blob: 924839fac0e6421a77839825a99833a342d3153c [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>
Andrea Arcangeli21292582017-09-06 16:25:00 -070047#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080049#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <asm/cacheflush.h>
51#include <asm/tlb.h>
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +020052#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Jan Beulich42b77722008-07-23 21:27:10 -070054#include "internal.h"
55
Kirill Korotaev3a459752006-09-07 14:17:04 +040056#ifndef arch_mmap_check
57#define arch_mmap_check(addr, len, flags) (0)
58#endif
59
Daniel Cashmand07e2252016-01-14 15:19:53 -080060#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
61const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
62const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
63int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
64#endif
65#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
66const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
67const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
68int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
69#endif
70
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -070071static bool ignore_rlimit_data;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -080072core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
Daniel Cashmand07e2252016-01-14 15:19:53 -080073
Hugh Dickinse0da3822005-04-19 13:29:15 -070074static void unmap_region(struct mm_struct *mm,
75 struct vm_area_struct *vma, struct vm_area_struct *prev,
76 unsigned long start, unsigned long end);
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/* description of effects of mapping type and prot in current implementation.
79 * this is due to the limited x86 page protection hardware. The expected
80 * behavior is in parens:
81 *
82 * map_type prot
83 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
84 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
85 * w: (no) no w: (no) no w: (yes) yes w: (no) no
86 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
vishnu.pscc71aba2014-10-09 15:26:29 -070087 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
89 * w: (no) no w: (no) no w: (copy) copy w: (no) no
90 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
91 *
Catalin Marinascab15ce2016-08-11 18:44:50 +010092 * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
93 * MAP_PRIVATE:
94 * r: (no) no
95 * w: (no) no
96 * x: (yes) yes
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 */
Daniel Micayac34cea2017-07-06 15:36:47 -070098pgprot_t protection_map[16] __ro_after_init = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
100 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
101};
102
Hugh Dickins804af2c2006-07-26 21:39:49 +0100103pgprot_t vm_get_page_prot(unsigned long vm_flags)
104{
Dave Kleikampb845f312008-07-08 00:28:51 +1000105 return __pgprot(pgprot_val(protection_map[vm_flags &
106 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
107 pgprot_val(arch_vm_get_page_prot(vm_flags)));
Hugh Dickins804af2c2006-07-26 21:39:49 +0100108}
109EXPORT_SYMBOL(vm_get_page_prot);
110
Peter Feiner64e45502014-10-13 15:55:46 -0700111static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
112{
113 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
114}
115
116/* Update vma->vm_page_prot to reflect vma->vm_flags. */
117void vma_set_page_prot(struct vm_area_struct *vma)
118{
119 unsigned long vm_flags = vma->vm_flags;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700120 pgprot_t vm_page_prot;
Peter Feiner64e45502014-10-13 15:55:46 -0700121
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700122 vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
123 if (vma_wants_writenotify(vma, vm_page_prot)) {
Peter Feiner64e45502014-10-13 15:55:46 -0700124 vm_flags &= ~VM_SHARED;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700125 vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
Peter Feiner64e45502014-10-13 15:55:46 -0700126 }
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700127 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
128 WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
Peter Feiner64e45502014-10-13 15:55:46 -0700129}
130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800132 * Requires inode->i_mapping->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 */
134static void __remove_shared_vm_struct(struct vm_area_struct *vma,
135 struct file *file, struct address_space *mapping)
136{
137 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500138 atomic_inc(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700140 mapping_unmap_writable(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800143 vma_interval_tree_remove(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 flush_dcache_mmap_unlock(mapping);
145}
146
147/*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700148 * Unlink a file-based vm structure from its interval tree, to hide
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700149 * vma from rmap and vmtruncate before freeing its page tables.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 */
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700151void unlink_file_vma(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{
153 struct file *file = vma->vm_file;
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (file) {
156 struct address_space *mapping = file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800157 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 __remove_shared_vm_struct(vma, file, mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800159 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 }
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700161}
162
163/*
164 * Close a vm structure and free it, returning the next.
165 */
166static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
167{
168 struct vm_area_struct *next = vma->vm_next;
169
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700170 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 if (vma->vm_ops && vma->vm_ops->close)
172 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700173 if (vma->vm_file)
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700174 fput(vma->vm_file);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700175 mpol_put(vma_policy(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kmem_cache_free(vm_area_cachep, vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -0700177 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800180static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -0700181
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100182SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700184 unsigned long retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 unsigned long newbrk, oldbrk;
186 struct mm_struct *mm = current->mm;
Hugh Dickins1be71072017-06-19 04:03:24 -0700187 struct vm_area_struct *next;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700188 unsigned long min_brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800189 bool populate;
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800190 LIST_HEAD(uf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700192 if (down_write_killable(&mm->mmap_sem))
193 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700195#ifdef CONFIG_COMPAT_BRK
Jiri Kosina5520e892011-01-13 15:47:23 -0800196 /*
197 * CONFIG_COMPAT_BRK can still be overridden by setting
198 * randomize_va_space to 2, which will still cause mm->start_brk
199 * to be arbitrarily shifted
200 */
Jiri Kosina4471a672011-04-14 15:22:09 -0700201 if (current->brk_randomized)
Jiri Kosina5520e892011-01-13 15:47:23 -0800202 min_brk = mm->start_brk;
203 else
204 min_brk = mm->end_data;
Jiri Kosinaa5b45922008-06-05 22:46:05 -0700205#else
206 min_brk = mm->start_brk;
207#endif
208 if (brk < min_brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 goto out;
Ram Gupta1e624192006-04-10 22:52:57 -0700210
211 /*
212 * Check against rlimit here. If this check is done later after the test
213 * of oldbrk with newbrk then it can escape the test and let the data
214 * segment grow beyond its set limit the in case where the limit is
215 * not page aligned -Ram Gupta
216 */
Cyrill Gorcunov8764b332014-10-09 15:27:32 -0700217 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
218 mm->end_data, mm->start_data))
Ram Gupta1e624192006-04-10 22:52:57 -0700219 goto out;
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 newbrk = PAGE_ALIGN(brk);
222 oldbrk = PAGE_ALIGN(mm->brk);
223 if (oldbrk == newbrk)
224 goto set_brk;
225
226 /* Always allow shrinking brk. */
227 if (brk <= mm->brk) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800228 if (!do_munmap(mm, newbrk, oldbrk-newbrk, &uf))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 goto set_brk;
230 goto out;
231 }
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 /* Check against existing mmap mappings. */
Hugh Dickins1be71072017-06-19 04:03:24 -0700234 next = find_vma(mm, oldbrk);
235 if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 goto out;
237
238 /* Ok, looks good - let it rip. */
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800239 if (do_brk(oldbrk, newbrk-oldbrk, &uf) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 goto out;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242set_brk:
243 mm->brk = brk;
Michel Lespinasse128557f2013-02-22 16:32:40 -0800244 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
245 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800246 userfaultfd_unmap_complete(mm, &uf);
Michel Lespinasse128557f2013-02-22 16:32:40 -0800247 if (populate)
248 mm_populate(oldbrk, newbrk - oldbrk);
249 return brk;
250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251out:
252 retval = mm->brk;
253 up_write(&mm->mmap_sem);
254 return retval;
255}
256
Michel Lespinassed3737182012-12-11 16:01:38 -0800257static long vma_compute_subtree_gap(struct vm_area_struct *vma)
258{
Hugh Dickins1be71072017-06-19 04:03:24 -0700259 unsigned long max, prev_end, subtree_gap;
260
261 /*
262 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
263 * allow two stack_guard_gaps between them here, and when choosing
264 * an unmapped area; whereas when expanding we only require one.
265 * That's a little inconsistent, but keeps the code here simpler.
266 */
267 max = vm_start_gap(vma);
268 if (vma->vm_prev) {
269 prev_end = vm_end_gap(vma->vm_prev);
270 if (max > prev_end)
271 max -= prev_end;
272 else
273 max = 0;
274 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800275 if (vma->vm_rb.rb_left) {
276 subtree_gap = rb_entry(vma->vm_rb.rb_left,
277 struct vm_area_struct, vm_rb)->rb_subtree_gap;
278 if (subtree_gap > max)
279 max = subtree_gap;
280 }
281 if (vma->vm_rb.rb_right) {
282 subtree_gap = rb_entry(vma->vm_rb.rb_right,
283 struct vm_area_struct, vm_rb)->rb_subtree_gap;
284 if (subtree_gap > max)
285 max = subtree_gap;
286 }
287 return max;
288}
289
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700290#ifdef CONFIG_DEBUG_VM_RB
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800291static int browse_rb(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800293 struct rb_root *root = &mm->mm_rb;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800294 int i = 0, j, bug = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 struct rb_node *nd, *pn = NULL;
296 unsigned long prev = 0, pend = 0;
297
298 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
299 struct vm_area_struct *vma;
300 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800301 if (vma->vm_start < prev) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700302 pr_emerg("vm_start %lx < prev %lx\n",
303 vma->vm_start, prev);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800304 bug = 1;
305 }
306 if (vma->vm_start < pend) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700307 pr_emerg("vm_start %lx < pend %lx\n",
308 vma->vm_start, pend);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800309 bug = 1;
310 }
311 if (vma->vm_start > vma->vm_end) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700312 pr_emerg("vm_start %lx > vm_end %lx\n",
313 vma->vm_start, vma->vm_end);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800314 bug = 1;
315 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800316 spin_lock(&mm->page_table_lock);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800317 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700318 pr_emerg("free gap %lx, correct %lx\n",
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800319 vma->rb_subtree_gap,
320 vma_compute_subtree_gap(vma));
321 bug = 1;
322 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800323 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 i++;
325 pn = nd;
David Millerd1af65d2007-02-28 20:13:13 -0800326 prev = vma->vm_start;
327 pend = vma->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 }
329 j = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800330 for (nd = pn; nd; nd = rb_prev(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 j++;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800332 if (i != j) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700333 pr_emerg("backwards %d, forwards %d\n", j, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800334 bug = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800336 return bug ? -1 : i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337}
338
Michel Lespinassed3737182012-12-11 16:01:38 -0800339static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
340{
341 struct rb_node *nd;
342
343 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
344 struct vm_area_struct *vma;
345 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
Sasha Levin96dad672014-10-09 15:28:39 -0700346 VM_BUG_ON_VMA(vma != ignore &&
347 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
348 vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
Rashika Kheriaeafd4dc2014-04-03 14:48:03 -0700352static void validate_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
354 int bug = 0;
355 int i = 0;
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800356 unsigned long highest_address = 0;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700357 struct vm_area_struct *vma = mm->mmap;
Andrew Mortonff26f702014-10-09 15:28:19 -0700358
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700359 while (vma) {
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800360 struct anon_vma *anon_vma = vma->anon_vma;
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700361 struct anon_vma_chain *avc;
Andrew Mortonff26f702014-10-09 15:28:19 -0700362
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -0800363 if (anon_vma) {
364 anon_vma_lock_read(anon_vma);
365 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
366 anon_vma_interval_tree_verify(avc);
367 anon_vma_unlock_read(anon_vma);
368 }
369
Hugh Dickins1be71072017-06-19 04:03:24 -0700370 highest_address = vm_end_gap(vma);
Michel Lespinasseed8ea812012-10-08 16:31:45 -0700371 vma = vma->vm_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 i++;
373 }
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800374 if (i != mm->map_count) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700375 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800376 bug = 1;
377 }
378 if (highest_address != mm->highest_vm_end) {
Sasha Levin8542bdf2014-09-09 14:50:59 -0700379 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
Andrew Mortonff26f702014-10-09 15:28:19 -0700380 mm->highest_vm_end, highest_address);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800381 bug = 1;
382 }
Andrea Arcangeliacf128d2016-02-05 15:36:13 -0800383 i = browse_rb(mm);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800384 if (i != mm->map_count) {
Andrew Mortonff26f702014-10-09 15:28:19 -0700385 if (i != -1)
386 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
Michel Lespinasse5a0768f2012-12-11 16:01:42 -0800387 bug = 1;
388 }
Sasha Levin96dad672014-10-09 15:28:39 -0700389 VM_BUG_ON_MM(bug, mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391#else
Michel Lespinassed3737182012-12-11 16:01:38 -0800392#define validate_mm_rb(root, ignore) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393#define validate_mm(mm) do { } while (0)
394#endif
395
Michel Lespinassed3737182012-12-11 16:01:38 -0800396RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
397 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
398
399/*
400 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
401 * vma->vm_prev->vm_end values changed, without modifying the vma's position
402 * in the rbtree.
403 */
404static void vma_gap_update(struct vm_area_struct *vma)
405{
406 /*
407 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
408 * function that does exacltly what we want.
409 */
410 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
411}
412
413static inline void vma_rb_insert(struct vm_area_struct *vma,
414 struct rb_root *root)
415{
416 /* All rb_subtree_gap values must be consistent prior to insertion */
417 validate_mm_rb(root, NULL);
418
419 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
420}
421
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700422static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
423{
424 /*
425 * Note rb_erase_augmented is a fairly large inline function,
426 * so make sure we instantiate it only once with our desired
427 * augmented rbtree callbacks.
428 */
429 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
430}
431
432static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
433 struct rb_root *root,
434 struct vm_area_struct *ignore)
435{
436 /*
437 * All rb_subtree_gap values must be consistent prior to erase,
438 * with the possible exception of the "next" vma being erased if
439 * next->vm_start was reduced.
440 */
441 validate_mm_rb(root, ignore);
442
443 __vma_rb_erase(vma, root);
444}
445
446static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
447 struct rb_root *root)
Michel Lespinassed3737182012-12-11 16:01:38 -0800448{
449 /*
450 * All rb_subtree_gap values must be consistent prior to erase,
451 * with the possible exception of the vma being erased.
452 */
453 validate_mm_rb(root, vma);
454
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700455 __vma_rb_erase(vma, root);
Michel Lespinassed3737182012-12-11 16:01:38 -0800456}
457
Michel Lespinassebf181b92012-10-08 16:31:39 -0700458/*
459 * vma has some anon_vma assigned, and is already inserted on that
460 * anon_vma's interval trees.
461 *
462 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
463 * vma must be removed from the anon_vma's interval trees using
464 * anon_vma_interval_tree_pre_update_vma().
465 *
466 * After the update, the vma will be reinserted using
467 * anon_vma_interval_tree_post_update_vma().
468 *
469 * The entire update must be protected by exclusive mmap_sem and by
470 * the root anon_vma's mutex.
471 */
472static inline void
473anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
474{
475 struct anon_vma_chain *avc;
476
477 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
478 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
479}
480
481static inline void
482anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
483{
484 struct anon_vma_chain *avc;
485
486 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
487 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
488}
489
Hugh Dickins6597d782012-10-08 16:29:07 -0700490static int find_vma_links(struct mm_struct *mm, unsigned long addr,
491 unsigned long end, struct vm_area_struct **pprev,
492 struct rb_node ***rb_link, struct rb_node **rb_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
Hugh Dickins6597d782012-10-08 16:29:07 -0700494 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 __rb_link = &mm->mm_rb.rb_node;
497 rb_prev = __rb_parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
499 while (*__rb_link) {
500 struct vm_area_struct *vma_tmp;
501
502 __rb_parent = *__rb_link;
503 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
504
505 if (vma_tmp->vm_end > addr) {
Hugh Dickins6597d782012-10-08 16:29:07 -0700506 /* Fail if an existing vma overlaps the area */
507 if (vma_tmp->vm_start < end)
508 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 __rb_link = &__rb_parent->rb_left;
510 } else {
511 rb_prev = __rb_parent;
512 __rb_link = &__rb_parent->rb_right;
513 }
514 }
515
516 *pprev = NULL;
517 if (rb_prev)
518 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
519 *rb_link = __rb_link;
520 *rb_parent = __rb_parent;
Hugh Dickins6597d782012-10-08 16:29:07 -0700521 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
Cyril Hrubise8420a82013-04-29 15:08:33 -0700524static unsigned long count_vma_pages_range(struct mm_struct *mm,
525 unsigned long addr, unsigned long end)
526{
527 unsigned long nr_pages = 0;
528 struct vm_area_struct *vma;
529
530 /* Find first overlaping mapping */
531 vma = find_vma_intersection(mm, addr, end);
532 if (!vma)
533 return 0;
534
535 nr_pages = (min(end, vma->vm_end) -
536 max(addr, vma->vm_start)) >> PAGE_SHIFT;
537
538 /* Iterate over the rest of the overlaps */
539 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
540 unsigned long overlap_len;
541
542 if (vma->vm_start > end)
543 break;
544
545 overlap_len = min(end, vma->vm_end) - vma->vm_start;
546 nr_pages += overlap_len >> PAGE_SHIFT;
547 }
548
549 return nr_pages;
550}
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
553 struct rb_node **rb_link, struct rb_node *rb_parent)
554{
Michel Lespinassed3737182012-12-11 16:01:38 -0800555 /* Update tracking information for the gap following the new vma. */
556 if (vma->vm_next)
557 vma_gap_update(vma->vm_next);
558 else
Hugh Dickins1be71072017-06-19 04:03:24 -0700559 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800560
561 /*
562 * vma->vm_prev wasn't known when we followed the rbtree to find the
563 * correct insertion point for that vma. As a result, we could not
564 * update the vma vm_rb parents rb_subtree_gap values on the way down.
565 * So, we first insert the vma with a zero rb_subtree_gap value
566 * (to be consistent with what we did on the way down), and then
567 * immediately update the gap to the correct value. Finally we
568 * rebalance the rbtree after all augmented values have been set.
569 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
Michel Lespinassed3737182012-12-11 16:01:38 -0800571 vma->rb_subtree_gap = 0;
572 vma_gap_update(vma);
573 vma_rb_insert(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
Denys Vlasenkocb8f4882008-10-18 20:27:01 -0700576static void __vma_link_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
ZhenwenXu48aae422009-01-06 14:40:21 -0800578 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 file = vma->vm_file;
581 if (file) {
582 struct address_space *mapping = file->f_mapping;
583
584 if (vma->vm_flags & VM_DENYWRITE)
Al Viro496ad9a2013-01-23 17:07:38 -0500585 atomic_dec(&file_inode(file)->i_writecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (vma->vm_flags & VM_SHARED)
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700587 atomic_inc(&mapping->i_mmap_writable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 flush_dcache_mmap_lock(mapping);
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800590 vma_interval_tree_insert(vma, &mapping->i_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 flush_dcache_mmap_unlock(mapping);
592 }
593}
594
595static void
596__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
597 struct vm_area_struct *prev, struct rb_node **rb_link,
598 struct rb_node *rb_parent)
599{
600 __vma_link_list(mm, vma, prev, rb_parent);
601 __vma_link_rb(mm, vma, rb_link, rb_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602}
603
604static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
605 struct vm_area_struct *prev, struct rb_node **rb_link,
606 struct rb_node *rb_parent)
607{
608 struct address_space *mapping = NULL;
609
Huang Shijie64ac4942014-06-04 16:07:33 -0700610 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800612 i_mmap_lock_write(mapping);
Huang Shijie64ac4942014-06-04 16:07:33 -0700613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615 __vma_link(mm, vma, prev, rb_link, rb_parent);
616 __vma_link_file(vma);
617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800619 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 mm->map_count++;
622 validate_mm(mm);
623}
624
625/*
Kautuk Consul88f6b4c2012-03-21 16:34:16 -0700626 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700627 * mm's list and rbtree. It has already been inserted into the interval tree.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
ZhenwenXu48aae422009-01-06 14:40:21 -0800629static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Hugh Dickins6597d782012-10-08 16:29:07 -0700631 struct vm_area_struct *prev;
ZhenwenXu48aae422009-01-06 14:40:21 -0800632 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Hugh Dickins6597d782012-10-08 16:29:07 -0700634 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
635 &prev, &rb_link, &rb_parent))
636 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 __vma_link(mm, vma, prev, rb_link, rb_parent);
638 mm->map_count++;
639}
640
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700641static __always_inline void __vma_unlink_common(struct mm_struct *mm,
642 struct vm_area_struct *vma,
643 struct vm_area_struct *prev,
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700644 bool has_prev,
645 struct vm_area_struct *ignore)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
Michel Lespinassed3737182012-12-11 16:01:38 -0800647 struct vm_area_struct *next;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700648
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700649 vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700650 next = vma->vm_next;
651 if (has_prev)
652 prev->vm_next = next;
653 else {
654 prev = vma->vm_prev;
655 if (prev)
656 prev->vm_next = next;
657 else
658 mm->mmap = next;
659 }
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700660 if (next)
661 next->vm_prev = prev;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700662
663 /* Kill the cache */
664 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700667static inline void __vma_unlink_prev(struct mm_struct *mm,
668 struct vm_area_struct *vma,
669 struct vm_area_struct *prev)
670{
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700671 __vma_unlink_common(mm, vma, prev, true, vma);
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700672}
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674/*
675 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
676 * is already present in an i_mmap tree without adjusting the tree.
677 * The following helper function should be used when such adjustments
678 * are necessary. The "insert" vma (if any) is to be inserted
679 * before we drop the necessary locks.
680 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700681int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
682 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
683 struct vm_area_struct *expand)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
685 struct mm_struct *mm = vma->vm_mm;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700686 struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 struct address_space *mapping = NULL;
Davidlohr Buesof808c132017-09-08 16:15:08 -0700688 struct rb_root_cached *root = NULL;
Rik van Riel012f18002010-08-09 17:18:40 -0700689 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 struct file *file = vma->vm_file;
Michel Lespinassed3737182012-12-11 16:01:38 -0800691 bool start_changed = false, end_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 long adjust_next = 0;
693 int remove_next = 0;
694
695 if (next && !insert) {
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700696 struct vm_area_struct *exporter = NULL, *importer = NULL;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700697
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 if (end >= next->vm_end) {
699 /*
700 * vma expands, overlapping all the next, and
701 * perhaps the one after too (mprotect case 6).
Andrea Arcangeli86d12e42016-10-07 17:01:34 -0700702 * The only other cases that gets here are
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700703 * case 1, case 7 and case 8.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700705 if (next == expand) {
706 /*
707 * The only case where we don't expand "vma"
708 * and we expand "next" instead is case 8.
709 */
710 VM_WARN_ON(end != next->vm_end);
711 /*
712 * remove_next == 3 means we're
713 * removing "vma" and that to do so we
714 * swapped "vma" and "next".
715 */
716 remove_next = 3;
717 VM_WARN_ON(file != next->vm_file);
718 swap(vma, next);
719 } else {
720 VM_WARN_ON(expand != vma);
721 /*
722 * case 1, 6, 7, remove_next == 2 is case 6,
723 * remove_next == 1 is case 1 or 7.
724 */
725 remove_next = 1 + (end > next->vm_end);
726 VM_WARN_ON(remove_next == 2 &&
727 end != next->vm_next->vm_end);
728 VM_WARN_ON(remove_next == 1 &&
729 end != next->vm_end);
730 /* trim end to next, for case 6 first pass */
731 end = next->vm_end;
732 }
733
Linus Torvalds287d97a2010-04-10 15:22:30 -0700734 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 importer = vma;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700736
737 /*
738 * If next doesn't have anon_vma, import from vma after
739 * next, if the vma overlaps with it.
740 */
Andrea Arcangeli97a42cd2016-10-07 17:01:31 -0700741 if (remove_next == 2 && !next->anon_vma)
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700742 exporter = next->vm_next;
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 } else if (end > next->vm_start) {
745 /*
746 * vma expands, overlapping part of the next:
747 * mprotect case 5 shifting the boundary up.
748 */
749 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
Linus Torvalds287d97a2010-04-10 15:22:30 -0700750 exporter = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 importer = vma;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700752 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 } else if (end < vma->vm_end) {
754 /*
755 * vma shrinks, and !insert tells it's not
756 * split_vma inserting another: so it must be
757 * mprotect case 4 shifting the boundary down.
758 */
vishnu.pscc71aba2014-10-09 15:26:29 -0700759 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
Linus Torvalds287d97a2010-04-10 15:22:30 -0700760 exporter = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 importer = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700762 VM_WARN_ON(expand != importer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Rik van Riel5beb4932010-03-05 13:42:07 -0800765 /*
766 * Easily overlooked: when mprotect shifts the boundary,
767 * make sure the expanding vma has anon_vma set if the
768 * shrinking vma had, to cover any anon pages imported.
769 */
Linus Torvalds287d97a2010-04-10 15:22:30 -0700770 if (exporter && exporter->anon_vma && !importer->anon_vma) {
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800771 int error;
772
Linus Torvalds287d97a2010-04-10 15:22:30 -0700773 importer->anon_vma = exporter->anon_vma;
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300774 error = anon_vma_clone(importer, exporter);
Leon Yu3fe89b32015-03-25 15:55:11 -0700775 if (error)
Konstantin Khlebnikovb800c912015-01-11 16:54:06 +0300776 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800777 }
778 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700779again:
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700780 vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
Kirill A. Shutemov37f9f552016-07-26 15:25:48 -0700781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 if (file) {
783 mapping = file->f_mapping;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800784 root = &mapping->i_mmap;
785 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530786
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800787 if (adjust_next)
788 uprobe_munmap(next, next->vm_start, next->vm_end);
Srikar Dronamraju682968e2012-03-30 23:56:46 +0530789
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800790 i_mmap_lock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if (insert) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 /*
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700793 * Put into interval tree now, so instantiated pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 * are visible to arm/parisc __flush_dcache_page
795 * throughout; but we cannot insert into address
796 * space until vma start or end is updated.
797 */
798 __vma_link_file(insert);
799 }
800 }
801
Michel Lespinassebf181b92012-10-08 16:31:39 -0700802 anon_vma = vma->anon_vma;
803 if (!anon_vma && adjust_next)
804 anon_vma = next->anon_vma;
805 if (anon_vma) {
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700806 VM_WARN_ON(adjust_next && next->anon_vma &&
807 anon_vma != next->anon_vma);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000808 anon_vma_lock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700809 anon_vma_interval_tree_pre_update_vma(vma);
810 if (adjust_next)
811 anon_vma_interval_tree_pre_update_vma(next);
812 }
Rik van Riel012f18002010-08-09 17:18:40 -0700813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 if (root) {
815 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700816 vma_interval_tree_remove(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700818 vma_interval_tree_remove(next, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 }
820
Michel Lespinassed3737182012-12-11 16:01:38 -0800821 if (start != vma->vm_start) {
822 vma->vm_start = start;
823 start_changed = true;
824 }
825 if (end != vma->vm_end) {
826 vma->vm_end = end;
827 end_changed = true;
828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 vma->vm_pgoff = pgoff;
830 if (adjust_next) {
831 next->vm_start += adjust_next << PAGE_SHIFT;
832 next->vm_pgoff += adjust_next;
833 }
834
835 if (root) {
836 if (adjust_next)
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700837 vma_interval_tree_insert(next, root);
838 vma_interval_tree_insert(vma, root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 flush_dcache_mmap_unlock(mapping);
840 }
841
842 if (remove_next) {
843 /*
844 * vma_merge has merged next into vma, and needs
845 * us to remove next before dropping the locks.
846 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700847 if (remove_next != 3)
848 __vma_unlink_prev(mm, next, vma);
849 else
Andrea Arcangeli8f26e0b2016-10-07 17:01:37 -0700850 /*
851 * vma is not before next if they've been
852 * swapped.
853 *
854 * pre-swap() next->vm_start was reduced so
855 * tell validate_mm_rb to ignore pre-swap()
856 * "next" (which is stored in post-swap()
857 * "vma").
858 */
859 __vma_unlink_common(mm, next, NULL, false, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 if (file)
861 __remove_shared_vm_struct(next, file, mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 } else if (insert) {
863 /*
864 * split_vma has split insert from vma, and needs
865 * us to insert it before dropping the locks
866 * (it may either follow vma or precede it).
867 */
868 __insert_vm_struct(mm, insert);
Michel Lespinassed3737182012-12-11 16:01:38 -0800869 } else {
870 if (start_changed)
871 vma_gap_update(vma);
872 if (end_changed) {
873 if (!next)
Hugh Dickins1be71072017-06-19 04:03:24 -0700874 mm->highest_vm_end = vm_end_gap(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -0800875 else if (!adjust_next)
876 vma_gap_update(next);
877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879
Michel Lespinassebf181b92012-10-08 16:31:39 -0700880 if (anon_vma) {
881 anon_vma_interval_tree_post_update_vma(vma);
882 if (adjust_next)
883 anon_vma_interval_tree_post_update_vma(next);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800884 anon_vma_unlock_write(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800887 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530889 if (root) {
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100890 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530891
892 if (adjust_next)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100893 uprobe_mmap(next);
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530894 }
895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (remove_next) {
Matt Helsley925d1c42008-04-29 01:01:36 -0700897 if (file) {
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +0530898 uprobe_munmap(next, next->vm_start, next->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 fput(file);
Matt Helsley925d1c42008-04-29 01:01:36 -0700900 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800901 if (next->anon_vma)
902 anon_vma_merge(vma, next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 mm->map_count--;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700904 mpol_put(vma_policy(next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 kmem_cache_free(vm_area_cachep, next);
906 /*
907 * In mprotect's case 6 (see comments on vma_merge),
908 * we must remove another next too. It would clutter
909 * up the code too much to do both in one go.
910 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700911 if (remove_next != 3) {
912 /*
913 * If "next" was removed and vma->vm_end was
914 * expanded (up) over it, in turn
915 * "next->vm_prev->vm_end" changed and the
916 * "vma->vm_next" gap must be updated.
917 */
918 next = vma->vm_next;
919 } else {
920 /*
921 * For the scope of the comment "next" and
922 * "vma" considered pre-swap(): if "vma" was
923 * removed, next->vm_start was expanded (down)
924 * over it and the "next" gap must be updated.
925 * Because of the swap() the post-swap() "vma"
926 * actually points to pre-swap() "next"
927 * (post-swap() "next" as opposed is now a
928 * dangling pointer).
929 */
930 next = vma;
931 }
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700932 if (remove_next == 2) {
933 remove_next = 1;
934 end = next->vm_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 goto again;
Kirill A. Shutemov734537c2016-07-28 15:49:01 -0700936 }
Michel Lespinassed3737182012-12-11 16:01:38 -0800937 else if (next)
938 vma_gap_update(next);
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700939 else {
940 /*
941 * If remove_next == 2 we obviously can't
942 * reach this path.
943 *
944 * If remove_next == 3 we can't reach this
945 * path because pre-swap() next is always not
946 * NULL. pre-swap() "next" is not being
947 * removed and its next->vm_end is not altered
948 * (and furthermore "end" already matches
949 * next->vm_end in remove_next == 3).
950 *
951 * We reach this only in the remove_next == 1
952 * case if the "next" vma that was removed was
953 * the highest vma of the mm. However in such
954 * case next->vm_end == "end" and the extended
955 * "vma" has vma->vm_end == next->vm_end so
956 * mm->highest_vm_end doesn't need any update
957 * in remove_next == 1 case.
958 */
Hugh Dickins1be71072017-06-19 04:03:24 -0700959 VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
Andrea Arcangelifb8c41e2016-10-07 17:01:25 -0700960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +0530962 if (insert && file)
Ingo Molnar7b2d81d2012-02-17 09:27:41 +0100963 uprobe_mmap(insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
965 validate_mm(mm);
Rik van Riel5beb4932010-03-05 13:42:07 -0800966
967 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968}
969
970/*
971 * If the vma has a ->close operation then the driver probably needs to release
972 * per-vma resources, so we don't attempt to merge those.
973 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974static inline int is_mergeable_vma(struct vm_area_struct *vma,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700975 struct file *file, unsigned long vm_flags,
976 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977{
Cyrill Gorcunov34228d42014-01-23 15:53:42 -0800978 /*
979 * VM_SOFTDIRTY should not prevent from VMA merging, if we
980 * match the flags but dirty bit -- the caller should mark
981 * merged VMA as dirty. If dirty bit won't be excluded from
982 * comparison, we increase pressue on the memory system forcing
983 * the kernel to generate new VMAs when old one could be
984 * extended instead.
985 */
986 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 return 0;
988 if (vma->vm_file != file)
989 return 0;
990 if (vma->vm_ops && vma->vm_ops->close)
991 return 0;
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700992 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
993 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 return 1;
995}
996
997static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
Shaohua Li965f55d2011-05-24 17:11:20 -0700998 struct anon_vma *anon_vma2,
999 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000{
Shaohua Li965f55d2011-05-24 17:11:20 -07001001 /*
1002 * The list_is_singular() test is to avoid merging VMA cloned from
1003 * parents. This can improve scalability caused by anon_vma lock.
1004 */
1005 if ((!anon_vma1 || !anon_vma2) && (!vma ||
1006 list_is_singular(&vma->anon_vma_chain)))
1007 return 1;
1008 return anon_vma1 == anon_vma2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009}
1010
1011/*
1012 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1013 * in front of (at a lower virtual address and file offset than) the vma.
1014 *
1015 * We cannot merge two vmas if they have differently assigned (non-NULL)
1016 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1017 *
1018 * We don't check here for the merged mmap wrapping around the end of pagecache
1019 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1020 * wrap, nor mmaps which cover the final page at index -1UL.
1021 */
1022static int
1023can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001024 struct anon_vma *anon_vma, struct file *file,
1025 pgoff_t vm_pgoff,
1026 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001028 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001029 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (vma->vm_pgoff == vm_pgoff)
1031 return 1;
1032 }
1033 return 0;
1034}
1035
1036/*
1037 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1038 * beyond (at a higher virtual address and file offset than) the vma.
1039 *
1040 * We cannot merge two vmas if they have differently assigned (non-NULL)
1041 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1042 */
1043static int
1044can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001045 struct anon_vma *anon_vma, struct file *file,
1046 pgoff_t vm_pgoff,
1047 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001049 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
Shaohua Li965f55d2011-05-24 17:11:20 -07001050 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 pgoff_t vm_pglen;
Libind6e93212013-07-03 15:01:26 -07001052 vm_pglen = vma_pages(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1054 return 1;
1055 }
1056 return 0;
1057}
1058
1059/*
1060 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1061 * whether that can be merged with its predecessor or its successor.
1062 * Or both (it neatly fills a hole).
1063 *
1064 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1065 * certain not to be mapped by the time vma_merge is called; but when
1066 * called for mprotect, it is certain to be already mapped (either at
1067 * an offset within prev, or at the start of next), and the flags of
1068 * this area are about to be changed to vm_flags - and the no-change
1069 * case has already been eliminated.
1070 *
1071 * The following mprotect cases have to be considered, where AAAA is
1072 * the area passed down from mprotect_fixup, never extending beyond one
1073 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1074 *
1075 * AAAA AAAA AAAA AAAA
1076 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1077 * cannot merge might become might become might become
1078 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1079 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001080 * mremap move: PPPPXXXXXXXX 8
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 * AAAA
1082 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1083 * might become case 1 below case 2 below case 3 below
1084 *
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001085 * It is important for case 8 that the the vma NNNN overlapping the
1086 * region AAAA is never going to extended over XXXX. Instead XXXX must
1087 * be extended in region AAAA and NNNN must be removed. This way in
1088 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1089 * rmap_locks, the properties of the merged vma will be already
1090 * correct for the whole merged range. Some of those properties like
1091 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1092 * be correct for the whole merged range immediately after the
1093 * rmap_locks are released. Otherwise if XXXX would be removed and
1094 * NNNN would be extended over the XXXX range, remove_migration_ptes
1095 * or other rmap walkers (if working on addresses beyond the "end"
1096 * parameter) may establish ptes with the wrong permissions of NNNN
1097 * instead of the right permissions of XXXX.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 */
1099struct vm_area_struct *vma_merge(struct mm_struct *mm,
1100 struct vm_area_struct *prev, unsigned long addr,
1101 unsigned long end, unsigned long vm_flags,
vishnu.pscc71aba2014-10-09 15:26:29 -07001102 struct anon_vma *anon_vma, struct file *file,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001103 pgoff_t pgoff, struct mempolicy *policy,
1104 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
1106 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1107 struct vm_area_struct *area, *next;
Rik van Riel5beb4932010-03-05 13:42:07 -08001108 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
1110 /*
1111 * We later require that vma->vm_flags == vm_flags,
1112 * so this tests vma->vm_flags & VM_SPECIAL, too.
1113 */
1114 if (vm_flags & VM_SPECIAL)
1115 return NULL;
1116
1117 if (prev)
1118 next = prev->vm_next;
1119 else
1120 next = mm->mmap;
1121 area = next;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001122 if (area && area->vm_end == end) /* cases 6, 7, 8 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 next = next->vm_next;
1124
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001125 /* verify some invariant that must be enforced by the caller */
1126 VM_WARN_ON(prev && addr <= prev->vm_start);
1127 VM_WARN_ON(area && end > area->vm_end);
1128 VM_WARN_ON(addr >= end);
1129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 /*
1131 * Can it merge with the predecessor?
1132 */
1133 if (prev && prev->vm_end == addr &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001134 mpol_equal(vma_policy(prev), policy) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 can_vma_merge_after(prev, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001136 anon_vma, file, pgoff,
1137 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 /*
1139 * OK, it can. Can we now merge in the successor as well?
1140 */
1141 if (next && end == next->vm_start &&
1142 mpol_equal(policy, vma_policy(next)) &&
1143 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001144 anon_vma, file,
1145 pgoff+pglen,
1146 vm_userfaultfd_ctx) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 is_mergeable_anon_vma(prev->anon_vma,
Shaohua Li965f55d2011-05-24 17:11:20 -07001148 next->anon_vma, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 /* cases 1, 6 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001150 err = __vma_adjust(prev, prev->vm_start,
1151 next->vm_end, prev->vm_pgoff, NULL,
1152 prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 } else /* cases 2, 5, 7 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001154 err = __vma_adjust(prev, prev->vm_start,
1155 end, prev->vm_pgoff, NULL, prev);
Rik van Riel5beb4932010-03-05 13:42:07 -08001156 if (err)
1157 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001158 khugepaged_enter_vma_merge(prev, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 return prev;
1160 }
1161
1162 /*
1163 * Can this new request be merged in front of next?
1164 */
1165 if (next && end == next->vm_start &&
vishnu.pscc71aba2014-10-09 15:26:29 -07001166 mpol_equal(policy, vma_policy(next)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 can_vma_merge_before(next, vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001168 anon_vma, file, pgoff+pglen,
1169 vm_userfaultfd_ctx)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 if (prev && addr < prev->vm_end) /* case 4 */
Andrea Arcangelie86f15e2016-10-07 17:01:28 -07001171 err = __vma_adjust(prev, prev->vm_start,
1172 addr, prev->vm_pgoff, NULL, next);
1173 else { /* cases 3, 8 */
1174 err = __vma_adjust(area, addr, next->vm_end,
1175 next->vm_pgoff - pglen, NULL, next);
1176 /*
1177 * In case 3 area is already equal to next and
1178 * this is a noop, but in case 8 "area" has
1179 * been removed and next was expanded over it.
1180 */
1181 area = next;
1182 }
Rik van Riel5beb4932010-03-05 13:42:07 -08001183 if (err)
1184 return NULL;
David Rientjes6d50e602014-10-29 14:50:31 -07001185 khugepaged_enter_vma_merge(area, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 return area;
1187 }
1188
1189 return NULL;
1190}
1191
1192/*
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001193 * Rough compatbility check to quickly see if it's even worth looking
1194 * at sharing an anon_vma.
1195 *
1196 * They need to have the same vm_file, and the flags can only differ
1197 * in things that mprotect may change.
1198 *
1199 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1200 * we can merge the two vma's. For example, we refuse to merge a vma if
1201 * there is a vm_ops->close() function, because that indicates that the
1202 * driver is doing some kind of reference counting. But that doesn't
1203 * really matter for the anon_vma sharing case.
1204 */
1205static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1206{
1207 return a->vm_end == b->vm_start &&
1208 mpol_equal(vma_policy(a), vma_policy(b)) &&
1209 a->vm_file == b->vm_file &&
Cyrill Gorcunov34228d42014-01-23 15:53:42 -08001210 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001211 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1212}
1213
1214/*
1215 * Do some basic sanity checking to see if we can re-use the anon_vma
1216 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1217 * the same as 'old', the other will be the new one that is trying
1218 * to share the anon_vma.
1219 *
1220 * NOTE! This runs with mm_sem held for reading, so it is possible that
1221 * the anon_vma of 'old' is concurrently in the process of being set up
1222 * by another page fault trying to merge _that_. But that's ok: if it
1223 * is being set up, that automatically means that it will be a singleton
1224 * acceptable for merging, so we can do all of this optimistically. But
Jason Low4db0c3c2015-04-15 16:14:08 -07001225 * we do that READ_ONCE() to make sure that we never re-load the pointer.
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001226 *
1227 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1228 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1229 * is to return an anon_vma that is "complex" due to having gone through
1230 * a fork).
1231 *
1232 * We also make sure that the two vma's are compatible (adjacent,
1233 * and with the same memory policies). That's all stable, even with just
1234 * a read lock on the mm_sem.
1235 */
1236static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1237{
1238 if (anon_vma_compatible(a, b)) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001239 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001240
1241 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1242 return anon_vma;
1243 }
1244 return NULL;
1245}
1246
1247/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1249 * neighbouring vmas for a suitable anon_vma, before it goes off
1250 * to allocate a new anon_vma. It checks because a repetitive
1251 * sequence of mprotects and faults may otherwise lead to distinct
1252 * anon_vmas being allocated, preventing vma merge in subsequent
1253 * mprotect.
1254 */
1255struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1256{
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001257 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 struct vm_area_struct *near;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
1260 near = vma->vm_next;
1261 if (!near)
1262 goto try_prev;
1263
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001264 anon_vma = reusable_anon_vma(near, vma, near);
1265 if (anon_vma)
1266 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267try_prev:
Linus Torvalds9be34c92011-06-16 00:35:09 -07001268 near = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 if (!near)
1270 goto none;
1271
Linus Torvaldsd0e9fe12010-04-10 10:36:19 -07001272 anon_vma = reusable_anon_vma(near, near, vma);
1273 if (anon_vma)
1274 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275none:
1276 /*
1277 * There's no absolute need to look only at touching neighbours:
1278 * we could search further afield for "compatible" anon_vmas.
1279 * But it would probably just be a waste of time searching,
1280 * or lead to too many vmas hanging off the same anon_vma.
1281 * We're trying to allow mprotect remerging later on,
1282 * not trying to minimize memory used for anon_vmas.
1283 */
1284 return NULL;
1285}
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287/*
Al Viro40401532012-02-13 03:58:52 +00001288 * If a hint addr is less than mmap_min_addr change hint to be as
1289 * low as possible but still greater than mmap_min_addr
1290 */
1291static inline unsigned long round_hint_to_min(unsigned long hint)
1292{
1293 hint &= PAGE_MASK;
1294 if (((void *)hint != NULL) &&
1295 (hint < mmap_min_addr))
1296 return PAGE_ALIGN(mmap_min_addr);
1297 return hint;
1298}
1299
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001300static inline int mlock_future_check(struct mm_struct *mm,
1301 unsigned long flags,
1302 unsigned long len)
1303{
1304 unsigned long locked, lock_limit;
1305
1306 /* mlock MCL_FUTURE? */
1307 if (flags & VM_LOCKED) {
1308 locked = len >> PAGE_SHIFT;
1309 locked += mm->locked_vm;
1310 lock_limit = rlimit(RLIMIT_MEMLOCK);
1311 lock_limit >>= PAGE_SHIFT;
1312 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1313 return -EAGAIN;
1314 }
1315 return 0;
1316}
1317
Al Viro40401532012-02-13 03:58:52 +00001318/*
Jianjun Kong27f5de72009-09-17 19:26:26 -07001319 * The caller must hold down_write(&current->mm->mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001321unsigned long do_mmap(struct file *file, unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 unsigned long len, unsigned long prot,
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001323 unsigned long flags, vm_flags_t vm_flags,
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001324 unsigned long pgoff, unsigned long *populate,
1325 struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
vishnu.pscc71aba2014-10-09 15:26:29 -07001327 struct mm_struct *mm = current->mm;
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001328 int pkey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Michel Lespinasse41badc12013-02-22 16:32:47 -08001330 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001331
Piotr Kwapulinskie37609b2015-06-24 16:58:39 -07001332 if (!len)
1333 return -EINVAL;
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 /*
1336 * Does the application expect PROT_READ to imply PROT_EXEC?
1337 *
1338 * (the exception is when the underlying filesystem is noexec
1339 * mounted, in which case we dont add PROT_EXEC.)
1340 */
1341 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Eric W. Biederman90f85722015-06-29 14:42:03 -05001342 if (!(file && path_noexec(&file->f_path)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 prot |= PROT_EXEC;
1344
Eric Paris7cd94142007-11-26 18:47:40 -05001345 if (!(flags & MAP_FIXED))
1346 addr = round_hint_to_min(addr);
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 /* Careful about overflows.. */
1349 len = PAGE_ALIGN(len);
Al Viro9206de92009-12-03 15:23:11 -05001350 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return -ENOMEM;
1352
1353 /* offset overflow? */
1354 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
vishnu.pscc71aba2014-10-09 15:26:29 -07001355 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
1357 /* Too many mappings? */
1358 if (mm->map_count > sysctl_max_map_count)
1359 return -ENOMEM;
1360
1361 /* Obtain the address to map to. we verify (or select) it and ensure
1362 * that it represents a valid section of the address space.
1363 */
1364 addr = get_unmapped_area(file, addr, len, pgoff, flags);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001365 if (offset_in_page(addr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 return addr;
1367
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001368 if (prot == PROT_EXEC) {
1369 pkey = execute_only_pkey(mm);
1370 if (pkey < 0)
1371 pkey = 0;
1372 }
1373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 /* Do simple checking here so the lower-level routines won't have
1375 * to. we assume access permissions have been handled by the open
1376 * of the memory object, so we don't do any here.
1377 */
Dave Hansen62b5f7d2016-02-12 13:02:40 -08001378 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1380
Huang Shijiecdf7b342009-09-21 17:03:36 -07001381 if (flags & MAP_LOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 if (!can_do_mlock())
1383 return -EPERM;
Rik van Rielba470de2008-10-18 20:26:50 -07001384
Davidlohr Bueso363ee172014-01-21 15:49:15 -08001385 if (mlock_future_check(mm, vm_flags, len))
1386 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (file) {
Oleg Nesterov077bf222013-09-11 14:20:19 -07001389 struct inode *inode = file_inode(file);
Dan Williams1c972592017-11-01 16:36:30 +01001390 unsigned long flags_mask;
1391
1392 flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
Oleg Nesterov077bf222013-09-11 14:20:19 -07001393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 switch (flags & MAP_TYPE) {
1395 case MAP_SHARED:
Dan Williams1c972592017-11-01 16:36:30 +01001396 /*
1397 * Force use of MAP_SHARED_VALIDATE with non-legacy
1398 * flags. E.g. MAP_SYNC is dangerous to use with
1399 * MAP_SHARED as you don't know which consistency model
1400 * you will get. We silently ignore unsupported flags
1401 * with MAP_SHARED to preserve backward compatibility.
1402 */
1403 flags &= LEGACY_MAP_MASK;
1404 /* fall through */
1405 case MAP_SHARED_VALIDATE:
1406 if (flags & ~flags_mask)
1407 return -EOPNOTSUPP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1409 return -EACCES;
1410
1411 /*
1412 * Make sure we don't allow writing to an append-only
1413 * file..
1414 */
1415 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1416 return -EACCES;
1417
1418 /*
1419 * Make sure there are no mandatory locks on the file.
1420 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001421 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 return -EAGAIN;
1423
1424 vm_flags |= VM_SHARED | VM_MAYSHARE;
1425 if (!(file->f_mode & FMODE_WRITE))
1426 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1427
1428 /* fall through */
1429 case MAP_PRIVATE:
1430 if (!(file->f_mode & FMODE_READ))
1431 return -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -05001432 if (path_noexec(&file->f_path)) {
Linus Torvalds80c56062006-10-15 14:09:55 -07001433 if (vm_flags & VM_EXEC)
1434 return -EPERM;
1435 vm_flags &= ~VM_MAYEXEC;
1436 }
Linus Torvalds80c56062006-10-15 14:09:55 -07001437
Al Viro72c2d532013-09-22 16:27:52 -04001438 if (!file->f_op->mmap)
Linus Torvalds80c56062006-10-15 14:09:55 -07001439 return -ENODEV;
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001440 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1441 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 break;
1443
1444 default:
1445 return -EINVAL;
1446 }
1447 } else {
1448 switch (flags & MAP_TYPE) {
1449 case MAP_SHARED:
Oleg Nesterovb2c56e42013-09-11 14:20:18 -07001450 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1451 return -EINVAL;
Tejun Heoce363942008-09-03 16:09:47 +02001452 /*
1453 * Ignore pgoff.
1454 */
1455 pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 vm_flags |= VM_SHARED | VM_MAYSHARE;
1457 break;
1458 case MAP_PRIVATE:
1459 /*
1460 * Set pgoff according to addr for anon_vma.
1461 */
1462 pgoff = addr >> PAGE_SHIFT;
1463 break;
1464 default:
1465 return -EINVAL;
1466 }
1467 }
1468
Michel Lespinassec22c0d62013-02-22 16:32:43 -08001469 /*
1470 * Set 'VM_NORESERVE' if we should not account for the
1471 * memory use of this mapping.
1472 */
1473 if (flags & MAP_NORESERVE) {
1474 /* We honor MAP_NORESERVE if allowed to overcommit */
1475 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1476 vm_flags |= VM_NORESERVE;
1477
1478 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1479 if (file && is_file_hugepages(file))
1480 vm_flags |= VM_NORESERVE;
1481 }
1482
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001483 addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -07001484 if (!IS_ERR_VALUE(addr) &&
1485 ((vm_flags & VM_LOCKED) ||
1486 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
Michel Lespinasse41badc12013-02-22 16:32:47 -08001487 *populate = len;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001488 return addr;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001489}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001490
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001491SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1492 unsigned long, prot, unsigned long, flags,
1493 unsigned long, fd, unsigned long, pgoff)
1494{
1495 struct file *file = NULL;
Chen Gang1e3ee142015-11-05 18:48:35 -08001496 unsigned long retval;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001497
1498 if (!(flags & MAP_ANONYMOUS)) {
Al Viro120a7952010-10-30 02:54:44 -04001499 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001500 file = fget(fd);
1501 if (!file)
Chen Gang1e3ee142015-11-05 18:48:35 -08001502 return -EBADF;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001503 if (is_file_hugepages(file))
1504 len = ALIGN(len, huge_page_size(hstate_file(file)));
Jörn Engel493af572013-07-08 16:00:26 -07001505 retval = -EINVAL;
1506 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1507 goto out_fput;
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001508 } else if (flags & MAP_HUGETLB) {
1509 struct user_struct *user = NULL;
Andrew Mortonc103a4d2013-07-08 16:01:08 -07001510 struct hstate *hs;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001511
Anshuman Khandual20ac2892017-05-03 14:55:00 -07001512 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +08001513 if (!hs)
1514 return -EINVAL;
1515
1516 len = ALIGN(len, huge_page_size(hs));
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001517 /*
1518 * VM_NORESERVE is used because the reservations will be
1519 * taken when vm_ops->mmap() is called
1520 * A dummy user value is used because we are not locking
1521 * memory so no accounting is necessary
1522 */
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001523 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
Andi Kleen42d73952012-12-11 16:01:34 -08001524 VM_NORESERVE,
1525 &user, HUGETLB_ANONHUGE_INODE,
1526 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001527 if (IS_ERR(file))
1528 return PTR_ERR(file);
1529 }
1530
1531 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1532
Michal Hocko9fbeb5a2016-05-23 16:25:30 -07001533 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Jörn Engel493af572013-07-08 16:00:26 -07001534out_fput:
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001535 if (file)
1536 fput(file);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001537 return retval;
1538}
1539
Christoph Hellwiga4679372010-03-10 15:21:15 -08001540#ifdef __ARCH_WANT_SYS_OLD_MMAP
1541struct mmap_arg_struct {
1542 unsigned long addr;
1543 unsigned long len;
1544 unsigned long prot;
1545 unsigned long flags;
1546 unsigned long fd;
1547 unsigned long offset;
1548};
1549
1550SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1551{
1552 struct mmap_arg_struct a;
1553
1554 if (copy_from_user(&a, arg, sizeof(a)))
1555 return -EFAULT;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08001556 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001557 return -EINVAL;
1558
1559 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1560 a.offset >> PAGE_SHIFT);
1561}
1562#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1563
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001564/*
1565 * Some shared mappigns will want the pages marked read-only
1566 * to track write events. If so, we'll downgrade vm_page_prot
1567 * to the private version (using protection_map[] without the
1568 * VM_SHARED bit).
1569 */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001570int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001571{
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001572 vm_flags_t vm_flags = vma->vm_flags;
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001573 const struct vm_operations_struct *vm_ops = vma->vm_ops;
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001574
1575 /* If it was private or non-writable, the write bit is already clear */
1576 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1577 return 0;
1578
1579 /* The backer wishes to know when pages are first written to? */
Kirill A. Shutemov8a044462015-09-22 14:59:12 -07001580 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001581 return 1;
1582
Peter Feiner64e45502014-10-13 15:55:46 -07001583 /* The open routine did something to the protections that pgprot_modify
1584 * won't preserve? */
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -07001585 if (pgprot_val(vm_page_prot) !=
1586 pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001587 return 0;
1588
Peter Feiner64e45502014-10-13 15:55:46 -07001589 /* Do we need to track softdirty? */
1590 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1591 return 1;
1592
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001593 /* Specialty mapping? */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001594 if (vm_flags & VM_PFNMAP)
Alexey Dobriyan4e950f62007-07-30 02:36:13 +04001595 return 0;
1596
1597 /* Can the mapping track the dirty pages? */
1598 return vma->vm_file && vma->vm_file->f_mapping &&
1599 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1600}
1601
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001602/*
1603 * We account for memory if it's a private writeable mapping,
Mel Gorman5a6fe122009-02-10 14:02:27 +00001604 * not hugepages and VM_NORESERVE wasn't set.
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001605 */
KOSAKI Motohiroca16d142011-05-26 19:16:19 +09001606static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001607{
Mel Gorman5a6fe122009-02-10 14:02:27 +00001608 /*
1609 * hugetlb has its own accounting separate from the core VM
1610 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1611 */
1612 if (file && is_file_hugepages(file))
1613 return 0;
1614
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001615 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1616}
1617
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001618unsigned long mmap_region(struct file *file, unsigned long addr,
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001619 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
1620 struct list_head *uf)
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001621{
1622 struct mm_struct *mm = current->mm;
1623 struct vm_area_struct *vma, *prev;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001624 int error;
1625 struct rb_node **rb_link, *rb_parent;
1626 unsigned long charged = 0;
Miklos Szeredi0165ab42007-07-15 23:38:26 -07001627
Cyril Hrubise8420a82013-04-29 15:08:33 -07001628 /* Check against address space limit. */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001629 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
Cyril Hrubise8420a82013-04-29 15:08:33 -07001630 unsigned long nr_pages;
1631
1632 /*
1633 * MAP_FIXED may remove pages of mappings that intersects with
1634 * requested mapping. Account for the pages it would unmap.
1635 */
Cyril Hrubise8420a82013-04-29 15:08:33 -07001636 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1637
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001638 if (!may_expand_vm(mm, vm_flags,
1639 (len >> PAGE_SHIFT) - nr_pages))
Cyril Hrubise8420a82013-04-29 15:08:33 -07001640 return -ENOMEM;
1641 }
1642
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 /* Clear old maps */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07001644 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1645 &rb_parent)) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001646 if (do_munmap(mm, addr, len, uf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
1649
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001650 /*
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001651 * Private writable mapping: check memory availability
1652 */
Mel Gorman5a6fe122009-02-10 14:02:27 +00001653 if (accountable_mapping(file, vm_flags)) {
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001654 charged = len >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +00001655 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvaldsfc8744a2009-01-31 15:08:56 -08001656 return -ENOMEM;
1657 vm_flags |= VM_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 }
1659
1660 /*
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001661 * Can we just expand an old mapping?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 */
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07001663 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1664 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001665 if (vma)
1666 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
1668 /*
1669 * Determine the object being mapped and call the appropriate
1670 * specific mapper. the address has already been validated, but
1671 * not unmapped, but the maps are removed from the list.
1672 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08001673 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 if (!vma) {
1675 error = -ENOMEM;
1676 goto unacct_error;
1677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
1679 vma->vm_mm = mm;
1680 vma->vm_start = addr;
1681 vma->vm_end = addr + len;
1682 vma->vm_flags = vm_flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07001683 vma->vm_page_prot = vm_get_page_prot(vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 vma->vm_pgoff = pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001685 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
1687 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 if (vm_flags & VM_DENYWRITE) {
1689 error = deny_write_access(file);
1690 if (error)
1691 goto free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 }
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001693 if (vm_flags & VM_SHARED) {
1694 error = mapping_map_writable(file->f_mapping);
1695 if (error)
1696 goto allow_write_and_free_vma;
1697 }
1698
1699 /* ->mmap() can change vma->vm_file, but must guarantee that
1700 * vma_link() below can deny write-access if VM_DENYWRITE is set
1701 * and map writably if VM_SHARED is set. This usually means the
1702 * new file must not have been exposed to user-space, yet.
1703 */
Al Virocb0942b2012-08-27 14:48:26 -04001704 vma->vm_file = get_file(file);
Miklos Szeredif74ac012017-02-20 16:51:23 +01001705 error = call_mmap(file, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 if (error)
1707 goto unmap_and_free_vma;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001708
1709 /* Can addr have changed??
1710 *
1711 * Answer: Yes, several device drivers can do it in their
1712 * f_op->mmap method. -DaveM
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001713 * Bug: If addr is changed, prev, rb_link, rb_parent should
1714 * be updated for vma_link()
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001715 */
Joonsoo Kim2897b4d2012-12-12 13:51:53 -08001716 WARN_ON_ONCE(addr != vma->vm_start);
1717
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001718 addr = vma->vm_start;
Huang Shijief8dbf0a72009-09-21 17:03:41 -07001719 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 } else if (vm_flags & VM_SHARED) {
1721 error = shmem_zero_setup(vma);
1722 if (error)
1723 goto free_vma;
1724 }
1725
Linus Torvaldsde33c8d2009-01-29 17:46:42 -08001726 vma_link(mm, vma, prev, rb_link, rb_parent);
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001727 /* Once vma denies write, undo our temporary denial count */
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001728 if (file) {
1729 if (vm_flags & VM_SHARED)
1730 mapping_unmap_writable(file->f_mapping);
1731 if (vm_flags & VM_DENYWRITE)
1732 allow_write_access(file);
1733 }
Oleg Nesterove8686772013-09-11 14:20:20 -07001734 file = vma->vm_file;
Oleg Nesterov4d3d5b42008-04-28 02:12:10 -07001735out:
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001736 perf_event_mmap(vma);
Peter Zijlstra0a4a9392009-03-30 19:07:05 +02001737
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08001738 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 if (vm_flags & VM_LOCKED) {
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001740 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1741 vma == get_gate_vma(current->mm)))
KOSAKI Motohiro06f9d8c2010-03-05 13:41:43 -08001742 mm->locked_vm += (len >> PAGE_SHIFT);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001743 else
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001744 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001745 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301746
Oleg Nesterovc7a3a882012-08-19 19:10:42 +02001747 if (file)
1748 uprobe_mmap(vma);
Srikar Dronamraju2b144492012-02-09 14:56:42 +05301749
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07001750 /*
1751 * New (or expanded) vma always get soft dirty status.
1752 * Otherwise user-space soft-dirty page tracker won't
1753 * be able to distinguish situation when vma area unmapped,
1754 * then new mapped in-place (which must be aimed as
1755 * a completely new data area).
1756 */
1757 vma->vm_flags |= VM_SOFTDIRTY;
1758
Peter Feiner64e45502014-10-13 15:55:46 -07001759 vma_set_page_prot(vma);
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 return addr;
1762
1763unmap_and_free_vma:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 vma->vm_file = NULL;
1765 fput(file);
1766
1767 /* Undo any partial mapping done by a device driver. */
Hugh Dickinse0da3822005-04-19 13:29:15 -07001768 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1769 charged = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -07001770 if (vm_flags & VM_SHARED)
1771 mapping_unmap_writable(file->f_mapping);
1772allow_write_and_free_vma:
1773 if (vm_flags & VM_DENYWRITE)
1774 allow_write_access(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775free_vma:
1776 kmem_cache_free(vm_area_cachep, vma);
1777unacct_error:
1778 if (charged)
1779 vm_unacct_memory(charged);
1780 return error;
1781}
1782
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001783unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1784{
1785 /*
1786 * We implement the search by looking for an rbtree node that
1787 * immediately follows a suitable gap. That is,
1788 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1789 * - gap_end = vma->vm_start >= info->low_limit + length;
1790 * - gap_end - gap_start >= length
1791 */
1792
1793 struct mm_struct *mm = current->mm;
1794 struct vm_area_struct *vma;
1795 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1796
1797 /* Adjust search length to account for worst case alignment overhead */
1798 length = info->length + info->align_mask;
1799 if (length < info->length)
1800 return -ENOMEM;
1801
1802 /* Adjust search limits by the desired length */
1803 if (info->high_limit < length)
1804 return -ENOMEM;
1805 high_limit = info->high_limit - length;
1806
1807 if (info->low_limit > high_limit)
1808 return -ENOMEM;
1809 low_limit = info->low_limit + length;
1810
1811 /* Check if rbtree root looks promising */
1812 if (RB_EMPTY_ROOT(&mm->mm_rb))
1813 goto check_highest;
1814 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1815 if (vma->rb_subtree_gap < length)
1816 goto check_highest;
1817
1818 while (true) {
1819 /* Visit left subtree if it looks promising */
Hugh Dickins1be71072017-06-19 04:03:24 -07001820 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001821 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1822 struct vm_area_struct *left =
1823 rb_entry(vma->vm_rb.rb_left,
1824 struct vm_area_struct, vm_rb);
1825 if (left->rb_subtree_gap >= length) {
1826 vma = left;
1827 continue;
1828 }
1829 }
1830
Hugh Dickins1be71072017-06-19 04:03:24 -07001831 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001832check_current:
1833 /* Check if current node has a suitable gap */
1834 if (gap_start > high_limit)
1835 return -ENOMEM;
Hugh Dickinsf4cb7672017-06-20 02:10:44 -07001836 if (gap_end >= low_limit &&
1837 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001838 goto found;
1839
1840 /* Visit right subtree if it looks promising */
1841 if (vma->vm_rb.rb_right) {
1842 struct vm_area_struct *right =
1843 rb_entry(vma->vm_rb.rb_right,
1844 struct vm_area_struct, vm_rb);
1845 if (right->rb_subtree_gap >= length) {
1846 vma = right;
1847 continue;
1848 }
1849 }
1850
1851 /* Go back up the rbtree to find next candidate node */
1852 while (true) {
1853 struct rb_node *prev = &vma->vm_rb;
1854 if (!rb_parent(prev))
1855 goto check_highest;
1856 vma = rb_entry(rb_parent(prev),
1857 struct vm_area_struct, vm_rb);
1858 if (prev == vma->vm_rb.rb_left) {
Hugh Dickins1be71072017-06-19 04:03:24 -07001859 gap_start = vm_end_gap(vma->vm_prev);
1860 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001861 goto check_current;
1862 }
1863 }
1864 }
1865
1866check_highest:
1867 /* Check highest gap, which does not precede any rbtree node */
1868 gap_start = mm->highest_vm_end;
1869 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1870 if (gap_start > high_limit)
1871 return -ENOMEM;
1872
1873found:
1874 /* We found a suitable gap. Clip it with the original low_limit. */
1875 if (gap_start < info->low_limit)
1876 gap_start = info->low_limit;
1877
1878 /* Adjust gap address to the desired alignment */
1879 gap_start += (info->align_offset - gap_start) & info->align_mask;
1880
1881 VM_BUG_ON(gap_start + info->length > info->high_limit);
1882 VM_BUG_ON(gap_start + info->length > gap_end);
1883 return gap_start;
1884}
1885
1886unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1887{
1888 struct mm_struct *mm = current->mm;
1889 struct vm_area_struct *vma;
1890 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1891
1892 /* Adjust search length to account for worst case alignment overhead */
1893 length = info->length + info->align_mask;
1894 if (length < info->length)
1895 return -ENOMEM;
1896
1897 /*
1898 * Adjust search limits by the desired length.
1899 * See implementation comment at top of unmapped_area().
1900 */
1901 gap_end = info->high_limit;
1902 if (gap_end < length)
1903 return -ENOMEM;
1904 high_limit = gap_end - length;
1905
1906 if (info->low_limit > high_limit)
1907 return -ENOMEM;
1908 low_limit = info->low_limit + length;
1909
1910 /* Check highest gap, which does not precede any rbtree node */
1911 gap_start = mm->highest_vm_end;
1912 if (gap_start <= high_limit)
1913 goto found_highest;
1914
1915 /* Check if rbtree root looks promising */
1916 if (RB_EMPTY_ROOT(&mm->mm_rb))
1917 return -ENOMEM;
1918 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1919 if (vma->rb_subtree_gap < length)
1920 return -ENOMEM;
1921
1922 while (true) {
1923 /* Visit right subtree if it looks promising */
Hugh Dickins1be71072017-06-19 04:03:24 -07001924 gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001925 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1926 struct vm_area_struct *right =
1927 rb_entry(vma->vm_rb.rb_right,
1928 struct vm_area_struct, vm_rb);
1929 if (right->rb_subtree_gap >= length) {
1930 vma = right;
1931 continue;
1932 }
1933 }
1934
1935check_current:
1936 /* Check if current node has a suitable gap */
Hugh Dickins1be71072017-06-19 04:03:24 -07001937 gap_end = vm_start_gap(vma);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001938 if (gap_end < low_limit)
1939 return -ENOMEM;
Hugh Dickinsf4cb7672017-06-20 02:10:44 -07001940 if (gap_start <= high_limit &&
1941 gap_end > gap_start && gap_end - gap_start >= length)
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001942 goto found;
1943
1944 /* Visit left subtree if it looks promising */
1945 if (vma->vm_rb.rb_left) {
1946 struct vm_area_struct *left =
1947 rb_entry(vma->vm_rb.rb_left,
1948 struct vm_area_struct, vm_rb);
1949 if (left->rb_subtree_gap >= length) {
1950 vma = left;
1951 continue;
1952 }
1953 }
1954
1955 /* Go back up the rbtree to find next candidate node */
1956 while (true) {
1957 struct rb_node *prev = &vma->vm_rb;
1958 if (!rb_parent(prev))
1959 return -ENOMEM;
1960 vma = rb_entry(rb_parent(prev),
1961 struct vm_area_struct, vm_rb);
1962 if (prev == vma->vm_rb.rb_right) {
1963 gap_start = vma->vm_prev ?
Hugh Dickins1be71072017-06-19 04:03:24 -07001964 vm_end_gap(vma->vm_prev) : 0;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08001965 goto check_current;
1966 }
1967 }
1968 }
1969
1970found:
1971 /* We found a suitable gap. Clip it with the original high_limit. */
1972 if (gap_end > info->high_limit)
1973 gap_end = info->high_limit;
1974
1975found_highest:
1976 /* Compute highest gap address at the desired alignment */
1977 gap_end -= info->length;
1978 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1979
1980 VM_BUG_ON(gap_end < info->low_limit);
1981 VM_BUG_ON(gap_end < gap_start);
1982 return gap_end;
1983}
1984
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985/* Get an address range which is currently unmapped.
1986 * For shmat() with addr=0.
1987 *
1988 * Ugly calling convention alert:
1989 * Return value with the low bits set means error value,
1990 * ie
1991 * if (ret & ~PAGE_MASK)
1992 * error = ret;
1993 *
1994 * This function "knows" that -ENOMEM has the bits set.
1995 */
1996#ifndef HAVE_ARCH_UNMAPPED_AREA
1997unsigned long
1998arch_get_unmapped_area(struct file *filp, unsigned long addr,
1999 unsigned long len, unsigned long pgoff, unsigned long flags)
2000{
2001 struct mm_struct *mm = current->mm;
Hugh Dickins1be71072017-06-19 04:03:24 -07002002 struct vm_area_struct *vma, *prev;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002003 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002005 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 return -ENOMEM;
2007
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002008 if (flags & MAP_FIXED)
2009 return addr;
2010
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 if (addr) {
2012 addr = PAGE_ALIGN(addr);
Hugh Dickins1be71072017-06-19 04:03:24 -07002013 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002014 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickins1be71072017-06-19 04:03:24 -07002015 (!vma || addr + len <= vm_start_gap(vma)) &&
2016 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 return addr;
2018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002020 info.flags = 0;
2021 info.length = len;
Heiko Carstens4e99b022013-11-12 15:07:54 -08002022 info.low_limit = mm->mmap_base;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002023 info.high_limit = TASK_SIZE;
2024 info.align_mask = 0;
2025 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026}
vishnu.pscc71aba2014-10-09 15:26:29 -07002027#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029/*
2030 * This mmap-allocator allocates new areas top-down from below the
2031 * stack's low limit (the base):
2032 */
2033#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2034unsigned long
2035arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
2036 const unsigned long len, const unsigned long pgoff,
2037 const unsigned long flags)
2038{
Hugh Dickins1be71072017-06-19 04:03:24 -07002039 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 struct mm_struct *mm = current->mm;
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002041 unsigned long addr = addr0;
2042 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044 /* requested length too big for entire address space */
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002045 if (len > TASK_SIZE - mmap_min_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 return -ENOMEM;
2047
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002048 if (flags & MAP_FIXED)
2049 return addr;
2050
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 /* requesting a specific address */
2052 if (addr) {
2053 addr = PAGE_ALIGN(addr);
Hugh Dickins1be71072017-06-19 04:03:24 -07002054 vma = find_vma_prev(mm, addr, &prev);
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002055 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
Hugh Dickins1be71072017-06-19 04:03:24 -07002056 (!vma || addr + len <= vm_start_gap(vma)) &&
2057 (!prev || addr >= vm_end_gap(prev)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 return addr;
2059 }
2060
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002061 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2062 info.length = len;
Akira Takeuchi2afc7452013-11-12 15:08:21 -08002063 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002064 info.high_limit = mm->mmap_base;
2065 info.align_mask = 0;
2066 addr = vm_unmapped_area(&info);
Xiao Guangrongb716ad92012-03-21 16:33:56 -07002067
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 /*
2069 * A failed mmap() very likely causes application failure,
2070 * so fall back to the bottom-up function here. This scenario
2071 * can happen with large stack limits and large mmap()
2072 * allocations.
2073 */
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002074 if (offset_in_page(addr)) {
Michel Lespinassedb4fbfb2012-12-11 16:01:49 -08002075 VM_BUG_ON(addr != -ENOMEM);
2076 info.flags = 0;
2077 info.low_limit = TASK_UNMAPPED_BASE;
2078 info.high_limit = TASK_SIZE;
2079 addr = vm_unmapped_area(&info);
2080 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
2082 return addr;
2083}
2084#endif
2085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086unsigned long
2087get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2088 unsigned long pgoff, unsigned long flags)
2089{
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002090 unsigned long (*get_area)(struct file *, unsigned long,
2091 unsigned long, unsigned long, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
Al Viro9206de92009-12-03 15:23:11 -05002093 unsigned long error = arch_mmap_check(addr, len, flags);
2094 if (error)
2095 return error;
2096
2097 /* Careful about overflows.. */
2098 if (len > TASK_SIZE)
2099 return -ENOMEM;
2100
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002101 get_area = current->mm->get_unmapped_area;
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002102 if (file) {
2103 if (file->f_op->get_unmapped_area)
2104 get_area = file->f_op->get_unmapped_area;
2105 } else if (flags & MAP_SHARED) {
2106 /*
2107 * mmap_region() will call shmem_zero_setup() to create a file,
2108 * so use shmem's get_unmapped_area in case it can be huge.
2109 * do_mmap_pgoff() will clear pgoff, so match alignment.
2110 */
2111 pgoff = 0;
2112 get_area = shmem_get_unmapped_area;
2113 }
2114
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002115 addr = get_area(file, addr, len, pgoff, flags);
2116 if (IS_ERR_VALUE(addr))
2117 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002119 if (addr > TASK_SIZE - len)
2120 return -ENOMEM;
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002121 if (offset_in_page(addr))
Linus Torvalds07ab67c2005-05-19 22:43:37 -07002122 return -EINVAL;
Benjamin Herrenschmidt06abdfb2007-05-06 14:50:13 -07002123
Al Viro9ac4ed42012-05-30 17:13:15 -04002124 error = security_mmap_addr(addr);
2125 return error ? error : addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126}
2127
2128EXPORT_SYMBOL(get_unmapped_area);
2129
2130/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
ZhenwenXu48aae422009-01-06 14:40:21 -08002131struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002133 struct rb_node *rb_node;
2134 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002136 /* Check the cache first. */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002137 vma = vmacache_find(mm, addr);
2138 if (likely(vma))
2139 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002141 rb_node = mm->mm_rb.rb_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002143 while (rb_node) {
2144 struct vm_area_struct *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002146 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
Rajman Mekaco841e31e2012-05-29 15:06:21 -07002147
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002148 if (tmp->vm_end > addr) {
2149 vma = tmp;
2150 if (tmp->vm_start <= addr)
2151 break;
2152 rb_node = rb_node->rb_left;
2153 } else
2154 rb_node = rb_node->rb_right;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 }
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002156
2157 if (vma)
2158 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 return vma;
2160}
2161
2162EXPORT_SYMBOL(find_vma);
2163
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002164/*
2165 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002166 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167struct vm_area_struct *
2168find_vma_prev(struct mm_struct *mm, unsigned long addr,
2169 struct vm_area_struct **pprev)
2170{
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002171 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002173 vma = find_vma(mm, addr);
Mikulas Patocka83cd9042012-03-04 19:52:03 -05002174 if (vma) {
2175 *pprev = vma->vm_prev;
2176 } else {
2177 struct rb_node *rb_node = mm->mm_rb.rb_node;
2178 *pprev = NULL;
2179 while (rb_node) {
2180 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2181 rb_node = rb_node->rb_right;
2182 }
2183 }
KOSAKI Motohiro6bd48372012-01-10 15:08:07 -08002184 return vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185}
2186
2187/*
2188 * Verify that the stack growth is acceptable and
2189 * update accounting. This is shared with both the
2190 * grow-up and grow-down cases.
2191 */
Hugh Dickins1be71072017-06-19 04:03:24 -07002192static int acct_stack_growth(struct vm_area_struct *vma,
2193 unsigned long size, unsigned long grow)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194{
2195 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins1be71072017-06-19 04:03:24 -07002196 unsigned long new_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
2198 /* address space limit tests */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002199 if (!may_expand_vm(mm, vma->vm_flags, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 return -ENOMEM;
2201
2202 /* Stack limit test */
Krzysztof Opasiak24c79d82017-07-10 15:50:03 -07002203 if (size > rlimit(RLIMIT_STACK))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 return -ENOMEM;
2205
2206 /* mlock limit tests */
2207 if (vma->vm_flags & VM_LOCKED) {
2208 unsigned long locked;
2209 unsigned long limit;
2210 locked = mm->locked_vm + grow;
Krzysztof Opasiak24c79d82017-07-10 15:50:03 -07002211 limit = rlimit(RLIMIT_MEMLOCK);
Jiri Slaby59e99e52010-03-05 13:41:44 -08002212 limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 if (locked > limit && !capable(CAP_IPC_LOCK))
2214 return -ENOMEM;
2215 }
2216
Adam Litke0d59a012007-01-30 14:35:39 -08002217 /* Check to ensure the stack will not grow into a hugetlb-only region */
2218 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2219 vma->vm_end - size;
2220 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2221 return -EFAULT;
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 /*
2224 * Overcommit.. This must be the final test, as it will
2225 * update security statistics.
2226 */
Hugh Dickins05fa1992009-04-16 21:58:12 +01002227 if (security_vm_enough_memory_mm(mm, grow))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 return -ENOMEM;
2229
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 return 0;
2231}
2232
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002233#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234/*
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002235 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2236 * vma is the last one with address > vma->vm_end. Have to extend vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 */
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002238int expand_upwards(struct vm_area_struct *vma, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239{
Oleg Nesterov09357812015-11-05 18:48:17 -08002240 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins1be71072017-06-19 04:03:24 -07002241 struct vm_area_struct *next;
2242 unsigned long gap_addr;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002243 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245 if (!(vma->vm_flags & VM_GROWSUP))
2246 return -EFAULT;
2247
Helge Dellerbd726c92017-06-19 17:34:05 +02002248 /* Guard against exceeding limits of the address space. */
Hugh Dickins1be71072017-06-19 04:03:24 -07002249 address &= PAGE_MASK;
Helge Deller37511fb2017-07-14 14:49:38 -07002250 if (address >= (TASK_SIZE & PAGE_MASK))
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002251 return -ENOMEM;
Helge Dellerbd726c92017-06-19 17:34:05 +02002252 address += PAGE_SIZE;
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002253
Hugh Dickins1be71072017-06-19 04:03:24 -07002254 /* Enforce stack_guard_gap */
2255 gap_addr = address + stack_guard_gap;
Helge Dellerbd726c92017-06-19 17:34:05 +02002256
2257 /* Guard against overflow */
2258 if (gap_addr < address || gap_addr > TASK_SIZE)
2259 gap_addr = TASK_SIZE;
2260
Hugh Dickins1be71072017-06-19 04:03:24 -07002261 next = vma->vm_next;
Michal Hocko561b5e02017-07-10 15:49:51 -07002262 if (next && next->vm_start < gap_addr &&
2263 (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Hugh Dickins1be71072017-06-19 04:03:24 -07002264 if (!(next->vm_flags & VM_GROWSUP))
2265 return -ENOMEM;
2266 /* Check that both stack segments have the same anon_vma? */
2267 }
2268
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002269 /* We must make sure the anon_vma is allocated. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 if (unlikely(anon_vma_prepare(vma)))
2271 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
2273 /*
2274 * vma->vm_start/vm_end cannot change under us because the caller
2275 * is required to hold the mmap_sem in read mode. We need the
2276 * anon_vma lock to serialize against concurrent expand_stacks.
2277 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002278 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
2280 /* Somebody else might have raced and expanded it already */
2281 if (address > vma->vm_end) {
2282 unsigned long size, grow;
2283
2284 size = address - vma->vm_start;
2285 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2286
Hugh Dickins42c36f62011-05-09 17:44:42 -07002287 error = -ENOMEM;
2288 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2289 error = acct_stack_growth(vma, size, grow);
2290 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002291 /*
2292 * vma_gap_update() doesn't support concurrent
2293 * updates, but we only hold a shared mmap_sem
2294 * lock here, so we need to protect against
2295 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002296 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002297 * we don't guarantee that all growable vmas
2298 * in a mm share the same root anon vma.
2299 * So, we reuse mm->page_table_lock to guard
2300 * against concurrent vma expansions.
2301 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002302 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002303 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002304 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002305 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002306 anon_vma_interval_tree_pre_update_vma(vma);
Hugh Dickins42c36f62011-05-09 17:44:42 -07002307 vma->vm_end = address;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002308 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002309 if (vma->vm_next)
2310 vma_gap_update(vma->vm_next);
2311 else
Hugh Dickins1be71072017-06-19 04:03:24 -07002312 mm->highest_vm_end = vm_end_gap(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002313 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002314
Hugh Dickins42c36f62011-05-09 17:44:42 -07002315 perf_event_mmap(vma);
2316 }
Eric B Munson3af9e852010-05-18 15:30:49 +01002317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002319 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002320 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002321 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 return error;
2323}
Hugh Dickins46dea3d2005-10-29 18:16:20 -07002324#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326/*
2327 * vma is the first one with address < vma->vm_start. Have to extend vma.
2328 */
Michal Hockod05f3162011-05-24 17:11:44 -07002329int expand_downwards(struct vm_area_struct *vma,
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002330 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331{
Oleg Nesterov09357812015-11-05 18:48:17 -08002332 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins1be71072017-06-19 04:03:24 -07002333 struct vm_area_struct *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 int error;
2335
Eric Paris88694772007-11-26 18:47:26 -05002336 address &= PAGE_MASK;
Al Viroe5467852012-05-30 13:30:51 -04002337 error = security_mmap_addr(address);
Eric Paris88694772007-11-26 18:47:26 -05002338 if (error)
2339 return error;
2340
Hugh Dickins1be71072017-06-19 04:03:24 -07002341 /* Enforce stack_guard_gap */
Hugh Dickins1be71072017-06-19 04:03:24 -07002342 prev = vma->vm_prev;
Oleg Nesterov32e4e6d2017-07-10 15:49:54 -07002343 /* Check that both stack segments have the same anon_vma? */
2344 if (prev && !(prev->vm_flags & VM_GROWSDOWN) &&
Michal Hocko561b5e02017-07-10 15:49:51 -07002345 (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
Oleg Nesterov32e4e6d2017-07-10 15:49:54 -07002346 if (address - prev->vm_end < stack_guard_gap)
Hugh Dickins1be71072017-06-19 04:03:24 -07002347 return -ENOMEM;
Hugh Dickins1be71072017-06-19 04:03:24 -07002348 }
2349
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002350 /* We must make sure the anon_vma is allocated. */
2351 if (unlikely(anon_vma_prepare(vma)))
2352 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
2354 /*
2355 * vma->vm_start/vm_end cannot change under us because the caller
2356 * is required to hold the mmap_sem in read mode. We need the
2357 * anon_vma lock to serialize against concurrent expand_stacks.
2358 */
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002359 anon_vma_lock_write(vma->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
2361 /* Somebody else might have raced and expanded it already */
2362 if (address < vma->vm_start) {
2363 unsigned long size, grow;
2364
2365 size = vma->vm_end - address;
2366 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2367
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002368 error = -ENOMEM;
2369 if (grow <= vma->vm_pgoff) {
2370 error = acct_stack_growth(vma, size, grow);
2371 if (!error) {
Michel Lespinasse41289972012-12-12 13:52:25 -08002372 /*
2373 * vma_gap_update() doesn't support concurrent
2374 * updates, but we only hold a shared mmap_sem
2375 * lock here, so we need to protect against
2376 * concurrent vma expansions.
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002377 * anon_vma_lock_write() doesn't help here, as
Michel Lespinasse41289972012-12-12 13:52:25 -08002378 * we don't guarantee that all growable vmas
2379 * in a mm share the same root anon vma.
2380 * So, we reuse mm->page_table_lock to guard
2381 * against concurrent vma expansions.
2382 */
Oleg Nesterov09357812015-11-05 18:48:17 -08002383 spin_lock(&mm->page_table_lock);
Oleg Nesterov87e88272015-11-05 18:48:14 -08002384 if (vma->vm_flags & VM_LOCKED)
Oleg Nesterov09357812015-11-05 18:48:17 -08002385 mm->locked_vm += grow;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002386 vm_stat_account(mm, vma->vm_flags, grow);
Michel Lespinassebf181b92012-10-08 16:31:39 -07002387 anon_vma_interval_tree_pre_update_vma(vma);
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002388 vma->vm_start = address;
2389 vma->vm_pgoff -= grow;
Michel Lespinassebf181b92012-10-08 16:31:39 -07002390 anon_vma_interval_tree_post_update_vma(vma);
Michel Lespinassed3737182012-12-11 16:01:38 -08002391 vma_gap_update(vma);
Oleg Nesterov09357812015-11-05 18:48:17 -08002392 spin_unlock(&mm->page_table_lock);
Michel Lespinasse41289972012-12-12 13:52:25 -08002393
Linus Torvaldsa626ca62011-04-13 08:07:28 -07002394 perf_event_mmap(vma);
2395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 }
2397 }
Konstantin Khlebnikov12352d32016-02-05 15:36:50 -08002398 anon_vma_unlock_write(vma->anon_vma);
David Rientjes6d50e602014-10-29 14:50:31 -07002399 khugepaged_enter_vma_merge(vma, vma->vm_flags);
Oleg Nesterov09357812015-11-05 18:48:17 -08002400 validate_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 return error;
2402}
2403
Hugh Dickins1be71072017-06-19 04:03:24 -07002404/* enforced gap between the expanding stack and other mappings. */
2405unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2406
2407static int __init cmdline_parse_stack_guard_gap(char *p)
2408{
2409 unsigned long val;
2410 char *endptr;
2411
2412 val = simple_strtoul(p, &endptr, 10);
2413 if (!*endptr)
2414 stack_guard_gap = val << PAGE_SHIFT;
2415
2416 return 0;
2417}
2418__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2419
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002420#ifdef CONFIG_STACK_GROWSUP
2421int expand_stack(struct vm_area_struct *vma, unsigned long address)
2422{
2423 return expand_upwards(vma, address);
2424}
2425
2426struct vm_area_struct *
2427find_extend_vma(struct mm_struct *mm, unsigned long addr)
2428{
2429 struct vm_area_struct *vma, *prev;
2430
2431 addr &= PAGE_MASK;
2432 vma = find_vma_prev(mm, addr, &prev);
2433 if (vma && (vma->vm_start <= addr))
2434 return vma;
Denys Vlasenko1c127182008-11-12 01:24:41 +01002435 if (!prev || expand_stack(prev, addr))
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002436 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002437 if (prev->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002438 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07002439 return prev;
2440}
2441#else
2442int expand_stack(struct vm_area_struct *vma, unsigned long address)
2443{
2444 return expand_downwards(vma, address);
2445}
2446
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447struct vm_area_struct *
vishnu.pscc71aba2014-10-09 15:26:29 -07002448find_extend_vma(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449{
vishnu.pscc71aba2014-10-09 15:26:29 -07002450 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 unsigned long start;
2452
2453 addr &= PAGE_MASK;
vishnu.pscc71aba2014-10-09 15:26:29 -07002454 vma = find_vma(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 if (!vma)
2456 return NULL;
2457 if (vma->vm_start <= addr)
2458 return vma;
2459 if (!(vma->vm_flags & VM_GROWSDOWN))
2460 return NULL;
2461 start = vma->vm_start;
2462 if (expand_stack(vma, addr))
2463 return NULL;
Michel Lespinassecea10a12013-02-22 16:32:44 -08002464 if (vma->vm_flags & VM_LOCKED)
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -07002465 populate_vma_page_range(vma, addr, start, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 return vma;
2467}
2468#endif
2469
Jesse Barnese1d6d012014-12-12 16:55:27 -08002470EXPORT_SYMBOL_GPL(find_extend_vma);
2471
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472/*
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002473 * Ok - we have the memory areas we should free on the vma list,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 * so release them, and do the vma updates.
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002475 *
2476 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002478static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002480 unsigned long nr_accounted = 0;
2481
Hugh Dickins365e9c872005-10-29 18:16:18 -07002482 /* Update high watermark before we lower total_vm */
2483 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 do {
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002485 long nrpages = vma_pages(vma);
2486
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002487 if (vma->vm_flags & VM_ACCOUNT)
2488 nr_accounted += nrpages;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002489 vm_stat_account(mm, vma->vm_flags, -nrpages);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07002490 vma = remove_vma(vma);
Hugh Dickins146425a2005-04-19 13:29:18 -07002491 } while (vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002492 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 validate_mm(mm);
2494}
2495
2496/*
2497 * Get rid of page table information in the indicated region.
2498 *
Paolo 'Blaisorblade' Giarrussof10df682005-09-21 09:55:37 -07002499 * Called with the mm semaphore held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 */
2501static void unmap_region(struct mm_struct *mm,
Hugh Dickinse0da3822005-04-19 13:29:15 -07002502 struct vm_area_struct *vma, struct vm_area_struct *prev,
2503 unsigned long start, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504{
vishnu.pscc71aba2014-10-09 15:26:29 -07002505 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002506 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
2508 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07002509 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07002510 update_hiwater_rss(mm);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07002511 unmap_vmas(&tlb, vma, start, end);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002512 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
Hugh Dickins6ee86302013-04-29 15:07:44 -07002513 next ? next->vm_start : USER_PGTABLES_CEILING);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002514 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515}
2516
2517/*
2518 * Create a list of vma's touched by the unmap, removing them from the mm's
2519 * vma list as we go..
2520 */
2521static void
2522detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2523 struct vm_area_struct *prev, unsigned long end)
2524{
2525 struct vm_area_struct **insertion_point;
2526 struct vm_area_struct *tail_vma = NULL;
2527
2528 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002529 vma->vm_prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 do {
Michel Lespinassed3737182012-12-11 16:01:38 -08002531 vma_rb_erase(vma, &mm->mm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 mm->map_count--;
2533 tail_vma = vma;
2534 vma = vma->vm_next;
2535 } while (vma && vma->vm_start < end);
2536 *insertion_point = vma;
Michel Lespinassed3737182012-12-11 16:01:38 -08002537 if (vma) {
Linus Torvalds297c5ee2010-08-20 16:24:55 -07002538 vma->vm_prev = prev;
Michel Lespinassed3737182012-12-11 16:01:38 -08002539 vma_gap_update(vma);
2540 } else
Hugh Dickins1be71072017-06-19 04:03:24 -07002541 mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 tail_vma->vm_next = NULL;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07002543
2544 /* Kill the cache */
2545 vmacache_invalidate(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546}
2547
2548/*
David Rientjesdef5efe2017-02-24 14:58:47 -08002549 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2550 * has already been checked or doesn't make sense to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 */
David Rientjesdef5efe2017-02-24 14:58:47 -08002552int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2553 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 struct vm_area_struct *new;
Chen Gange3975892015-09-08 15:03:38 -07002556 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
Andi Kleena5516432008-07-23 21:27:41 -07002558 if (is_vm_hugetlb_page(vma) && (addr &
2559 ~(huge_page_mask(hstate_vma(vma)))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 return -EINVAL;
2561
Christoph Lametere94b1762006-12-06 20:33:17 -08002562 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 if (!new)
Chen Gange3975892015-09-08 15:03:38 -07002564 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565
2566 /* most fields are the same, copy all, and then fixup */
2567 *new = *vma;
2568
Rik van Riel5beb4932010-03-05 13:42:07 -08002569 INIT_LIST_HEAD(&new->anon_vma_chain);
2570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 if (new_below)
2572 new->vm_end = addr;
2573 else {
2574 new->vm_start = addr;
2575 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2576 }
2577
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002578 err = vma_dup_policy(vma, new);
2579 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002580 goto out_free_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
Daniel Forrestc4ea95d2014-12-02 15:59:42 -08002582 err = anon_vma_clone(new, vma);
2583 if (err)
Rik van Riel5beb4932010-03-05 13:42:07 -08002584 goto out_free_mpol;
2585
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002586 if (new->vm_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 get_file(new->vm_file);
2588
2589 if (new->vm_ops && new->vm_ops->open)
2590 new->vm_ops->open(new);
2591
2592 if (new_below)
Rik van Riel5beb4932010-03-05 13:42:07 -08002593 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2595 else
Rik van Riel5beb4932010-03-05 13:42:07 -08002596 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597
Rik van Riel5beb4932010-03-05 13:42:07 -08002598 /* Success. */
2599 if (!err)
2600 return 0;
2601
2602 /* Clean everything up if vma_adjust failed. */
Rik van Riel58927532010-04-26 12:33:03 -04002603 if (new->vm_ops && new->vm_ops->close)
2604 new->vm_ops->close(new);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -07002605 if (new->vm_file)
Rik van Riel5beb4932010-03-05 13:42:07 -08002606 fput(new->vm_file);
Andrea Arcangeli2aeadc32010-09-22 13:05:12 -07002607 unlink_anon_vmas(new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002608 out_free_mpol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002609 mpol_put(vma_policy(new));
Rik van Riel5beb4932010-03-05 13:42:07 -08002610 out_free_vma:
2611 kmem_cache_free(vm_area_cachep, new);
Rik van Riel5beb4932010-03-05 13:42:07 -08002612 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613}
2614
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002615/*
2616 * Split a vma into two pieces at address 'addr', a new vma is allocated
2617 * either for the first part or the tail.
2618 */
2619int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2620 unsigned long addr, int new_below)
2621{
2622 if (mm->map_count >= sysctl_max_map_count)
2623 return -ENOMEM;
2624
2625 return __split_vma(mm, vma, addr, new_below);
2626}
2627
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628/* Munmap is split into 2 main parts -- this part which finds
2629 * what needs doing, and the areas themselves, which do the
2630 * work. This now handles partial unmappings.
2631 * Jeremy Fitzhardinge <jeremy@goop.org>
2632 */
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002633int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2634 struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635{
2636 unsigned long end;
Hugh Dickins146425a2005-04-19 13:29:18 -07002637 struct vm_area_struct *vma, *prev, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002639 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 return -EINVAL;
2641
vishnu.pscc71aba2014-10-09 15:26:29 -07002642 len = PAGE_ALIGN(len);
2643 if (len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 return -EINVAL;
2645
2646 /* Find the first overlapping VMA */
Linus Torvalds9be34c92011-06-16 00:35:09 -07002647 vma = find_vma(mm, start);
Hugh Dickins146425a2005-04-19 13:29:18 -07002648 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 return 0;
Linus Torvalds9be34c92011-06-16 00:35:09 -07002650 prev = vma->vm_prev;
Hugh Dickins146425a2005-04-19 13:29:18 -07002651 /* we have start < vma->vm_end */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
2653 /* if it doesn't overlap, we have nothing.. */
2654 end = start + len;
Hugh Dickins146425a2005-04-19 13:29:18 -07002655 if (vma->vm_start >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 return 0;
2657
2658 /*
2659 * If we need to split any vma, do it now to save pain later.
2660 *
2661 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2662 * unmapped vm_area_struct will remain in use: so lower split_vma
2663 * places tmp vma above, and higher split_vma places tmp vma below.
2664 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002665 if (start > vma->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002666 int error;
2667
2668 /*
2669 * Make sure that map_count on return from munmap() will
2670 * not exceed its limit; but let map_count go just above
2671 * its limit temporarily, to help free resources as expected.
2672 */
2673 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2674 return -ENOMEM;
2675
2676 error = __split_vma(mm, vma, start, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 if (error)
2678 return error;
Hugh Dickins146425a2005-04-19 13:29:18 -07002679 prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 }
2681
2682 /* Does it split the last one? */
2683 last = find_vma(mm, end);
2684 if (last && end > last->vm_start) {
KOSAKI Motohiro659ace52009-12-14 17:57:56 -08002685 int error = __split_vma(mm, last, end, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 if (error)
2687 return error;
2688 }
vishnu.pscc71aba2014-10-09 15:26:29 -07002689 vma = prev ? prev->vm_next : mm->mmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
Andrea Arcangeli2376dd72017-09-06 16:23:53 -07002691 if (unlikely(uf)) {
2692 /*
2693 * If userfaultfd_unmap_prep returns an error the vmas
2694 * will remain splitted, but userland will get a
2695 * highly unexpected error anyway. This is no
2696 * different than the case where the first of the two
2697 * __split_vma fails, but we don't undo the first
2698 * split, despite we could. This is unlikely enough
2699 * failure that it's not worth optimizing it for.
2700 */
2701 int error = userfaultfd_unmap_prep(vma, start, end, uf);
2702 if (error)
2703 return error;
2704 }
2705
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 /*
Rik van Rielba470de2008-10-18 20:26:50 -07002707 * unlock any mlock()ed ranges before detaching vmas
2708 */
2709 if (mm->locked_vm) {
2710 struct vm_area_struct *tmp = vma;
2711 while (tmp && tmp->vm_start < end) {
2712 if (tmp->vm_flags & VM_LOCKED) {
2713 mm->locked_vm -= vma_pages(tmp);
2714 munlock_vma_pages_all(tmp);
2715 }
2716 tmp = tmp->vm_next;
2717 }
2718 }
2719
2720 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 * Remove the vma's, and unmap the actual pages
2722 */
Hugh Dickins146425a2005-04-19 13:29:18 -07002723 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2724 unmap_region(mm, vma, prev, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Dave Hansen1de4fa12014-11-14 07:18:31 -08002726 arch_unmap(mm, vma, start, end);
2727
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 /* Fix up all other VM information */
Hugh Dickins2c0b3812005-10-29 18:15:56 -07002729 remove_vma_list(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730
2731 return 0;
2732}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733
Al Virobfce2812012-04-20 21:57:04 -04002734int vm_munmap(unsigned long start, size_t len)
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002735{
2736 int ret;
Al Virobfce2812012-04-20 21:57:04 -04002737 struct mm_struct *mm = current->mm;
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002738 LIST_HEAD(uf);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002739
Michal Hockoae798782016-05-23 16:25:33 -07002740 if (down_write_killable(&mm->mmap_sem))
2741 return -EINTR;
2742
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002743 ret = do_munmap(mm, start, len, &uf);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002744 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002745 userfaultfd_unmap_complete(mm, &uf);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07002746 return ret;
2747}
2748EXPORT_SYMBOL(vm_munmap);
2749
Heiko Carstens6a6160a2009-01-14 14:14:15 +01002750SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 profile_munmap(addr);
Mike Rapoport846b1a02017-02-24 14:58:19 -08002753 return vm_munmap(addr, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754}
2755
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002756
2757/*
2758 * Emulation of deprecated remap_file_pages() syscall.
2759 */
2760SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2761 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2762{
2763
2764 struct mm_struct *mm = current->mm;
2765 struct vm_area_struct *vma;
2766 unsigned long populate = 0;
2767 unsigned long ret = -EINVAL;
2768 struct file *file;
2769
Joe Perches756a0252016-03-17 14:19:47 -07002770 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2771 current->comm, current->pid);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002772
2773 if (prot)
2774 return ret;
2775 start = start & PAGE_MASK;
2776 size = size & PAGE_MASK;
2777
2778 if (start + size <= start)
2779 return ret;
2780
2781 /* Does pgoff wrap? */
2782 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2783 return ret;
2784
Michal Hockodc0ef0d2016-05-23 16:25:27 -07002785 if (down_write_killable(&mm->mmap_sem))
2786 return -EINTR;
2787
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002788 vma = find_vma(mm, start);
2789
2790 if (!vma || !(vma->vm_flags & VM_SHARED))
2791 goto out;
2792
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002793 if (start < vma->vm_start)
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002794 goto out;
2795
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002796 if (start + size > vma->vm_end) {
2797 struct vm_area_struct *next;
2798
2799 for (next = vma->vm_next; next; next = next->vm_next) {
2800 /* hole between vmas ? */
2801 if (next->vm_start != next->vm_prev->vm_end)
2802 goto out;
2803
2804 if (next->vm_file != vma->vm_file)
2805 goto out;
2806
2807 if (next->vm_flags != vma->vm_flags)
2808 goto out;
2809
2810 if (start + size <= next->vm_end)
2811 break;
2812 }
2813
2814 if (!next)
2815 goto out;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002816 }
2817
2818 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2819 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2820 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2821
2822 flags &= MAP_NONBLOCK;
2823 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2824 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002825 struct vm_area_struct *tmp;
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002826 flags |= MAP_LOCKED;
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002827
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002828 /* drop PG_Mlocked flag for over-mapped range */
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002829 for (tmp = vma; tmp->vm_start >= start + size;
2830 tmp = tmp->vm_next) {
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07002831 /*
2832 * Split pmd and munlock page on the border
2833 * of the range.
2834 */
2835 vma_adjust_trans_huge(tmp, start, start + size, 0);
2836
Kirill A. Shutemov48f7df32016-02-17 13:11:15 -08002837 munlock_vma_pages_range(tmp,
2838 max(tmp->vm_start, start),
2839 min(tmp->vm_end, start + size));
2840 }
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002841 }
2842
2843 file = get_file(vma->vm_file);
2844 ret = do_mmap_pgoff(vma->vm_file, start, size,
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002845 prot, flags, pgoff, &populate, NULL);
Kirill A. Shutemovc8d78c12015-02-10 14:09:46 -08002846 fput(file);
2847out:
2848 up_write(&mm->mmap_sem);
2849 if (populate)
2850 mm_populate(ret, populate);
2851 if (!IS_ERR_VALUE(ret))
2852 ret = 0;
2853 return ret;
2854}
2855
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856static inline void verify_mm_writelocked(struct mm_struct *mm)
2857{
Paul E. McKenneya241ec62005-10-30 15:03:12 -08002858#ifdef CONFIG_DEBUG_VM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2860 WARN_ON(1);
2861 up_read(&mm->mmap_sem);
2862 }
2863#endif
2864}
2865
2866/*
2867 * this is really a simplified "do_mmap". it only handles
2868 * anonymous maps. eventually we may be able to do some
2869 * brk-specific accounting here.
2870 */
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002871static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872{
vishnu.pscc71aba2014-10-09 15:26:29 -07002873 struct mm_struct *mm = current->mm;
2874 struct vm_area_struct *vma, *prev;
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002875 unsigned long len;
vishnu.pscc71aba2014-10-09 15:26:29 -07002876 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 pgoff_t pgoff = addr >> PAGE_SHIFT;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002878 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879
Kees Cookba093a62016-08-02 14:04:54 -07002880 len = PAGE_ALIGN(request);
2881 if (len < request)
2882 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 if (!len)
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002884 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002886 /* Until we need other flags, refuse anything except VM_EXEC. */
2887 if ((flags & (~VM_EXEC)) != 0)
2888 return -EINVAL;
2889 flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
Kirill Korotaev3a459752006-09-07 14:17:04 +04002890
Al Viro2c6a1012009-12-03 19:40:46 -05002891 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
Alexander Kuleshovde1741a2015-11-05 18:46:54 -08002892 if (offset_in_page(error))
Kirill Korotaev3a459752006-09-07 14:17:04 +04002893 return error;
2894
Davidlohr Bueso363ee172014-01-21 15:49:15 -08002895 error = mlock_future_check(mm, mm->def_flags, len);
2896 if (error)
2897 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898
2899 /*
2900 * mm->mmap_sem is required to protect against another thread
2901 * changing the mappings in case we sleep.
2902 */
2903 verify_mm_writelocked(mm);
2904
2905 /*
2906 * Clear old maps. this also does some error checking for us
2907 */
Rasmus Villemoes9fcd1452015-04-15 16:14:32 -07002908 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2909 &rb_parent)) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002910 if (do_munmap(mm, addr, len, uf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 }
2913
2914 /* Check against address space limits *after* clearing old maps... */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002915 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 return -ENOMEM;
2917
2918 if (mm->map_count > sysctl_max_map_count)
2919 return -ENOMEM;
2920
Al Viro191c5422012-02-13 03:58:52 +00002921 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 return -ENOMEM;
2923
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 /* Can we just expand an old private anonymous mapping? */
Rik van Rielba470de2008-10-18 20:26:50 -07002925 vma = vma_merge(mm, prev, addr, addr + len, flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07002926 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
Rik van Rielba470de2008-10-18 20:26:50 -07002927 if (vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 goto out;
2929
2930 /*
2931 * create a vma struct for an anonymous mapping
2932 */
Pekka Enbergc5e3b832006-03-25 03:06:43 -08002933 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 if (!vma) {
2935 vm_unacct_memory(len >> PAGE_SHIFT);
2936 return -ENOMEM;
2937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938
Rik van Riel5beb4932010-03-05 13:42:07 -08002939 INIT_LIST_HEAD(&vma->anon_vma_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 vma->vm_mm = mm;
2941 vma->vm_start = addr;
2942 vma->vm_end = addr + len;
2943 vma->vm_pgoff = pgoff;
2944 vma->vm_flags = flags;
Coly Li3ed75eb2007-10-18 23:39:15 -07002945 vma->vm_page_prot = vm_get_page_prot(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 vma_link(mm, vma, prev, rb_link, rb_parent);
2947out:
Eric B Munson3af9e852010-05-18 15:30:49 +01002948 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 mm->total_vm += len >> PAGE_SHIFT;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08002950 mm->data_vm += len >> PAGE_SHIFT;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002951 if (flags & VM_LOCKED)
2952 mm->locked_vm += (len >> PAGE_SHIFT);
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07002953 vma->vm_flags |= VM_SOFTDIRTY;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002954 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955}
2956
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002957static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf)
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002958{
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002959 return do_brk_flags(addr, len, 0, uf);
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002960}
2961
2962int vm_brk_flags(unsigned long addr, unsigned long len, unsigned long flags)
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002963{
2964 struct mm_struct *mm = current->mm;
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002965 int ret;
Michel Lespinasse128557f2013-02-22 16:32:40 -08002966 bool populate;
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002967 LIST_HEAD(uf);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002968
Michal Hocko2d6c9282016-05-23 16:25:42 -07002969 if (down_write_killable(&mm->mmap_sem))
2970 return -EINTR;
2971
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002972 ret = do_brk_flags(addr, len, flags, &uf);
Michel Lespinasse128557f2013-02-22 16:32:40 -08002973 populate = ((mm->def_flags & VM_LOCKED) != 0);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002974 up_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -08002975 userfaultfd_unmap_complete(mm, &uf);
Linus Torvalds5d22fc22016-05-27 15:57:31 -07002976 if (populate && !ret)
Michel Lespinasse128557f2013-02-22 16:32:40 -08002977 mm_populate(addr, len);
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002978 return ret;
2979}
Denys Vlasenko16e72e92017-02-22 15:45:16 -08002980EXPORT_SYMBOL(vm_brk_flags);
2981
2982int vm_brk(unsigned long addr, unsigned long len)
2983{
2984 return vm_brk_flags(addr, len, 0);
2985}
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07002986EXPORT_SYMBOL(vm_brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
2988/* Release all mmaps. */
2989void exit_mmap(struct mm_struct *mm)
2990{
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07002991 struct mmu_gather tlb;
Rik van Rielba470de2008-10-18 20:26:50 -07002992 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 unsigned long nr_accounted = 0;
2994
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002995 /* mm's last user has gone, and its about to be pulled down */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002996 mmu_notifier_release(mm);
Jeremy Fitzhardinged6dd61c2007-05-02 19:27:14 +02002997
Rik van Rielba470de2008-10-18 20:26:50 -07002998 if (mm->locked_vm) {
2999 vma = mm->mmap;
3000 while (vma) {
3001 if (vma->vm_flags & VM_LOCKED)
3002 munlock_vma_pages_all(vma);
3003 vma = vma->vm_next;
3004 }
3005 }
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08003006
3007 arch_exit_mmap(mm);
3008
Rik van Rielba470de2008-10-18 20:26:50 -07003009 vma = mm->mmap;
Jeremy Fitzhardinge9480c532009-02-11 13:04:41 -08003010 if (!vma) /* Can happen if dup_mmap() received an OOM */
3011 return;
3012
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 flush_cache_mm(mm);
Linus Torvalds2b047252013-08-15 11:42:25 -07003015 tlb_gather_mmu(&tlb, mm, 0, -1);
Oleg Nesterov901608d2009-01-06 14:40:29 -08003016 /* update_hiwater_rss(mm) here? but nobody should be looking */
Hugh Dickinse0da3822005-04-19 13:29:15 -07003017 /* Use -1 here to ensure all VMAs in the mm are unmapped */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003018 unmap_vmas(&tlb, vma, 0, -1);
Hugh Dickins9ba69292009-09-21 17:02:20 -07003019
Andrea Arcangeli21292582017-09-06 16:25:00 -07003020 set_bit(MMF_OOM_SKIP, &mm->flags);
3021 if (unlikely(tsk_is_oom_victim(current))) {
3022 /*
3023 * Wait for oom_reap_task() to stop working on this
3024 * mm. Because MMF_OOM_SKIP is already set before
3025 * calling down_read(), oom_reap_task() will not run
3026 * on this "mm" post up_write().
3027 *
3028 * tsk_is_oom_victim() cannot be set from under us
3029 * either because current->mm is already set to NULL
3030 * under task_lock before calling mmput and oom_mm is
3031 * set not NULL by the OOM killer only if current->mm
3032 * is found not NULL while holding the task_lock.
3033 */
3034 down_write(&mm->mmap_sem);
3035 up_write(&mm->mmap_sem);
3036 }
Hugh Dickins6ee86302013-04-29 15:07:44 -07003037 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
Al Viro853f5e22012-03-05 14:03:47 -05003038 tlb_finish_mmu(&tlb, 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 /*
Hugh Dickins8f4f8c12005-10-29 18:16:29 -07003041 * Walk the list again, actually closing and freeing it,
3042 * with preemption enabled, without holding any MM locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 */
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003044 while (vma) {
3045 if (vma->vm_flags & VM_ACCOUNT)
3046 nr_accounted += vma_pages(vma);
Hugh Dickinsa8fb5612005-10-29 18:15:57 -07003047 vma = remove_vma(vma);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07003048 }
3049 vm_unacct_memory(nr_accounted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050}
3051
3052/* Insert vm structure into process list sorted by address
3053 * and into the inode's i_mmap tree. If vm_file is non-NULL
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003054 * then i_mmap_rwsem is taken here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055 */
Hugh Dickins6597d782012-10-08 16:29:07 -07003056int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057{
Hugh Dickins6597d782012-10-08 16:29:07 -07003058 struct vm_area_struct *prev;
3059 struct rb_node **rb_link, *rb_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
Chen Gangc9d13f52015-09-08 15:04:08 -07003061 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3062 &prev, &rb_link, &rb_parent))
3063 return -ENOMEM;
3064 if ((vma->vm_flags & VM_ACCOUNT) &&
3065 security_vm_enough_memory_mm(mm, vma_pages(vma)))
3066 return -ENOMEM;
3067
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 /*
3069 * The vm_pgoff of a purely anonymous vma should be irrelevant
3070 * until its first write fault, when page's anon_vma and index
3071 * are set. But now set the vm_pgoff it will almost certainly
3072 * end up with (unless mremap moves it elsewhere before that
3073 * first wfault), so /proc/pid/maps tells a consistent story.
3074 *
3075 * By setting it to reflect the virtual start address of the
3076 * vma, merges and splits can happen in a seamless way, just
3077 * using the existing file pgoff checks and manipulations.
3078 * Similarly in do_mmap_pgoff and in do_brk.
3079 */
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003080 if (vma_is_anonymous(vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 BUG_ON(vma->anon_vma);
3082 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3083 }
Srikar Dronamraju2b144492012-02-09 14:56:42 +05303084
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 vma_link(mm, vma, prev, rb_link, rb_parent);
3086 return 0;
3087}
3088
3089/*
3090 * Copy the vma structure to a new location in the same mm,
3091 * prior to moving page table entries, to effect an mremap move.
3092 */
3093struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
Michel Lespinasse38a76012012-10-08 16:31:50 -07003094 unsigned long addr, unsigned long len, pgoff_t pgoff,
3095 bool *need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096{
3097 struct vm_area_struct *vma = *vmap;
3098 unsigned long vma_start = vma->vm_start;
3099 struct mm_struct *mm = vma->vm_mm;
3100 struct vm_area_struct *new_vma, *prev;
3101 struct rb_node **rb_link, *rb_parent;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003102 bool faulted_in_anon_vma = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103
3104 /*
3105 * If anonymous vma has not yet been faulted, update new pgoff
3106 * to match new location, to increase its chance of merging.
3107 */
Oleg Nesterovce757992015-09-08 14:58:34 -07003108 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 pgoff = addr >> PAGE_SHIFT;
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003110 faulted_in_anon_vma = false;
3111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
Hugh Dickins6597d782012-10-08 16:29:07 -07003113 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3114 return NULL; /* should never get here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -07003116 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3117 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 if (new_vma) {
3119 /*
3120 * Source vma may have been merged into new_vma
3121 */
Andrea Arcangeli948f0172012-01-10 15:08:05 -08003122 if (unlikely(vma_start >= new_vma->vm_start &&
3123 vma_start < new_vma->vm_end)) {
3124 /*
3125 * The only way we can get a vma_merge with
3126 * self during an mremap is if the vma hasn't
3127 * been faulted in yet and we were allowed to
3128 * reset the dst vma->vm_pgoff to the
3129 * destination address of the mremap to allow
3130 * the merge to happen. mremap must change the
3131 * vm_pgoff linearity between src and dst vmas
3132 * (in turn preventing a vma_merge) to be
3133 * safe. It is only safe to keep the vm_pgoff
3134 * linear if there are no pages mapped yet.
3135 */
Sasha Levin81d1b092014-10-09 15:28:10 -07003136 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -07003137 *vmap = vma = new_vma;
Michel Lespinasse108d6642012-10-08 16:31:36 -07003138 }
Michel Lespinasse38a76012012-10-08 16:31:50 -07003139 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 } else {
Christoph Lametere94b1762006-12-06 20:33:17 -08003141 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
Chen Gange3975892015-09-08 15:03:38 -07003142 if (!new_vma)
3143 goto out;
3144 *new_vma = *vma;
3145 new_vma->vm_start = addr;
3146 new_vma->vm_end = addr + len;
3147 new_vma->vm_pgoff = pgoff;
3148 if (vma_dup_policy(vma, new_vma))
3149 goto out_free_vma;
3150 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
3151 if (anon_vma_clone(new_vma, vma))
3152 goto out_free_mempol;
3153 if (new_vma->vm_file)
3154 get_file(new_vma->vm_file);
3155 if (new_vma->vm_ops && new_vma->vm_ops->open)
3156 new_vma->vm_ops->open(new_vma);
3157 vma_link(mm, new_vma, prev, rb_link, rb_parent);
3158 *need_rmap_locks = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 }
3160 return new_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003161
Chen Gange3975892015-09-08 15:03:38 -07003162out_free_mempol:
Oleg Nesterovef0855d2013-09-11 14:20:14 -07003163 mpol_put(vma_policy(new_vma));
Chen Gange3975892015-09-08 15:03:38 -07003164out_free_vma:
Rik van Riel5beb4932010-03-05 13:42:07 -08003165 kmem_cache_free(vm_area_cachep, new_vma);
Chen Gange3975892015-09-08 15:03:38 -07003166out:
Rik van Riel5beb4932010-03-05 13:42:07 -08003167 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168}
akpm@osdl.org119f6572005-05-01 08:58:35 -07003169
3170/*
3171 * Return true if the calling process may expand its vm space by the passed
3172 * number of pages
3173 */
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003174bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
akpm@osdl.org119f6572005-05-01 08:58:35 -07003175{
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003176 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3177 return false;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003178
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003179 if (is_data_mapping(flags) &&
3180 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003181 /* Workaround for Valgrind */
3182 if (rlimit(RLIMIT_DATA) == 0 &&
3183 mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3184 return true;
3185 if (!ignore_rlimit_data) {
3186 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 -08003187 current->comm, current->pid,
3188 (mm->data_vm + npages) << PAGE_SHIFT,
3189 rlimit(RLIMIT_DATA));
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003190 return false;
Konstantin Khlebnikovf4fcd552016-05-20 16:57:45 -07003191 }
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003192 }
akpm@osdl.org119f6572005-05-01 08:58:35 -07003193
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003194 return true;
3195}
3196
3197void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3198{
3199 mm->total_vm += npages;
3200
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003201 if (is_exec_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003202 mm->exec_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003203 else if (is_stack_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003204 mm->stack_vm += npages;
Konstantin Khlebnikovd977d562016-02-02 16:57:43 -08003205 else if (is_data_mapping(flags))
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003206 mm->data_vm += npages;
akpm@osdl.org119f6572005-05-01 08:58:35 -07003207}
Roland McGrathfa5dc222007-02-08 14:20:41 -08003208
Dave Jiang11bac802017-02-24 14:56:41 -08003209static int special_mapping_fault(struct vm_fault *vmf);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003210
3211/*
3212 * Having a close hook prevents vma merging regardless of flags.
3213 */
3214static void special_mapping_close(struct vm_area_struct *vma)
3215{
3216}
3217
3218static const char *special_mapping_name(struct vm_area_struct *vma)
3219{
3220 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3221}
3222
Dmitry Safonovb059a452016-06-28 14:35:38 +03003223static int special_mapping_mremap(struct vm_area_struct *new_vma)
3224{
3225 struct vm_special_mapping *sm = new_vma->vm_private_data;
3226
Dmitry Safonov280e87e2017-06-19 17:32:42 +01003227 if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3228 return -EFAULT;
3229
Dmitry Safonovb059a452016-06-28 14:35:38 +03003230 if (sm->mremap)
3231 return sm->mremap(sm, new_vma);
Dmitry Safonov280e87e2017-06-19 17:32:42 +01003232
Dmitry Safonovb059a452016-06-28 14:35:38 +03003233 return 0;
3234}
3235
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003236static const struct vm_operations_struct special_mapping_vmops = {
3237 .close = special_mapping_close,
3238 .fault = special_mapping_fault,
Dmitry Safonovb059a452016-06-28 14:35:38 +03003239 .mremap = special_mapping_mremap,
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003240 .name = special_mapping_name,
3241};
3242
3243static const struct vm_operations_struct legacy_special_mapping_vmops = {
3244 .close = special_mapping_close,
3245 .fault = special_mapping_fault,
3246};
Roland McGrathfa5dc222007-02-08 14:20:41 -08003247
Dave Jiang11bac802017-02-24 14:56:41 -08003248static int special_mapping_fault(struct vm_fault *vmf)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003249{
Dave Jiang11bac802017-02-24 14:56:41 -08003250 struct vm_area_struct *vma = vmf->vma;
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003251 pgoff_t pgoff;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003252 struct page **pages;
3253
Andy Lutomirskif872f542015-12-29 20:12:19 -08003254 if (vma->vm_ops == &legacy_special_mapping_vmops) {
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003255 pages = vma->vm_private_data;
Andy Lutomirskif872f542015-12-29 20:12:19 -08003256 } else {
3257 struct vm_special_mapping *sm = vma->vm_private_data;
3258
3259 if (sm->fault)
Dave Jiang11bac802017-02-24 14:56:41 -08003260 return sm->fault(sm, vmf->vma, vmf);
Andy Lutomirskif872f542015-12-29 20:12:19 -08003261
3262 pages = sm->pages;
3263 }
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003264
Oleg Nesterov8a9cc3b2015-09-08 14:58:31 -07003265 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003266 pgoff--;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003267
3268 if (*pages) {
3269 struct page *page = *pages;
3270 get_page(page);
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003271 vmf->page = page;
3272 return 0;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003273 }
3274
Nick Pigginb1d0e4f2008-02-09 01:15:19 +01003275 return VM_FAULT_SIGBUS;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003276}
3277
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003278static struct vm_area_struct *__install_special_mapping(
3279 struct mm_struct *mm,
3280 unsigned long addr, unsigned long len,
Chen Gang27f28b92015-11-05 18:48:41 -08003281 unsigned long vm_flags, void *priv,
3282 const struct vm_operations_struct *ops)
Roland McGrathfa5dc222007-02-08 14:20:41 -08003283{
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003284 int ret;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003285 struct vm_area_struct *vma;
3286
3287 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3288 if (unlikely(vma == NULL))
Stefani Seibold3935ed62014-03-17 23:22:02 +01003289 return ERR_PTR(-ENOMEM);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003290
Rik van Riel5beb4932010-03-05 13:42:07 -08003291 INIT_LIST_HEAD(&vma->anon_vma_chain);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003292 vma->vm_mm = mm;
3293 vma->vm_start = addr;
3294 vma->vm_end = addr + len;
3295
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -07003296 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
Coly Li3ed75eb2007-10-18 23:39:15 -07003297 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003298
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003299 vma->vm_ops = ops;
3300 vma->vm_private_data = priv;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003301
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003302 ret = insert_vm_struct(mm, vma);
3303 if (ret)
3304 goto out;
Roland McGrathfa5dc222007-02-08 14:20:41 -08003305
Konstantin Khlebnikov84638332016-01-14 15:22:07 -08003306 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003307
Ingo Molnarcdd6c482009-09-21 12:02:48 +02003308 perf_event_mmap(vma);
Peter Zijlstra089dd792009-06-05 14:04:55 +02003309
Stefani Seibold3935ed62014-03-17 23:22:02 +01003310 return vma;
Tavis Ormandy462e635e2010-12-09 15:29:42 +01003311
3312out:
3313 kmem_cache_free(vm_area_cachep, vma);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003314 return ERR_PTR(ret);
3315}
3316
Dmitry Safonov2eefd872016-09-05 16:33:05 +03003317bool vma_is_special_mapping(const struct vm_area_struct *vma,
3318 const struct vm_special_mapping *sm)
3319{
3320 return vma->vm_private_data == sm &&
3321 (vma->vm_ops == &special_mapping_vmops ||
3322 vma->vm_ops == &legacy_special_mapping_vmops);
3323}
3324
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003325/*
3326 * Called with mm->mmap_sem held for writing.
3327 * Insert a new vma covering the given region, with the given flags.
3328 * Its pages are supplied by the given array of struct page *.
3329 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3330 * The region past the last page supplied will always produce SIGBUS.
3331 * The array pointer and the pages it points to are assumed to stay alive
3332 * for as long as this mapping might exist.
3333 */
3334struct vm_area_struct *_install_special_mapping(
3335 struct mm_struct *mm,
3336 unsigned long addr, unsigned long len,
3337 unsigned long vm_flags, const struct vm_special_mapping *spec)
3338{
Chen Gang27f28b92015-11-05 18:48:41 -08003339 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3340 &special_mapping_vmops);
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003341}
3342
Stefani Seibold3935ed62014-03-17 23:22:02 +01003343int install_special_mapping(struct mm_struct *mm,
3344 unsigned long addr, unsigned long len,
3345 unsigned long vm_flags, struct page **pages)
3346{
Andy Lutomirskia62c34b2014-05-19 15:58:33 -07003347 struct vm_area_struct *vma = __install_special_mapping(
Chen Gang27f28b92015-11-05 18:48:41 -08003348 mm, addr, len, vm_flags, (void *)pages,
3349 &legacy_special_mapping_vmops);
Stefani Seibold3935ed62014-03-17 23:22:02 +01003350
Duan Jiong14bd5b42014-06-04 16:07:05 -07003351 return PTR_ERR_OR_ZERO(vma);
Roland McGrathfa5dc222007-02-08 14:20:41 -08003352}
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003353
3354static DEFINE_MUTEX(mm_all_locks_mutex);
3355
Peter Zijlstra454ed842008-08-11 09:30:25 +02003356static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003357{
Davidlohr Buesof808c132017-09-08 16:15:08 -07003358 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003359 /*
3360 * The LSB of head.next can't change from under us
3361 * because we hold the mm_all_locks_mutex.
3362 */
Jiri Kosina572043c2013-01-11 14:31:59 -08003363 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003364 /*
3365 * We can safely modify head.next after taking the
Ingo Molnar5a505082012-12-02 19:56:46 +00003366 * anon_vma->root->rwsem. If some other vma in this mm shares
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003367 * the same anon_vma we won't take it again.
3368 *
3369 * No need of atomic instructions here, head.next
3370 * can't change from under us thanks to the
Ingo Molnar5a505082012-12-02 19:56:46 +00003371 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003372 */
3373 if (__test_and_set_bit(0, (unsigned long *)
Davidlohr Buesof808c132017-09-08 16:15:08 -07003374 &anon_vma->root->rb_root.rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003375 BUG();
3376 }
3377}
3378
Peter Zijlstra454ed842008-08-11 09:30:25 +02003379static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003380{
3381 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3382 /*
3383 * AS_MM_ALL_LOCKS can't change from under us because
3384 * we hold the mm_all_locks_mutex.
3385 *
3386 * Operations on ->flags have to be atomic because
3387 * even if AS_MM_ALL_LOCKS is stable thanks to the
3388 * mm_all_locks_mutex, there may be other cpus
3389 * changing other bitflags in parallel to us.
3390 */
3391 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3392 BUG();
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08003393 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003394 }
3395}
3396
3397/*
3398 * This operation locks against the VM for all pte/vma/mm related
3399 * operations that could ever happen on a certain mm. This includes
3400 * vmtruncate, try_to_unmap, and all page faults.
3401 *
3402 * The caller must take the mmap_sem in write mode before calling
3403 * mm_take_all_locks(). The caller isn't allowed to release the
3404 * mmap_sem until mm_drop_all_locks() returns.
3405 *
3406 * mmap_sem in write mode is required in order to block all operations
3407 * that could modify pagetables and free pages without need of
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08003408 * altering the vma layout. It's also needed in write mode to avoid new
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003409 * anon_vmas to be associated with existing vmas.
3410 *
3411 * A single task can't take more than one mm_take_all_locks() in a row
3412 * or it would deadlock.
3413 *
Michel Lespinassebf181b92012-10-08 16:31:39 -07003414 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003415 * mapping->flags avoid to take the same lock twice, if more than one
3416 * vma in this mm is backed by the same anon_vma or address_space.
3417 *
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003418 * We take locks in following order, accordingly to comment at beginning
3419 * of mm/rmap.c:
3420 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3421 * hugetlb mapping);
3422 * - all i_mmap_rwsem locks;
3423 * - all anon_vma->rwseml
3424 *
3425 * We can take all locks within these types randomly because the VM code
3426 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3427 * mm_all_locks_mutex.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003428 *
3429 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3430 * that may have to take thousand of locks.
3431 *
3432 * mm_take_all_locks() can fail if it's interrupted by signals.
3433 */
3434int mm_take_all_locks(struct mm_struct *mm)
3435{
3436 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003437 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003438
3439 BUG_ON(down_read_trylock(&mm->mmap_sem));
3440
3441 mutex_lock(&mm_all_locks_mutex);
3442
3443 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3444 if (signal_pending(current))
3445 goto out_unlock;
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -08003446 if (vma->vm_file && vma->vm_file->f_mapping &&
3447 is_vm_hugetlb_page(vma))
3448 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3449 }
3450
3451 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3452 if (signal_pending(current))
3453 goto out_unlock;
3454 if (vma->vm_file && vma->vm_file->f_mapping &&
3455 !is_vm_hugetlb_page(vma))
Peter Zijlstra454ed842008-08-11 09:30:25 +02003456 vm_lock_mapping(mm, vma->vm_file->f_mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003457 }
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003458
3459 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3460 if (signal_pending(current))
3461 goto out_unlock;
3462 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003463 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3464 vm_lock_anon_vma(mm, avc->anon_vma);
Peter Zijlstra7cd5a022008-08-11 09:30:25 +02003465 }
3466
Kautuk Consul584cff52011-10-31 17:08:59 -07003467 return 0;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003468
3469out_unlock:
Kautuk Consul584cff52011-10-31 17:08:59 -07003470 mm_drop_all_locks(mm);
3471 return -EINTR;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003472}
3473
3474static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3475{
Davidlohr Buesof808c132017-09-08 16:15:08 -07003476 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003477 /*
3478 * The LSB of head.next can't change to 0 from under
3479 * us because we hold the mm_all_locks_mutex.
3480 *
3481 * We must however clear the bitflag before unlocking
Michel Lespinassebf181b92012-10-08 16:31:39 -07003482 * the vma so the users using the anon_vma->rb_root will
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003483 * never see our bitflag.
3484 *
3485 * No need of atomic instructions here, head.next
3486 * can't change from under us until we release the
Ingo Molnar5a505082012-12-02 19:56:46 +00003487 * anon_vma->root->rwsem.
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003488 */
3489 if (!__test_and_clear_bit(0, (unsigned long *)
Davidlohr Buesof808c132017-09-08 16:15:08 -07003490 &anon_vma->root->rb_root.rb_root.rb_node))
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003491 BUG();
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -08003492 anon_vma_unlock_write(anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003493 }
3494}
3495
3496static void vm_unlock_mapping(struct address_space *mapping)
3497{
3498 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3499 /*
3500 * AS_MM_ALL_LOCKS can't change to 0 from under us
3501 * because we hold the mm_all_locks_mutex.
3502 */
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08003503 i_mmap_unlock_write(mapping);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003504 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3505 &mapping->flags))
3506 BUG();
3507 }
3508}
3509
3510/*
3511 * The mmap_sem cannot be released by the caller until
3512 * mm_drop_all_locks() returns.
3513 */
3514void mm_drop_all_locks(struct mm_struct *mm)
3515{
3516 struct vm_area_struct *vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08003517 struct anon_vma_chain *avc;
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003518
3519 BUG_ON(down_read_trylock(&mm->mmap_sem));
3520 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3521
3522 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3523 if (vma->anon_vma)
Rik van Riel5beb4932010-03-05 13:42:07 -08003524 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3525 vm_unlock_anon_vma(avc->anon_vma);
Andrea Arcangeli7906d002008-07-28 15:46:26 -07003526 if (vma->vm_file && vma->vm_file->f_mapping)
3527 vm_unlock_mapping(vma->vm_file->f_mapping);
3528 }
3529
3530 mutex_unlock(&mm_all_locks_mutex);
3531}
David Howells8feae132009-01-08 12:04:47 +00003532
3533/*
seokhoon.yoon3edf41d2017-02-24 14:56:44 -08003534 * initialise the percpu counter for VM
David Howells8feae132009-01-08 12:04:47 +00003535 */
3536void __init mmap_init(void)
3537{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003538 int ret;
3539
Tejun Heo908c7f12014-09-08 09:51:29 +09003540 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07003541 VM_BUG_ON(ret);
David Howells8feae132009-01-08 12:04:47 +00003542}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003543
3544/*
3545 * Initialise sysctl_user_reserve_kbytes.
3546 *
3547 * This is intended to prevent a user from starting a single memory hogging
3548 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3549 * mode.
3550 *
3551 * The default value is min(3% of free memory, 128MB)
3552 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3553 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003554static int init_user_reserve(void)
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003555{
3556 unsigned long free_kbytes;
3557
Michal Hockoc41f0122017-09-06 16:23:36 -07003558 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07003559
3560 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3561 return 0;
3562}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003563subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003564
3565/*
3566 * Initialise sysctl_admin_reserve_kbytes.
3567 *
3568 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3569 * to log in and kill a memory hogging process.
3570 *
3571 * Systems with more than 256MB will reserve 8MB, enough to recover
3572 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3573 * only reserve 3% of free pages by default.
3574 */
Andrew Shewmaker16408792013-04-29 15:08:12 -07003575static int init_admin_reserve(void)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003576{
3577 unsigned long free_kbytes;
3578
Michal Hockoc41f0122017-09-06 16:23:36 -07003579 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07003580
3581 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3582 return 0;
3583}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003584subsys_initcall(init_admin_reserve);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003585
3586/*
3587 * Reinititalise user and admin reserves if memory is added or removed.
3588 *
3589 * The default user reserve max is 128MB, and the default max for the
3590 * admin reserve is 8MB. These are usually, but not always, enough to
3591 * enable recovery from a memory hogging process using login/sshd, a shell,
3592 * and tools like top. It may make sense to increase or even disable the
3593 * reserve depending on the existence of swap or variations in the recovery
3594 * tools. So, the admin may have changed them.
3595 *
3596 * If memory is added and the reserves have been eliminated or increased above
3597 * the default max, then we'll trust the admin.
3598 *
3599 * If memory is removed and there isn't enough free memory, then we
3600 * need to reset the reserves.
3601 *
3602 * Otherwise keep the reserve set by the admin.
3603 */
3604static int reserve_mem_notifier(struct notifier_block *nb,
3605 unsigned long action, void *data)
3606{
3607 unsigned long tmp, free_kbytes;
3608
3609 switch (action) {
3610 case MEM_ONLINE:
3611 /* Default max is 128MB. Leave alone if modified by operator. */
3612 tmp = sysctl_user_reserve_kbytes;
3613 if (0 < tmp && tmp < (1UL << 17))
3614 init_user_reserve();
3615
3616 /* Default max is 8MB. Leave alone if modified by operator. */
3617 tmp = sysctl_admin_reserve_kbytes;
3618 if (0 < tmp && tmp < (1UL << 13))
3619 init_admin_reserve();
3620
3621 break;
3622 case MEM_OFFLINE:
Michal Hockoc41f0122017-09-06 16:23:36 -07003623 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
Andrew Shewmaker16408792013-04-29 15:08:12 -07003624
3625 if (sysctl_user_reserve_kbytes > free_kbytes) {
3626 init_user_reserve();
3627 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3628 sysctl_user_reserve_kbytes);
3629 }
3630
3631 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3632 init_admin_reserve();
3633 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3634 sysctl_admin_reserve_kbytes);
3635 }
3636 break;
3637 default:
3638 break;
3639 }
3640 return NOTIFY_OK;
3641}
3642
3643static struct notifier_block reserve_mem_nb = {
3644 .notifier_call = reserve_mem_notifier,
3645};
3646
3647static int __meminit init_reserve_notifier(void)
3648{
3649 if (register_hotmemory_notifier(&reserve_mem_nb))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07003650 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
Andrew Shewmaker16408792013-04-29 15:08:12 -07003651
3652 return 0;
3653}
Paul Gortmakera64fb3c2014-01-23 15:53:30 -08003654subsys_initcall(init_reserve_notifier);