(file->f_mapping);
1661 if (error)
1662 goto allow_write_and_free_vma;
1663 }
1664
1665 /* ->mmap() can change vma->vm_file, but must guarantee that
1666 * vma_link() below can deny write-access if VM_DENYWRITE is set
1667 * and map writably if VM_SHARED is set. This usually means the
1668 * new file must not have been exposed to user-space, yet.
1669 */
Al Virocb0942b2012-08-27 14:48:26 -04001670 vma->vm_file = get_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 error = file->f_op->mmap(file, vma);
1672 if (error)
1673 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001674
1675 /* Can addr have changed??
1676 *
1677 * Answer: Yes, several device drivers can do it in their
1678 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001679 * Bug: If addr is changed, prev, rb_link, rb_parent should
1680 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001681 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001682 WARN_ON_ONCE(addr != vma->vm_start);
1683
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001684 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001685 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 } else if (vm_flags & VM_SHARED) {
1687 error = shmem_zero_setup(vma);
1688 if (error)
1689 goto free_vma;
1690 }
1691
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001692 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001693 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001694 if (file) {
1695 if (vm_flags & VM_SHARED)
1696 mapping_unmap_writable(file->f_mapping);
1697 if (vm_flags & VM_DENYWRITE)
1698 allow_write_access(file);
1699 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001700 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001701out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001702 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001703
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001704 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001706 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1707 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001708 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001709 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001710 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001711 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301712
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001713 if (file)
1714 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301715
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001716 /*
1717 * New (or expanded) vma always get soft dirty status.
1718 * Otherwise user-space soft-dirty page tracker won't
1719 * be able to distinguish situation when vma area unmapped,
1720 * then new mapped in-place (which must be aimed as
1721 * a completely new data area).
1722 */
1723 vma->vm_flags |= VM_SOFTDIRTY;
1724
Peter Feiner64e45502014-10-13 15:55:46 -07001725 vma_set_page_prot(vma);
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 return addr;
1728
1729unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 vma->vm_file = NULL;
1731 fput(file);
1732
1733 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001734 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1735 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001736 if (vm_flags & VM_SHARED)
1737 mapping_unmap_writable(file->f_mapping);
1738allow_write_and_free_vma:
1739 if (vm_flags & VM_DENYWRITE)
1740 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741free_vma:
1742 kmem_cache_free(vm_area_cachep, vma);
1743unacct_error:
1744 if (charged)
1745 vm_unacct_memory(charged);
1746 return error;
1747}
1748
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001749unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1750{
1751 /*
1752 * We implement the search by looking for an rbtree node that
1753 * immediately follows a suitable gap. That is,
1754 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1755 * - gap_end = vma->vm_start >= info->low_limit + length;
1756 * - gap_end - gap_start >= length
1757 */
1758
1759 struct mm_struct *mm = current->mm;
1760 struct vm_area_struct *vma;
1761 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1762
1763 /* Adjust search length to account for worst case alignment overhead */
1764 length = info->length + info->align_mask;
1765 if (length < info->length)
1766 return -ENOMEM;
1767
1768 /* Adjust search limits by the desired length */
1769 if (info->high_limit < length)
1770 return -ENOMEM;
1771 high_limit = info->high_limit - length;
1772
1773 if (info->low_limit > high_limit)
1774 return -ENOMEM;
1775 low_limit = info->low_limit + length;
1776
1777 /* Check if rbtree root looks promising */
1778 if (RB_EMPTY_ROOT(&mm->mm_rb))
1779 goto check_highest;
1780 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1781 if (vma->rb_subtree_gap < length)
1782 goto check_highest;
1783
1784 while (true) {
1785 /* Visit left subtree if it looks promising */
1786 gap_end = vma->vm_start;
1787 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1788 struct vm_area_struct *left =
1789 rb_entry(vma->vm_rb.rb_left,
1790 struct vm_area_struct, vm_rb);
1791 if (left->rb_subtree_gap >= length) {
1792 vma = left;
1793 continue;
1794 }
1795 }
1796
1797 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1798check_current:
1799 /* Check if current node has a suitable gap */
1800 if (gap_start > high_limit)
1801 return -ENOMEM;
1802 if (gap_end >= low_limit && gap_end - gap_start >= length)
1803 goto found;
1804
1805 /* Visit right subtree if it looks promising */
1806 if (vma->vm_rb.rb_right) {
1807 struct vm_area_struct *right =
1808 rb_entry(vma->vm_rb.rb_right,
1809 struct vm_area_struct, vm_rb);
1810 if (right->rb_subtree_gap >= length) {
1811 vma = right;
1812 continue;
1813 }
1814 }
1815
1816 /* Go back up the rbtree to find next candidate node */
1817 while (true) {
1818 struct rb_node *prev = &vma->vm_rb;
1819 if (!rb_parent(prev))
1820 goto check_highest;
1821 vma = rb_entry(rb_parent(prev),
1822 struct vm_area_struct, vm_rb);
1823 if (prev == vma->vm_rb.rb_left) {
1824 gap_start = vma->vm_prev->vm_end;
1825 gap_end = vma->vm_start;
1826 goto check_current;
1827 }
1828 }
1829 }
1830
1831check_highest:
1832 /* Check highest gap, which does not precede any rbtree node */
1833 gap_start = mm->highest_vm_end;
1834 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1835 if (gap_start > high_limit)
1836 return -ENOMEM;
1837
1838found:
1839 /* We found a suitable gap. Clip it with the original low_limit. */
1840 if (gap_start < info->low_limit)
1841 gap_start = info->low_limit;
1842
1843 /* Adjust gap address to the desired alignment */
1844 gap_start += (info->align_offset - gap_start) & info->align_mask;
1845
1846 VM_BUG_ON(gap_start + info->length > info->high_limit);
1847 VM_BUG_ON(gap_start + info->length > gap_end);
1848 return gap_start;
1849}
1850
1851unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1852{
1853 struct mm_struct *mm = current->mm;
1854 struct vm_area_struct *vma;
1855 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1856
1857 /* Adjust search length to account for worst case alignment overhead */
1858 length = info->length + info->align_mask;
1859 if (length < info->length)
1860 return -ENOMEM;
1861
1862 /*
1863 * Adjust search limits by the desired length.
1864 * See implementation comment at top of unmapped_area().
1865 */
1866 gap_end = info->high_limit;
1867 if (gap_end < length)
1868 return -ENOMEM;
1869 high_limit = gap_end - length;
1870
1871 if (info->low_limit > high_limit)
1872 return -ENOMEM;
1873 low_limit = info->low_limit + length;
1874
1875 /* Check highest gap, which does not precede any rbtree node */
1876 gap_start = mm->highest_vm_end;
1877 if (gap_start <= high_limit)
1878 goto found_highest;
1879
1880 /* Check if rbtree root looks promising */
1881 if (RB_EMPTY_ROOT(&mm->mm_rb))
1882 return -ENOMEM;
1883 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1884 if (vma->rb_subtree_gap < length)
1885 return -ENOMEM;
1886
1887 while (true) {
1888 /* Visit right subtree if it looks promising */
1889 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1890 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1891 struct vm_area_struct *right =
1892 rb_entry(vma->vm_rb.rb_right,
1893 struct vm_area_struct, vm_rb);
1894 if (right->rb_subtree_gap >= length) {
1895 vma = right;
1896 continue;
1897 }
1898 }
1899
1900check_current:
1901 /* Check if current node has a suitable gap */
1902 gap_end = vma->vm_start;
1903 if (gap_end < low_limit)
1904 return -ENOMEM;
1905 if (gap_start <= high_limit && gap_end - gap_start >= length)
1906 goto found;
1907
1908 /* Visit left subtree if it looks promising */
1909 if (vma->vm_rb.rb_left) {
1910 struct vm_area_struct *left =
1911 rb_entry(vma->vm_rb.rb_left,
1912 struct vm_area_struct, vm_rb);
1913 if (left->rb_subtree_gap >= length) {
1914 vma = left;
1915 continue;
1916 }
1917 }
1918
1919 /* Go back up the rbtree to find next candidate node */
1920 while (true) {
1921 struct rb_node *prev = &vma->vm_rb;
1922 if (!rb_parent(prev))
1923 return -ENOMEM;
1924 vma = rb_entry(rb_parent(prev),
1925 struct vm_area_struct, vm_rb);
1926 if (prev == vma->vm_rb.rb_right) {
1927 gap_start = vma->vm_prev ?
1928 vma->vm_prev->vm_end : 0;
1929 goto check_current;
1930 }
1931 }
1932 }
1933
1934found:
1935 /* We found a suitable gap. Clip it with the original high_limit. */
1936 if (gap_end > info->high_limit)
1937 gap_end = info->high_limit;
1938
1939found_highest:
1940 /* Compute highest gap address at the desired alignment */
1941 gap_end -= info->length;
1942 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1943
1944 VM_BUG_ON(gap_end < info->low_limit);
1945 VM_BUG_ON(gap_end < gap_start);
1946 return gap_end;
1947}
1948
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949/* Get an address range which is currently unmapped.
1950 * For shmat() with addr=0.
1951 *
1952 * Ugly calling convention alert:
1953 * Return value with the low bits set means error value,
1954 * ie
1955 * if (ret & ~PAGE_MASK)
1956 * error = ret;
1957 *
1958 * This function "knows" that -ENOMEM has the bits set.
1959 */
1960#ifndef HAVE_ARCH_UNMAPPED_AREA
1961unsigned long
1962arch_get_unmapped_area(struct file *filp, unsigned long addr,
1963 unsigned long len, unsigned long pgoff, unsigned long flags)
1964{
1965 struct mm_struct *mm = current->mm;
1966 struct vm_area_struct *vma;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001967 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001969 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 return -ENOMEM;
1971
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07001972 if (flags & MAP_FIXED)
1973 return addr;
1974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 if (addr) {
1976 addr = PAGE_ALIGN(addr);
1977 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08001978 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 (!vma || addr + len <= vma->vm_start))
1980 return addr;
1981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001983 info.flags = 0;
1984 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08001985 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001986 info.high_limit = TASK_SIZE;
1987 info.align_mask = 0;
1988 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989}
vishnu.pscc71aba2014-10-09 15:26:29 -07001990#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992/*
1993 * This mmap-allocator allocates new areas top-down from below the
1994 * stack's low limit (the base):
1995 */
1996#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1997unsigned long
1998arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1999 const unsigned long len, const unsigned long pgoff,
2000 const unsigned long flags)
2001{
2002 struct vm_area_struct *vma;
2003 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002004 unsigned long addr = addr0;
2005 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
2007 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002008 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 return -ENOMEM;
2010
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002011 if (flags & MAP_FIXED)
2012 return addr;
2013
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 /* requesting a specific address */
2015 if (addr) {
2016 addr = PAGE_ALIGN(addr);
2017 vma = find_vma(mm, addr);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002018 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 (!vma || addr + len <= vma->vm_start))
2020 return addr;
2021 }
2022
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002023 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2024 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002025 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002026 info.high_limit = mm->mmap_base;
2027 info.align_mask = 0;
2028 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07002029
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 /*
2031 * A failed mmap() very likely causes application failure,
2032 * so fall back to the bottom-up function here. This scenario
2033 * can happen with large stack limits and large mmap()
2034 * allocations.
2035 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002036 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002037 VM_BUG_ON(addr != -ENOMEM);
2038 info.flags = 0;
2039 info.low_limit = TASK_UNMAPPED_BASE;
2040 info.high_limit = TASK_SIZE;
2041 addr = vm_unmapped_area(&info);
2042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044 return addr;
2045}
2046#endif
2047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048unsigned long
2049get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2050 unsigned long pgoff, unsigned long flags)
2051{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002052 unsigned long (*get_area)(struct file *, unsigned long,
2053 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Al Viro9206de92009-12-03 15:23:11 -05002055 unsigned long error = arch_mmap_check(addr, len, flags);
2056 if (error)
2057 return error;
2058
2059 /* Careful about overflows.. */
2060 if (len > TASK_SIZE)
2061 return -ENOMEM;
2062
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002063 get_area = current->mm->get_unmapped_area;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002064 if (file) {
2065 if (file->f_op->get_unmapped_area)
2066 get_area = file->f_op->get_unmapped_area;
2067 } else if (flags & MAP_SHARED) {
2068 /*
2069 * mmap_region() will call shmem_zero_setup() to create a file,
2070 * so use shmem's get_unmapped_area in case it can be huge.
2071 * do_mmap_pgoff() will clear pgoff, so match alignment.
2072 */
2073 pgoff = 0;
2074 get_area = shmem_get_unmapped_area;
2075 }
2076
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002077 addr = get_area(file, addr, len, pgoff, flags);
2078 if (IS_ERR_VALUE(addr))
2079 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002081 if (addr > TASK_SIZE - len)
2082 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002083 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002084 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002085
Al Viro9ac4ed42012-05-30 17:13:15 -04002086 error = security_mmap_addr(addr);
2087 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088}
2089
2090EXPORT_SYMBOL(get_unmapped_area);
2091
2092/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002093struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002095 struct rb_node *rb_node;
2096 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002098 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002099 vma = vmacache_find(mm, addr);
2100 if (likely(vma))
2101 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002103 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002105 while (rb_node) {
2106 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002108 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002109
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002110 if (tmp->vm_end > addr) {
2111 vma = tmp;
2112 if (tmp->vm_start <= addr)
2113 break;
2114 rb_node = rb_node->rb_left;
2115 } else
2116 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002118
2119 if (vma)
2120 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 return vma;
2122}
2123
2124EXPORT_SYMBOL(find_vma);
2125
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002126/*
2127 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002128 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129struct vm_area_struct *
2130find_vma_prev(struct mm_struct *mm, unsigned long addr,
2131 struct vm_area_struct **pprev)
2132{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002133 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002135 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002136 if (vma) {
2137 *pprev = vma->vm_prev;
2138 } else {
2139 struct rb_node *rb_node = mm->mm_rb.rb_node;
2140 *pprev = NULL;
2141 while (rb_node) {
2142 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2143 rb_node = rb_node->rb_right;
2144 }
2145 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002146 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147}
2148
2149/*
2150 * Verify that the stack growth is acceptable and
2151 * update accounting. This is shared with both the
2152 * grow-up and grow-down cases.
2153 */
ZhenwenXu48aae422009-01-06 14:40:21 -08002154static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155{
2156 struct mm_struct *mm = vma->vm_mm;
2157 struct rlimit *rlim = current->signal->rlim;
Linus Torvalds690eac52015-01-11 11:33:57 -08002158 unsigned long new_start, actual_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
2160 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002161 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 return -ENOMEM;
2163
2164 /* Stack limit test */
Linus Torvalds690eac52015-01-11 11:33:57 -08002165 actual_size = size;
2166 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2167 actual_size -= PAGE_SIZE;
Jason Low4db0c3c2015-04-15 16:14:08 -07002168 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 return -ENOMEM;
2170
2171 /* mlock limit tests */
2172 if (vma->vm_flags & VM_LOCKED) {
2173 unsigned long locked;
2174 unsigned long limit;
2175 locked = mm->locked_vm + grow;
Jason Low4db0c3c2015-04-15 16:14:08 -07002176 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002177 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 if (locked > limit && !capable(CAP_IPC_LOCK))
2179 return -ENOMEM;
2180 }
2181
Adam Litke0d59a012007-01-30 14:35:39 -08002182 /* Check to ensure the stack will not grow into a hugetlb-only region */
2183 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2184 vma->vm_end - size;
2185 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2186 return -EFAULT;
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 /*
2189 * Overcommit.. This must be the final test, as it will
2190 * update security statistics.
2191 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002192 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 return -ENOMEM;
2194
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 return 0;
2196}
2197
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002198#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002200 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2201 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002203int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204{
Oleg Nesterov09357812015-11-05 18:48:17 -08002205 struct mm_struct *mm = vma->vm_mm;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002206 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
2208 if (!(vma->vm_flags & VM_GROWSUP))
2209 return -EFAULT;
2210
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002211 /* Guard against wrapping around to address 0. */
2212 if (address < PAGE_ALIGN(address+4))
2213 address = PAGE_ALIGN(address+4);
2214 else
2215 return -ENOMEM;
2216
2217 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 if (unlikely(anon_vma_prepare(vma)))
2219 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
2221 /*
2222 * vma->vm_start/vm_end cannot change under us because the caller
2223 * is required to hold the mmap_sem in read mode. We need the
2224 * anon_vma lock to serialize against concurrent expand_stacks.
2225 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002226 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
2228 /* Somebody else might have raced and expanded it already */
2229 if (address > vma->vm_end) {
2230 unsigned long size, grow;
2231
2232 size = address - vma->vm_start;
2233 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2234
Hugh Dickins42c36f62011-05-09 17:44:42 -07002235 error = -ENOMEM;
2236 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2237 error = acct_stack_growth(vma, size, grow);
2238 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002239 /*
2240 * vma_gap_update() doesn't support concurrent
2241 * updates, but we only hold a shared mmap_sem
2242 * lock here, so we need to protect against
2243 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002244 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002245 * we don't guarantee that all growable vmas
2246 * in a mm share the same root anon vma.
2247 * So, we reuse mm->page_table_lock to guard
2248 * against concurrent vma expansions.
2249 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002250 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002251 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002252 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002253 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002254 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002255 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002256 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002257 if (vma->vm_next)
2258 vma_gap_update(vma->vm_next);
2259 else
Oleg Nesterov09357812015-11-05 18:48:17 -08002260 mm->highest_vm_end = address;
2261 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002262
Hugh Dickins42c36f62011-05-09 17:44:42 -07002263 perf_event_mmap(vma);
2264 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002267 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002268 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002269 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 return error;
2271}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002272#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274/*
2275 * vma is the first one with address < vma->vm_start. Have to extend vma.
2276 */
Michal Hockod05f3162011-05-24 17:11:44 -07002277int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002278 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279{
Oleg Nesterov09357812015-11-05 18:48:17 -08002280 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 int error;
2282
Eric Paris88694772007-11-26 18:47:26 -05002283 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002284 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002285 if (error)
2286 return error;
2287
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002288 /* We must make sure the anon_vma is allocated. */
2289 if (unlikely(anon_vma_prepare(vma)))
2290 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
2292 /*
2293 * vma->vm_start/vm_end cannot change under us because the caller
2294 * is required to hold the mmap_sem in read mode. We need the
2295 * anon_vma lock to serialize against concurrent expand_stacks.
2296 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002297 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 /* Somebody else might have raced and expanded it already */
2300 if (address < vma->vm_start) {
2301 unsigned long size, grow;
2302
2303 size = vma->vm_end - address;
2304 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2305
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002306 error = -ENOMEM;
2307 if (grow <= vma->vm_pgoff) {
2308 error = acct_stack_growth(vma, size, grow);
2309 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002310 /*
2311 * vma_gap_update() doesn't support concurrent
2312 * updates, but we only hold a shared mmap_sem
2313 * lock here, so we need to protect against
2314 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002315 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002316 * we don't guarantee that all growable vmas
2317 * in a mm share the same root anon vma.
2318 * So, we reuse mm->page_table_lock to guard
2319 * against concurrent vma expansions.
2320 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002321 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002322 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002323 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002324 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002325 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002326 vma->vm_start = address;
2327 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002328 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002329 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002330 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002331
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002332 perf_event_mmap(vma);
2333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 }
2335 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002336 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002337 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002338 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 return error;
2340}
2341
Linus Torvalds09884962013-02-27 08:36:04 -08002342/*
2343 * Note how expand_stack() refuses to expand the stack all the way to
2344 * abut the next virtual mapping, *unless* that mapping itself is also
2345 * a stack mapping. We want to leave room for a guard page, after all
2346 * (the guard page itself is not added here, that is done by the
2347 * actual page faulting logic)
2348 *
2349 * This matches the behavior of the guard page logic (see mm/memory.c:
2350 * check_stack_guard_page()), which only allows the guard page to be
2351 * removed under these circumstances.
2352 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002353#ifdef CONFIG_STACK_GROWSUP
2354int expand_stack(struct vm_area_struct *vma, unsigned long address)
2355{
Linus Torvalds09884962013-02-27 08:36:04 -08002356 struct vm_area_struct *next;
2357
2358 address &= PAGE_MASK;
2359 next = vma->vm_next;
2360 if (next && next->vm_start == address + PAGE_SIZE) {
2361 if (!(next->vm_flags & VM_GROWSUP))
2362 return -ENOMEM;
2363 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002364 return expand_upwards(vma, address);
2365}
2366
2367struct vm_area_struct *
2368find_extend_vma(struct mm_struct *mm, unsigned long addr)
2369{
2370 struct vm_area_struct *vma, *prev;
2371
2372 addr &= PAGE_MASK;
2373 vma = find_vma_prev(mm, addr, &prev);
2374 if (vma && (vma->vm_start <= addr))
2375 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002376 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002377 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002378 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002379 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002380 return prev;
2381}
2382#else
2383int expand_stack(struct vm_area_struct *vma, unsigned long address)
2384{
Linus Torvalds09884962013-02-27 08:36:04 -08002385 struct vm_area_struct *prev;
2386
2387 address &= PAGE_MASK;
2388 prev = vma->vm_prev;
2389 if (prev && prev->vm_end == address) {
2390 if (!(prev->vm_flags & VM_GROWSDOWN))
2391 return -ENOMEM;
2392 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002393 return expand_downwards(vma, address);
2394}
2395
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002397find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398{
vishnu.pscc71aba2014-10-09 15:26:29 -07002399 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 unsigned long start;
2401
2402 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002403 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 if (!vma)
2405 return NULL;
2406 if (vma->vm_start <= addr)
2407 return vma;
2408 if (!(vma->vm_flags & VM_GROWSDOWN))
2409 return NULL;
2410 start = vma->vm_start;
2411 if (expand_stack(vma, addr))
2412 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002413 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002414 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 return vma;
2416}
2417#endif
2418
Jesse Barnese1d6d012014-12-12 16:55:27 -08002419EXPORT_SYMBOL_GPL(find_extend_vma);
2420
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002422 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002424 *
2425 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002427static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002429 unsigned long nr_accounted = 0;
2430
Hugh Dickins365e9c872005-10-29 18:16:18 -07002431 /* Update high watermark before we lower total_vm */
2432 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002434 long nrpages = vma_pages(vma);
2435
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002436 if (vma->vm_flags & VM_ACCOUNT)
2437 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002438 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002439 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002440 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002441 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 validate_mm(mm);
2443}
2444
2445/*
2446 * Get rid of page table information in the indicated region.
2447 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002448 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 */
2450static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002451 struct vm_area_struct *vma, struct vm_area_struct *prev,
2452 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453{
vishnu.pscc71aba2014-10-09 15:26:29 -07002454 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002455 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456
2457 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002458 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002459 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002460 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002461 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002462 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002463 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464}
2465
2466/*
2467 * Create a list of vma's touched by the unmap, removing them from the mm's
2468 * vma list as we go..
2469 */
2470static void
2471detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2472 struct vm_area_struct *prev, unsigned long end)
2473{
2474 struct vm_area_struct **insertion_point;
2475 struct vm_area_struct *tail_vma = NULL;
2476
2477 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002478 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002480 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 mm->map_count--;
2482 tail_vma = vma;
2483 vma = vma->vm_next;
2484 } while (vma && vma->vm_start < end);
2485 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002486 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002487 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002488 vma_gap_update(vma);
2489 } else
2490 mm->highest_vm_end = prev ? prev->vm_end : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002492
2493 /* Kill the cache */
2494 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495}
2496
2497/*
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002498 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2499 * munmap path where it doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 */
vishnu.pscc71aba2014-10-09 15:26:29 -07002501static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 unsigned long addr, int new_below)
2503{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002505 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506
Andi Kleena5516432008-07-23 21:27:41 -07002507 if (is_vm_hugetlb_page(vma) && (addr &
2508 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 return -EINVAL;
2510
Christoph Lametere94b1762006-12-06 20:33:17 -08002511 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002513 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
2515 /* most fields are the same, copy all, and then fixup */
2516 *new = *vma;
2517
Rik van Riel5beb4932010-03-05 13:42:07 -08002518 INIT_LIST_HEAD(&new->anon_vma_chain);
2519
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 if (new_below)
2521 new->vm_end = addr;
2522 else {
2523 new->vm_start = addr;
2524 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2525 }
2526
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002527 err = vma_dup_policy(vma, new);
2528 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002529 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002531 err = anon_vma_clone(new, vma);
2532 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002533 goto out_free_mpol;
2534
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002535 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 get_file(new->vm_file);
2537
2538 if (new->vm_ops && new->vm_ops->open)
2539 new->vm_ops->open(new);
2540
2541 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002542 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2544 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002545 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Rik van Riel5beb4932010-03-05 13:42:07 -08002547 /* Success. */
2548 if (!err)
2549 return 0;
2550
2551 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002552 if (new->vm_ops && new->vm_ops->close)
2553 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002554 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002555 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002556 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002557 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002558 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002559 out_free_vma:
2560 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002561 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562}
2563
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002564/*
2565 * Split a vma into two pieces at address 'addr', a new vma is allocated
2566 * either for the first part or the tail.
2567 */
2568int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2569 unsigned long addr, int new_below)
2570{
2571 if (mm->map_count >= sysctl_max_map_count)
2572 return -ENOMEM;
2573
2574 return __split_vma(mm, vma, addr, new_below);
2575}
2576
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577/* Munmap is split into 2 main parts -- this part which finds
2578 * what needs doing, and the areas themselves, which do the
2579 * work. This now handles partial unmappings.
2580 * Jeremy Fitzhardinge <jeremy@goop.org>
2581 */
2582int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2583{
2584 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002585 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002587 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 return -EINVAL;
2589
vishnu.pscc71aba2014-10-09 15:26:29 -07002590 len = PAGE_ALIGN(len);
2591 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 return -EINVAL;
2593
2594 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002595 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002596 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002598 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002599 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600
2601 /* if it doesn't overlap, we have nothing.. */
2602 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002603 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 return 0;
2605
2606 /*
2607 * If we need to split any vma, do it now to save pain later.
2608 *
2609 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2610 * unmapped vm_area_struct will remain in use: so lower split_vma
2611 * places tmp vma above, and higher split_vma places tmp vma below.
2612 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002613 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002614 int error;
2615
2616 /*
2617 * Make sure that map_count on return from munmap() will
2618 * not exceed its limit; but let map_count go just above
2619 * its limit temporarily, to help free resources as expected.
2620 */
2621 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2622 return -ENOMEM;
2623
2624 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 if (error)
2626 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002627 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 }
2629
2630 /* Does it split the last one? */
2631 last = find_vma(mm, end);
2632 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002633 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 if (error)
2635 return error;
2636 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002637 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
2639 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002640 * unlock any mlock()ed ranges before detaching vmas
2641 */
2642 if (mm->locked_vm) {
2643 struct vm_area_struct *tmp = vma;
2644 while (tmp && tmp->vm_start < end) {
2645 if (tmp->vm_flags & VM_LOCKED) {
2646 mm->locked_vm -= vma_pages(tmp);
2647 munlock_vma_pages_all(tmp);
2648 }
2649 tmp = tmp->vm_next;
2650 }
2651 }
2652
2653 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 * Remove the vma's, and unmap the actual pages
2655 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002656 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2657 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658
Dave Hansen1de4fa12014-11-14 07:18:31 -08002659 arch_unmap(mm, vma, start, end);
2660
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002662 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
2664 return 0;
2665}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
Al Virobfce2812012-04-20 21:57:04 -04002667int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002668{
2669 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002670 struct mm_struct *mm = current->mm;
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002671
Michal Hockoae798782016-05-23 16:25:33 -07002672 if (down_write_killable(&mm->mmap_sem))
2673 return -EINTR;
2674
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002675 ret = do_munmap(mm, start, len);
2676 up_write(&mm->mmap_sem);
2677 return ret;
2678}
2679EXPORT_SYMBOL(vm_munmap);
2680
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002681SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682{
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002683 int ret;
2684 struct mm_struct *mm = current->mm;
2685
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 profile_munmap(addr);
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002687 if (down_write_killable(&mm->mmap_sem))
2688 return -EINTR;
2689 ret = do_munmap(mm, addr, len);
2690 up_write(&mm->mmap_sem);
2691 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692}
2693
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002694
2695/*
2696 * Emulation of deprecated remap_file_pages() syscall.
2697 */
2698SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2699 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2700{
2701
2702 struct mm_struct *mm = current->mm;
2703 struct vm_area_struct *vma;
2704 unsigned long populate = 0;
2705 unsigned long ret = -EINVAL;
2706 struct file *file;
2707
Joe Perches756a0252016-03-17 14:19:47 -07002708 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2709 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002710
2711 if (prot)
2712 return ret;
2713 start = start & PAGE_MASK;
2714 size = size & PAGE_MASK;
2715
2716 if (start + size <= start)
2717 return ret;
2718
2719 /* Does pgoff wrap? */
2720 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2721 return ret;
2722
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002723 if (down_write_killable(&mm->mmap_sem))
2724 return -EINTR;
2725
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002726 vma = find_vma(mm, start);
2727
2728 if (!vma || !(vma->vm_flags & VM_SHARED))
2729 goto out;
2730
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002731 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002732 goto out;
2733
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002734 if (start + size > vma->vm_end) {
2735 struct vm_area_struct *next;
2736
2737 for (next = vma->vm_next; next; next = next->vm_next) {
2738 /* hole between vmas ? */
2739 if (next->vm_start != next->vm_prev->vm_end)
2740 goto out;
2741
2742 if (next->vm_file != vma->vm_file)
2743 goto out;
2744
2745 if (next->vm_flags != vma->vm_flags)
2746 goto out;
2747
2748 if (start + size <= next->vm_end)
2749 break;
2750 }
2751
2752 if (!next)
2753 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002754 }
2755
2756 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2757 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2758 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2759
2760 flags &= MAP_NONBLOCK;
2761 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2762 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002763 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002764 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002765
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002766 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002767 for (tmp = vma; tmp->vm_start >= start + size;
2768 tmp = tmp->vm_next) {
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07002769 /*
2770 * Split pmd and munlock page on the border
2771 * of the range.
2772 */
2773 vma_adjust_trans_huge(tmp, start, start + size, 0);
2774
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002775 munlock_vma_pages_range(tmp,
2776 max(tmp->vm_start, start),
2777 min(tmp->vm_end, start + size));
2778 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002779 }
2780
2781 file = get_file(vma->vm_file);
2782 ret = do_mmap_pgoff(vma->vm_file, start, size,
2783 prot, flags, pgoff, &populate);
2784 fput(file);
2785out:
2786 up_write(&mm->mmap_sem);
2787 if (populate)
2788 mm_populate(ret, populate);
2789 if (!IS_ERR_VALUE(ret))
2790 ret = 0;
2791 return ret;
2792}
2793
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794static inline void verify_mm_writelocked(struct mm_struct *mm)
2795{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002796#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2798 WARN_ON(1);
2799 up_read(&mm->mmap_sem);
2800 }
2801#endif
2802}
2803
2804/*
2805 * this is really a simplified "do_mmap". it only handles
2806 * anonymous maps. eventually we may be able to do some
2807 * brk-specific accounting here.
2808 */
Kees Cookba093a62016-08-02 14:04:54 -07002809static int do_brk(unsigned long addr, unsigned long request)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810{
vishnu.pscc71aba2014-10-09 15:26:29 -07002811 struct mm_struct *mm = current->mm;
2812 struct vm_area_struct *vma, *prev;
Kees Cookba093a62016-08-02 14:04:54 -07002813 unsigned long flags, len;
vishnu.pscc71aba2014-10-09 15:26:29 -07002814 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002816 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817
Kees Cookba093a62016-08-02 14:04:54 -07002818 len = PAGE_ALIGN(request);
2819 if (len < request)
2820 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002822 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823
Kirill Korotaev3a459752006-09-07 14:17:04 +04002824 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2825
Al Viro2c6a1012009-12-03 19:40:46 -05002826 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002827 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002828 return error;
2829
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002830 error = mlock_future_check(mm, mm->def_flags, len);
2831 if (error)
2832 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
2834 /*
2835 * mm->mmap_sem is required to protect against another thread
2836 * changing the mappings in case we sleep.
2837 */
2838 verify_mm_writelocked(mm);
2839
2840 /*
2841 * Clear old maps. this also does some error checking for us
2842 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002843 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2844 &rb_parent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 if (do_munmap(mm, addr, len))
2846 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 }
2848
2849 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002850 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 return -ENOMEM;
2852
2853 if (mm->map_count > sysctl_max_map_count)
2854 return -ENOMEM;
2855
Al Viro191c5422012-02-13 03:58:52 +00002856 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 return -ENOMEM;
2858
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002860 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002861 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002862 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 goto out;
2864
2865 /*
2866 * create a vma struct for an anonymous mapping
2867 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002868 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 if (!vma) {
2870 vm_unacct_memory(len >> PAGE_SHIFT);
2871 return -ENOMEM;
2872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873
Rik van Riel5beb4932010-03-05 13:42:07 -08002874 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 vma->vm_mm = mm;
2876 vma->vm_start = addr;
2877 vma->vm_end = addr + len;
2878 vma->vm_pgoff = pgoff;
2879 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002880 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 vma_link(mm, vma, prev, rb_link, rb_parent);
2882out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002883 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002885 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002886 if (flags & VM_LOCKED)
2887 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002888 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002889 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890}
2891
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002892int vm_brk(unsigned long addr, unsigned long len)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002893{
2894 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002895 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002896 bool populate;
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002897
Michal Hocko2d6c9282016-05-23 16:25:42 -07002898 if (down_write_killable(&mm->mmap_sem))
2899 return -EINTR;
2900
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002901 ret = do_brk(addr, len);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002902 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002903 up_write(&mm->mmap_sem);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002904 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002905 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002906 return ret;
2907}
2908EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909
2910/* Release all mmaps. */
2911void exit_mmap(struct mm_struct *mm)
2912{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002913 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002914 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 unsigned long nr_accounted = 0;
2916
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002917 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002918 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002919
Rik van Rielba470de2008-10-18 20:26:50 -07002920 if (mm->locked_vm) {
2921 vma = mm->mmap;
2922 while (vma) {
2923 if (vma->vm_flags & VM_LOCKED)
2924 munlock_vma_pages_all(vma);
2925 vma = vma->vm_next;
2926 }
2927 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002928
2929 arch_exit_mmap(mm);
2930
Rik van Rielba470de2008-10-18 20:26:50 -07002931 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08002932 if (!vma) /* Can happen if dup_mmap() received an OOM */
2933 return;
2934
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07002937 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08002938 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07002939 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002940 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07002941
Hugh Dickins6ee86302013-04-29 15:07:44 -07002942 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05002943 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07002946 * Walk the list again, actually closing and freeing it,
2947 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002949 while (vma) {
2950 if (vma->vm_flags & VM_ACCOUNT)
2951 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002952 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002953 }
2954 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955}
2956
2957/* Insert vm structure into process list sorted by address
2958 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08002959 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 */
Hugh Dickins6597d782012-10-08 16:29:07 -07002961int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962{
Hugh Dickins6597d782012-10-08 16:29:07 -07002963 struct vm_area_struct *prev;
2964 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965
Chen Gangc9d13f52015-09-08 15:04:08 -07002966 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2967 &prev, &rb_link, &rb_parent))
2968 return -ENOMEM;
2969 if ((vma->vm_flags & VM_ACCOUNT) &&
2970 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2971 return -ENOMEM;
2972
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 /*
2974 * The vm_pgoff of a purely anonymous vma should be irrelevant
2975 * until its first write fault, when page's anon_vma and index
2976 * are set. But now set the vm_pgoff it will almost certainly
2977 * end up with (unless mremap moves it elsewhere before that
2978 * first wfault), so /proc/pid/maps tells a consistent story.
2979 *
2980 * By setting it to reflect the virtual start address of the
2981 * vma, merges and splits can happen in a seamless way, just
2982 * using the existing file pgoff checks and manipulations.
2983 * Similarly in do_mmap_pgoff and in do_brk.
2984 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07002985 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 BUG_ON(vma->anon_vma);
2987 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2988 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05302989
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 vma_link(mm, vma, prev, rb_link, rb_parent);
2991 return 0;
2992}
2993
2994/*
2995 * Copy the vma structure to a new location in the same mm,
2996 * prior to moving page table entries, to effect an mremap move.
2997 */
2998struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07002999 unsigned long addr, unsigned long len, pgoff_t pgoff,
3000 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001{
3002 struct vm_area_struct *vma = *vmap;
3003 unsigned long vma_start = vma->vm_start;
3004 struct mm_struct *mm = vma->vm_mm;
3005 struct vm_area_struct *new_vma, *prev;
3006 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003007 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
3009 /*
3010 * If anonymous vma has not yet been faulted, update new pgoff
3011 * to match new location, to increase its chance of merging.
3012 */
Oleg Nesterovce757992015-09-08 14:58:34 -07003013 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003015 faulted_in_anon_vma = false;
3016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
Hugh Dickins6597d782012-10-08 16:29:07 -07003018 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3019 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07003021 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3022 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 if (new_vma) {
3024 /*
3025 * Source vma may have been merged into new_vma
3026 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003027 if (unlikely(vma_start >= new_vma->vm_start &&
3028 vma_start < new_vma->vm_end)) {
3029 /*
3030 * The only way we can get a vma_merge with
3031 * self during an mremap is if the vma hasn't
3032 * been faulted in yet and we were allowed to
3033 * reset the dst vma->vm_pgoff to the
3034 * destination address of the mremap to allow
3035 * the merge to happen. mremap must change the
3036 * vm_pgoff linearity between src and dst vmas
3037 * (in turn preventing a vma_merge) to be
3038 * safe. It is only safe to keep the vm_pgoff
3039 * linear if there are no pages mapped yet.
3040 */
Sasha Levin81d1b092014-10-09 15:28:10 -07003041 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07003042 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07003043 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07003044 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08003046 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07003047 if (!new_vma)
3048 goto out;
3049 *new_vma = *vma;
3050 new_vma->vm_start = addr;
3051 new_vma->vm_end = addr + len;
3052 new_vma->vm_pgoff = pgoff;
3053 if (vma_dup_policy(vma, new_vma))
3054 goto out_free_vma;
3055 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3056 if (anon_vma_clone(new_vma, vma))
3057 goto out_free_mempol;
3058 if (new_vma->vm_file)
3059 get_file(new_vma->vm_file);
3060 if (new_vma->vm_ops && new_vma->vm_ops->open)
3061 new_vma->vm_ops->open(new_vma);
3062 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3063 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 }
3065 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003066
Chen Gange3975892015-09-08 15:03:38 -07003067out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07003068 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07003069out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08003070 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07003071out:
Rik van Riel5beb4932010-03-05 13:42:07 -08003072 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003074
3075/*
3076 * Return true if the calling process may expand its vm space by the passed
3077 * number of pages
3078 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003079bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003080{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003081 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3082 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003083
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003084 if (is_data_mapping(flags) &&
3085 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003086 /* Workaround for Valgrind */
3087 if (rlimit(RLIMIT_DATA) == 0 &&
3088 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3089 return true;
3090 if (!ignore_rlimit_data) {
3091 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 -08003092 current->comm, current->pid,
3093 (mm->data_vm + npages) << PAGE_SHIFT,
3094 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003095 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003096 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003097 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07003098
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003099 return true;
3100}
3101
3102void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3103{
3104 mm->total_vm += npages;
3105
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003106 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003107 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003108 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003109 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003110 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003111 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003112}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003113
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003114static int special_mapping_fault(struct vm_area_struct *vma,
3115 struct vm_fault *vmf);
3116
3117/*
3118 * Having a close hook prevents vma merging regardless of flags.
3119 */
3120static void special_mapping_close(struct vm_area_struct *vma)
3121{
3122}
3123
3124static const char *special_mapping_name(struct vm_area_struct *vma)
3125{
3126 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3127}
3128
Dmitry Safonovb059a452016-06-28 14:35:38 +03003129static int special_mapping_mremap(struct vm_area_struct *new_vma)
3130{
3131 struct vm_special_mapping *sm = new_vma->vm_private_data;
3132
3133 if (sm->mremap)
3134 return sm->mremap(sm, new_vma);
3135 return 0;
3136}
3137
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003138static const struct vm_operations_struct special_mapping_vmops = {
3139 .close = special_mapping_close,
3140 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03003141 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003142 .name = special_mapping_name,
3143};
3144
3145static const struct vm_operations_struct legacy_special_mapping_vmops = {
3146 .close = special_mapping_close,
3147 .fault = special_mapping_fault,
3148};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003149
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003150static int special_mapping_fault(struct vm_area_struct *vma,
3151 struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003152{
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003153 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003154 struct page **pages;
3155
Andy Lutomirskif872f542015-12-29 20:12:19 -08003156 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003157 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08003158 } else {
3159 struct vm_special_mapping *sm = vma->vm_private_data;
3160
3161 if (sm->fault)
3162 return sm->fault(sm, vma, vmf);
3163
3164 pages = sm->pages;
3165 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003166
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003167 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003168 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003169
3170 if (*pages) {
3171 struct page *page = *pages;
3172 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003173 vmf->page = page;
3174 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003175 }
3176
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003177 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003178}
3179
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003180static struct vm_area_struct *__install_special_mapping(
3181 struct mm_struct *mm,
3182 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003183 unsigned long vm_flags, void *priv,
3184 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003185{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003186 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003187 struct vm_area_struct *vma;
3188
3189 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3190 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003191 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003192
Rik van Riel5beb4932010-03-05 13:42:07 -08003193 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003194 vma->vm_mm = mm;
3195 vma->vm_start = addr;
3196 vma->vm_end = addr + len;
3197
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003198 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003199 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003200
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003201 vma->vm_ops = ops;
3202 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003203
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003204 ret = insert_vm_struct(mm, vma);
3205 if (ret)
3206 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003207
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003208 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003209
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003210 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003211
Stefani Seibold3935ed62014-03-17 23:22:02 +01003212 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003213
3214out:
3215 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003216 return ERR_PTR(ret);
3217}
3218
Dmitry Safonov2eefd872016-09-05 16:33:05 +03003219bool vma_is_special_mapping(const struct vm_area_struct *vma,
3220 const struct vm_special_mapping *sm)
3221{
3222 return vma->vm_private_data == sm &&
3223 (vma->vm_ops == &special_mapping_vmops ||
3224 vma->vm_ops == &legacy_special_mapping_vmops);
3225}
3226
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003227/*
3228 * Called with mm->mmap_sem held for writing.
3229 * Insert a new vma covering the given region, with the given flags.
3230 * Its pages are supplied by the given array of struct page *.
3231 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3232 * The region past the last page supplied will always produce SIGBUS.
3233 * The array pointer and the pages it points to are assumed to stay alive
3234 * for as long as this mapping might exist.
3235 */
3236struct vm_area_struct *_install_special_mapping(
3237 struct mm_struct *mm,
3238 unsigned long addr, unsigned long len,
3239 unsigned long vm_flags, const struct vm_special_mapping *spec)
3240{
Chen Gang27f28b92015-11-05 18:48:41 -08003241 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3242 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003243}
3244
Stefani Seibold3935ed62014-03-17 23:22:02 +01003245int install_special_mapping(struct mm_struct *mm,
3246 unsigned long addr, unsigned long len,
3247 unsigned long vm_flags, struct page **pages)
3248{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003249 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003250 mm, addr, len, vm_flags, (void *)pages,
3251 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003252
Duan Jiong14bd5b42014-06-04 16:07:05 -07003253 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003254}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003255
3256static DEFINE_MUTEX(mm_all_locks_mutex);
3257
Peter Zijlstra454ed842008-08-11 09:30:25 +02003258static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003259{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003260 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003261 /*
3262 * The LSB of head.next can't change from under us
3263 * because we hold the mm_all_locks_mutex.
3264 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003265 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003266 /*
3267 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003268 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003269 * the same anon_vma we won't take it again.
3270 *
3271 * No need of atomic instructions here, head.next
3272 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003273 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003274 */
3275 if (__test_and_set_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003276 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003277 BUG();
3278 }
3279}
3280
Peter Zijlstra454ed842008-08-11 09:30:25 +02003281static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003282{
3283 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3284 /*
3285 * AS_MM_ALL_LOCKS can't change from under us because
3286 * we hold the mm_all_locks_mutex.
3287 *
3288 * Operations on ->flags have to be atomic because
3289 * even if AS_MM_ALL_LOCKS is stable thanks to the
3290 * mm_all_locks_mutex, there may be other cpus
3291 * changing other bitflags in parallel to us.
3292 */
3293 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3294 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003295 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003296 }
3297}
3298
3299/*
3300 * This operation locks against the VM for all pte/vma/mm related
3301 * operations that could ever happen on a certain mm. This includes
3302 * vmtruncate, try_to_unmap, and all page faults.
3303 *
3304 * The caller must take the mmap_sem in write mode before calling
3305 * mm_take_all_locks(). The caller isn't allowed to release the
3306 * mmap_sem until mm_drop_all_locks() returns.
3307 *
3308 * mmap_sem in write mode is required in order to block all operations
3309 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003310 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003311 * anon_vmas to be associated with existing vmas.
3312 *
3313 * A single task can't take more than one mm_take_all_locks() in a row
3314 * or it would deadlock.
3315 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003316 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003317 * mapping->flags avoid to take the same lock twice, if more than one
3318 * vma in this mm is backed by the same anon_vma or address_space.
3319 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003320 * We take locks in following order, accordingly to comment at beginning
3321 * of mm/rmap.c:
3322 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3323 * hugetlb mapping);
3324 * - all i_mmap_rwsem locks;
3325 * - all anon_vma->rwseml
3326 *
3327 * We can take all locks within these types randomly because the VM code
3328 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3329 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003330 *
3331 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3332 * that may have to take thousand of locks.
3333 *
3334 * mm_take_all_locks() can fail if it's interrupted by signals.
3335 */
3336int mm_take_all_locks(struct mm_struct *mm)
3337{
3338 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003339 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003340
3341 BUG_ON(down_read_trylock(&mm->mmap_sem));
3342
3343 mutex_lock(&mm_all_locks_mutex);
3344
3345 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3346 if (signal_pending(current))
3347 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003348 if (vma->vm_file && vma->vm_file->f_mapping &&
3349 is_vm_hugetlb_page(vma))
3350 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3351 }
3352
3353 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3354 if (signal_pending(current))
3355 goto out_unlock;
3356 if (vma->vm_file && vma->vm_file->f_mapping &&
3357 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003358 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003359 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003360
3361 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3362 if (signal_pending(current))
3363 goto out_unlock;
3364 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003365 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3366 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003367 }
3368
Kautuk Consul584cff52011-10-31 17:08:59 -07003369 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003370
3371out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003372 mm_drop_all_locks(mm);
3373 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003374}
3375
3376static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3377{
Michel Lespinassebf181b92012-10-08 16:31:39 -07003378 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003379 /*
3380 * The LSB of head.next can't change to 0 from under
3381 * us because we hold the mm_all_locks_mutex.
3382 *
3383 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003384 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003385 * never see our bitflag.
3386 *
3387 * No need of atomic instructions here, head.next
3388 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003389 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003390 */
3391 if (!__test_and_clear_bit(0, (unsigned long *)
Michel Lespinassebf181b92012-10-08 16:31:39 -07003392 &anon_vma->root->rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003393 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003394 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003395 }
3396}
3397
3398static void vm_unlock_mapping(struct address_space *mapping)
3399{
3400 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3401 /*
3402 * AS_MM_ALL_LOCKS can't change to 0 from under us
3403 * because we hold the mm_all_locks_mutex.
3404 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003405 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003406 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3407 &mapping->flags))
3408 BUG();
3409 }
3410}
3411
3412/*
3413 * The mmap_sem cannot be released by the caller until
3414 * mm_drop_all_locks() returns.
3415 */
3416void mm_drop_all_locks(struct mm_struct *mm)
3417{
3418 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003419 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003420
3421 BUG_ON(down_read_trylock(&mm->mmap_sem));
3422 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3423
3424 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3425 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003426 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3427 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003428 if (vma->vm_file && vma->vm_file->f_mapping)
3429 vm_unlock_mapping(vma->vm_file->f_mapping);
3430 }
3431
3432 mutex_unlock(&mm_all_locks_mutex);
3433}
David Howells8feae132009-01-08 12:04:47 +00003434
3435/*
3436 * initialise the VMA slab
3437 */
3438void __init mmap_init(void)
3439{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003440 int ret;
3441
Tejun Heo908c7f12014-09-08 09:51:29 +09003442 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003443 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003444}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003445
3446/*
3447 * Initialise sysctl_user_reserve_kbytes.
3448 *
3449 * This is intended to prevent a user from starting a single memory hogging
3450 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3451 * mode.
3452 *
3453 * The default value is min(3% of free memory, 128MB)
3454 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3455 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003456static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003457{
3458 unsigned long free_kbytes;
3459
3460 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3461
3462 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3463 return 0;
3464}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003465subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003466
3467/*
3468 * Initialise sysctl_admin_reserve_kbytes.
3469 *
3470 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3471 * to log in and kill a memory hogging process.
3472 *
3473 * Systems with more than 256MB will reserve 8MB, enough to recover
3474 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3475 * only reserve 3% of free pages by default.
3476 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003477static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003478{
3479 unsigned long free_kbytes;
3480
3481 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3482
3483 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3484 return 0;
3485}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003486subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003487
3488/*
3489 * Reinititalise user and admin reserves if memory is added or removed.
3490 *
3491 * The default user reserve max is 128MB, and the default max for the
3492 * admin reserve is 8MB. These are usually, but not always, enough to
3493 * enable recovery from a memory hogging process using login/sshd, a shell,
3494 * and tools like top. It may make sense to increase or even disable the
3495 * reserve depending on the existence of swap or variations in the recovery
3496 * tools. So, the admin may have changed them.
3497 *
3498 * If memory is added and the reserves have been eliminated or increased above
3499 * the default max, then we'll trust the admin.
3500 *
3501 * If memory is removed and there isn't enough free memory, then we
3502 * need to reset the reserves.
3503 *
3504 * Otherwise keep the reserve set by the admin.
3505 */
3506static int reserve_mem_notifier(struct notifier_block *nb,
3507 unsigned long action, void *data)
3508{
3509 unsigned long tmp, free_kbytes;
3510
3511 switch (action) {
3512 case MEM_ONLINE:
3513 /* Default max is 128MB. Leave alone if modified by operator. */
3514 tmp = sysctl_user_reserve_kbytes;
3515 if (0 < tmp && tmp < (1UL << 17))
3516 init_user_reserve();
3517
3518 /* Default max is 8MB. Leave alone if modified by operator. */
3519 tmp = sysctl_admin_reserve_kbytes;
3520 if (0 < tmp && tmp < (1UL << 13))
3521 init_admin_reserve();
3522
3523 break;
3524 case MEM_OFFLINE:
3525 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3526
3527 if (sysctl_user_reserve_kbytes > free_kbytes) {
3528 init_user_reserve();
3529 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3530 sysctl_user_reserve_kbytes);
3531 }
3532
3533 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3534 init_admin_reserve();
3535 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3536 sysctl_admin_reserve_kbytes);
3537 }
3538 break;
3539 default:
3540 break;
3541 }
3542 return NOTIFY_OK;
3543}
3544
3545static struct notifier_block reserve_mem_nb = {
3546 .notifier_call = reserve_mem_notifier,
3547};
3548
3549static int __meminit init_reserve_notifier(void)
3550{
3551 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003552 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003553
3554 return 0;
3555}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003556subsys_initcall(init_reserve_notifier);